**Early Screening of the Autism Spectrum Disorders: Validity Properties and Cross-Cultural Generalizability of the First Year Inventory in Italy**


Received: 13 January 2020; Accepted: 13 February 2020; Published: 18 February 2020

**Abstract:** This study examined the cross-cultural generalisability of the First Year Inventory (FYI) on an Italian sample, testing its construct validity, consistency, and structural validity. Six hundred ninety-eight parents of children aged 11–13 months completed the questionnaire. Similarities between analyses of Italian and American/Israeli samples were found, as were demonstrations of the instrument's construct validity and internal consistency with both groups. The original factorial structure was not demonstrated; thus, a new factorial structure was tested, and a short version of the FYI was demonstrated via confirmatory factor analysis. The findings supported the generalisability of the Italian version of the FYI and its validity. The FYI may aid in medical decision-making on further steps for referral of the child to an early diagnostic assessment.

**Keywords:** First Year Inventory; autism spectrum disorders; early screening; risk; cross-cultural generalisability; validity

### **1. Introduction**

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterised by (a) persistent deficits in social communication and interaction and (b) restricted and repetitive patterns of behaviours, interests, and/or activities [1]. Recent epidemiological data [2] suggested that the prevalence of ASD reaches the proportion of 1/59 at age 4. To promote early detection of the risk of ASD, as recommended also by the American Academy of Pediatrics [3], several researchers [4–8] developed ad hoc measures for children under 24 months of age that are able to identify behaviours deviating from typical development.

In this vein, a recent systematic review [9] identified 16 Level 1 and 2 screening measures for the early detection of signs of ASD: 4 observational checklists, 2 interviews, and 10 questionnaires. Level 1 screening tools have been developed for the general population to detect children at risk of developmental disorders, including ASD. Level 2 screening measures have been developed to

detect children who are at risk for ASD, since they are already referred to the health service for developmental concerns (i.e., low-risk children) or because they are siblings of children with ASD (i.e., high-risk children). This review identified five promising instruments: the First Year Inventory (FYI) [8], the Modified-CHecklist for Autism in Toddler and its revised/follow-up form (M-CHAT and M-CHAT-R/F) [6,10], the Parental Observation of Early Markers Scale (POEMS) [11], and the Quantitative-CHecklist for Autism in Toddler (Q-CHAT) [4]. Analyses of the psychometric properties of these measures evaluated them as good. At the same time, however, the authors stressed that, for several such measures, further validation studies were needed to evaluate certain methodological properties that, as yet, were not adequately investigated.

The highest number of validation studies retrieved in the literature were for the M-CHAT and the M-CHAT-R/F [6,10,12–26]. The Q-CHAT has been validated by five studies [4,27–30] and the FYI by five studies [8,31–34]. The POEMS has been validated by one study [11].

The M-CHAT and the M-CHAT-R/F can be administered from 16 months of life, the POEMS from 1–24 months of life, the Q-CHAT is administrable when the child is 18 months old, and the FYI when he/she is 11–13 months old. The POEMS requires more administration time as it uses multiple parental observations. The present study focused on the FYI since it allows the earliest screening but—in contrast to the POEMS—requires less administration time and can be completed by parents during regular well-child visits as part of pediatric surveillance.

### *The First Year Inventory: Measure Description and Critical Analysis of the Validation Studies*

The FYI is a Level 1 screening measure designed to detect the ASD risk on the general population. It was developed through a systematic review of the literature conducted by Reznick and colleagues [8], who identified a list of behaviours comprised in the two core diagnostic criteria of the ASD (i.e., socio-communication and social interaction deficit and restricted, repetitive patterns of behaviour) [1]. Specifically, the authors analysed several retrospective studies and descriptive reports provided by parents, which assessed the first months of life of children with a later diagnosis of ASD, and prospective studies on children who had an older sibling with a diagnosis of ASD. As the authors highlighted, two sets of behaviours, clustered in two categories labelled 'Social–Communication' and 'Sensory–Regulatory Functions', detect children who are at risk of developing, at an early age, an ASD [8]. The Social–Communication domain was further differentiated into four constructs (Social Orienting & Receptive Communication, Social-Affective Engagement, Imitation, and Expressive Communication) as well as the Sensory–Regulatory Functions domain (Sensory Processing, Regulatory Patterns, Reactivity, and Repetitive Behaviors). For a detailed description of the domains and constructs, please refer to the Appendix of Reznick and colleagues' paper [8].

The 63 items of the FYI include 46 questions with response options on a four-point Likert scale (from 1—'never'—to 4—'often') and 14 items with answers in a three or four ad hoc multiple-choice format (see Appendix in [8]). Three additional open-ended questions were on (a) the number of consonants used by the child (Item 61); (b) parental concerns or interests about the child's development (Item 62); and (c) the presence of a specific medical condition (Item 63). Item 61 is scored from 0 (i.e., if the child uses more than three consonants) to 2 points (if the child uses only one or any consonants). The two last open-ended questions (Items 62 and 63) did not receive a score because they were used for qualitative evaluation.

This first study of the FYI was on an American sample (N = 1300) selected from the general population [8], with the purposes of (a) defining the scoring procedure; (b) identifying the risk cut-offs; and (c) evaluating the factorial structure of the instrument. With regard to the scoring procedure, according to the response distribution of the sample, the authors assigned 0 or 1 point to the answers corresponding to behaviours with the highest frequency expected in typically developing children (i.e., low risk). For example, Item 1 ('Does he/she look at you when his/her name is called?') received 0 or 1 point when the answer is respectively 'always' or 'sometimes' because it is expected that a typically developing child looks at the person who calls his/her name. Two points are assigned to answers

that have either low frequency (<5%) or correspond to behaviours unusual in typically developing children. For Item 1, the answers 'never' and 'seldom' receive 2 points because they represent unusual behaviours for a typically developing child.

To identify the cutoffs of risk, the authors [8] observed that the distribution had a chi-squared shape and identified a significant shape inflection corresponding to the score of 17 (at the 95th percentile of the distribution). Finally, they conducted an exploratory factor analysis (EFA), applying the principal factor method followed by a promax (oblique) rotation. The EFA accounted for six factors corresponding to four constructs of the Socio-Communication domain (Social-Affective Engagement—six items, Imitation—four items, Social Orientation—two items, and Expressive Communication—two items) and two constructs of the Sensory–Regulatory Functions domain (Regulatory Patterns—four items and Repetitive Behaviors—eight items). Thirty items did not load for any factor or loaded for more than one factor. To accomplish the broader goal of developing a measure for the early detection of ASD, the authors sorted the 61 items into the hypothesised eight constructs and two domains according to the theoretical model. After the EFA, each of the nonassigned items was allocated to a construct if the item theoretically fitted with that construct, the item–total correlation was higher than 0.30, and the change to Cronbach's alpha was negligible. After that procedure, nine items were assigned to an uncategorised group because they did not fit any of these criteria.

The FYI was tested in four other studies [31–34]. One [33] was a follow-up investigation of the Reznick and colleagues' sample [8] developed three years later. Two were retrospective studies on an American [34] and an Italian sample [32] of children with ASD. Finally, a more recent study [31] was published using an Israeli sample from the general population.

In the following section, we reported a critical comparison between validation studies.

All validation studies carried out analysis on children's (gender and family size) and parents' (educational level, ethnicity, and marital status) socio-demographic variables. With regard to the children's gender, all validation studies found a similar result: all males reached a higher score than females, both in the general [8,32] and clinical [35] population. Only Reznick and colleagues [8] found no significant impact of the family size variable on FYI score. With regard to the parental variables, only two studies [8,32] evaluated them. Specifically, both Reznick and colleagues [8] and Ben-Sasson and [32] found a negative and significant impact of low maternal educational level on FYI score. For this reason, both validation studies suggested rewriting several items. Furthermore, the study by Reznick and colleagues [8] found a significant and positive impact on FYI score for black mothers, whereas Ben-Sasson and Carter [32] found a significant and positive impact of single status mothers on screening measure score. As suggested by these authors [8,32], these variables could be monitored by researchers and professionals to interpret the FYI score adequately.

With regard to the questionnaire psychometric properties, it was worth noting possible detected similarities and differences between the validation studies. The convergent validity was demonstrated by two studies [32,34]. The first study [34] carried the analysis on a sample of the general population recruited by Reznick' study [8]; the second study [32] analyzed a sample of the Israeli general population. Both validation studies administered the observation and standardized measures to assess the child' autistic traits (ADOS 2—Autism Diagnostic Observation Schedule-Second Edition—and AOSI—Autism Observation Scale for Infants [36]—respectively) and his/her global functioning (MSEL). Furthermore, both validation studies suggested developing a short version of the FYI. Only Turner-Brown and colleagues [34] examined the accuracy of the screening measure applying a Receiver Operating Characteristic (ROC) analysis: they stated that the combined score on Social–Communicative and Sensory–Regulatory Functions domains was the optimal threshold to detect child at risk at 12 months. In addition, Muratori and colleagues [33] evaluated the FYI accuracy on a clinical sample, and they stated that a two-domains approach of social-communicative and total domains was the optimal threshold to detect cases of early-onset autism. Finally, only Reznick and colleagues [8] demonstrated the questionnaire structural validity and carried out an Explorative Factor Analysis (EFA). As anticipated above, the factorial structure was developed according to the results of two different statistical analysis:

the EFA and the Item–Total Correlation (ITC). Nevertheless, this statistical strategy was not adequate to define a factorial structure, and not one validation study carried out a Confirmatory Factor Analysis (CFA).

According to the systematic review findings and the present critical analysis of the validation studies of the FYI, the measure seems to show some promising characteristics and several limitations. The FYI is an effective tool requiring little administration time that can be applied starting from 11 months of life, both in general populations and those at risk. Therefore, the FYI is a cost-effective measure, appropriate for administration to parents during regular well-child visits as part of pediatric surveillance. Finally, according to the longitudinal research [33], the instrument seems to be an efficient measure for detecting behaviours that deviate from those characterising typical development (and, as such, can be a sign of the risk of ASD).

Nevertheless, the above-mentioned studies have several limitations. First, the cross-cultural generalisability of the FYI was studied on Israeli children. One study [32], involving Italian children, used a retrospective design. It is well known that parental memories may influence the quality of data derived through retrospective methods [37]; thus, further studies are needed to study the cross-cultural generalisability of this measure in a non-American sample. Second, the factorial analysis of the FYI [8] has not confirmed a structure based on the expected eight constructs. It should be noted that the authors did not report the results from the EFA (i.e., factor loadings, percentage of variance explained), and the final structure of the questionnaire was derived from a combination of evidence from the item–total correlations and what they theoretically expected to find. Establishing a psychometrically sound factorial structure of the FYI is not a secondary issue since the calculation of the risk cutoff is based on it. Finally, none of the other studies [31–34] analysed the factorial structure of the FYI, but rather took for granted what Reznick and colleagues [8] had found. Thus, further demonstrations of the factorial structure are particularly needed.

Therefore, the general aim of the current study was to conduct a screening of the signs of risk of ASD, applying the FYI on an Italian sample (from the general population) undergoing regular well-child visits as part of pediatric surveillance. The study purposes were to (a) examine the cross-cultural generalisability of the FYI, comparing the Italian findings with those of US and Israeli samples (specifically, comparisons of the analyses of socio-demographic variables, response distributions, and cut-offs); (b) demonstrate the construct validity of the FYI; and (c) demonstrate the internal consistency and structural validity of the FYI.

### **2. Materials and Methods**

### *2.1. Procedure*

The study was carried out in a large urban area in the south of Italy. The Ethical Committee of the Local Public Health Service gave its approval for this research (n◦ 528/8 March 2017). One hundred fifteen paediatricians of the local public health service received via mail a description of the research project, with a request to collaborate with it. Sixty-four of them (55.6%) participated in the research and received instructions for the recruitment of participants. All families treated by those paediatricians with a child born between February and September of 2016 were invited to participate in the study (*n* = 800). They received a description of the research project and signed informed consent. Data collection was conducted when the parents were at the paediatrician's office (in a quiet place before the visit); the paediatrician was not present during the administration of the questionnaire.

### *2.2. Measure*

Socio-Demographic Variables. The first part of the FYI allows identification of the following information: the child's gender, date of birth, weight at birth, order of birth, term birth vs. preterm birth, parents' marital status, their educational level, and their ethnicity. Finally, information was collected as well on who completed the questionnaire (e.g., mother, father, or both). Early identification of signs of risk of ASD. The 63 items of the FYI [8] (Italian translation by Muratori and Narzisi, 2009) allow evaluation of the child's functioning within two domains: Social–Communication and Sensory–Regulatory Functions. Each domain consists of four constructs. The Social–Communication domain includes the constructs of Social Orienting & Receptive Communication (nine items), Social–Affective Engagement (eight items), Imitation (six items), and Expressive Communication (five items). The Sensory–Regulatory Functions domain includes the constructs of Sensory Processing (six items), Regulatory Patterns (four items), Reactivity (three items), and Repetitive Behaviors (eleven items). According to Reznick and colleagues [8], the final score was calculated through a weighted average of the raw score for each construct and domain. a total score was calculated as an average of the two domains, with higher scores indicating higher risk.

### *2.3. Participants*

The convenience sample was composed of 698 returned questionnaires with a response rate of 86.1%. Forty-one questionnaires were excluded from the analyses because they were completed by mothers of children with Down's Syndrome (*n* = 2) or by mothers of preterm children (i.e., born before the 37th gestation week; *n* = 39). Those children were excluded from the sample since the study purpose was to validate the FYI as a Level 1 screening measure administrable to the general population, that is, children not referred for other developmental concerns. Specifically, the two children with Down's Syndrome were excluded from the sample because of their genetic disease. Furthermore, the 39 preterm children were excluded since—as in [8] and [32]—they were too immature at 12 months to be evaluated on social and behavioural functioning.

The final sample was comprised of 657 questionnaires (Figure 1) completed by mothers (69.9%), fathers (5.3%), or both parents together (24.2%) when the children were from 11 to 13 months old (M = 12.4 months; SD = 1 month). Three hundred forty-one of them were boys, 309 were girls. The toddlers' mean weight at birth was 3.32 kg (SD = 0.51; range 3–4.93 kg); 40.3% of the children were first-born, and 43.7% were second-born or more. The mothers' mean age was 33.83 (SD = 5.6; range 18–49), and their educational level was low (up to eight years of education) for 26.9% and high (nine or more years of education) for 73.8%. The fathers' mean age was 37.42 (SD = 6.4; range 19–67), and their mean of the educational level was low (up to eight years of education) for 32.1% and high (nine or more years of education) for 61.3%. The majority of the parents were married (92.8%), whereas 6.4% were single or divorced. The parents were European–White (88.1% of the mothers; 85.7% of the fathers), African (0.6% of the mothers; 1.1% of the fathers), or Asian (1.1% for mothers; 0.3% of the fathers).

**Figure 1.** Flowchart of study sample and design.

### *2.4. Analytic Strategy*

Independent sample t-tests were carried out to analyse the differences in the two domains of the FYI (Social–Communication and Sensory–Regulatory Functions), the total score, and the eight constructs based on the socio-demographic variables. When a difference was found as statistically significant, a Cohen's *d* was reported. To compare the Italian and American (or Israeli) response distributions, a chi-square analysis was run for each item. The null hypothesis (H0), that the response distribution of the Italian and American sample (or Israeli) for each item was not different, is what we aimed to demonstrate. Thus, a nonsignificant chi-square is a demonstration that the distributions are comparable. The analyses were conducted in SPSS v.25.

The data were screened to investigate the missing data distribution, normality distribution, and outliers. Exploratory factor analyses (EFA) and confirmatory factor analyses (CFA) through SEM (Structural Equation Modelling) were carried out in Mplus v.8 applying WLSMV because the data were ordinal. Geomin rotation was applied to the EFAs with the Weighted Least Square Mean and Variance (WLSMV) as estimator since the data were ordinal and missing data were also found. The Kaiser–Meyer–Olkin (KMO) statistic was computed on the 60 items of the FYI to evaluate if the data were suitable the data for the factor analysis.

### **3. Results**

### *3.1. Preliminary Analysis*

Less than 5% of the socio-demographic variables and less than 1.7% for the items of the FYI were missing. Among the latter, those with the highest percentages of missing data were Item 40 (1.7%), Item 5 (1.3%; 'Does your baby seem to have trouble hearing?'), and Item 16 (1.1%; 'Is it easy to understand your baby's facial expressions?'). The 'Little's missing completely at random' test was significant, χ<sup>2</sup> (3367) = 4008.438; *p* = 0.000; this means that missing data were nonrandomly distributed. For this reason, and given the low percentages, they were not imputed. Comparing our missing patterns with those of the US sample [8], only Item 40 ('Do your baby's eyes line up together when looking at an object?') had a similar percentage of missing data (1.7% in the Italian sample and 2% in the American sample). For all the other items, we had less missing data than the US sample.

### *3.2. Generalisability*

Analyses on the socio-demographic variables. The t-tests showed no significant effects by the childbirth order (i.e., first-born = 40.3%; second-born or more = 43.7%) on the FYI domains, the total score, or the eight constructs. With regard to the children's gender, the t-test showed a significant difference on the Reactivity construct. Boys obtained higher scores than girls. Boys reached higher scores also on the two domains and on the total score. Table 1 shows the results of the t-tests, with means and standard deviations.


**Table 1.** Independent-sample t-test by gender on the FYI domains, total score, and constructs.

Note: \* *p* < 0.05.

Considering the parental socio-demographic variables, the t-tests showed differences for maternal educational level and marital status. Specifically, mothers with a low educational level (up to eight years of education), compared to those with high educational level (nine or more years of education), obtained higher scores in the two FYI domains, the total score, and all constructs, with the exception of Social–Orienting and Receptive Communication (part of the Socio-Communication domain) and Regulatory Patterns (part of the Sensory–Regulatory Functions domain) constructs. Table 2 shows the results of these analyses.

**Table 2.** Independent-sample t-tests by maternal educational level on the FYI domains, total score, and constructs.


Note: \* *p* < 0.05.

Mothers without a partner showed higher scores (M = 8.03; ds = 7.90) on the Repetitive Behaviors construct (part of the Sensory–Regulatory Functions domain) than mothers with a partner (M = 5.40; ds = 6.48), *t*(44.876) = −2.109, *p* = 0.041.

Comparisons between distributions. We aimed to demonstrate the null hypothesis (H0), that the percentage of response distribution for each item for the Italian and American (see Table 3) and Israeli (see Table 4) samples was not different. Indeed, the first column of the Table 3 reports the content of the items, and the second to the fifth columns report the percentages for each response for the two samples. **Table 3.** Chi-square comparison between American (AS; Reznick et al., 2007) and Italian (ItS) sample distribution (%).


**Table 3.** *Cont*.


**Table 3.** *Cont*.


Note: IS = Italian Sample; AS = American Sample. The bold line identifies the items with three or four multiple-choice answers.

**Table 4.** Chi-square comparison between Italian (ItS) and Israeli (IS; Ben-Sasson and Carter, 2012) sample response distribution (%).


Note: ITS = Italian Sample; IS = Israeli Sample. The bold line identifies the items with three or four multiple-choice answers.

Comparing the Italian and the American distributions, the χ<sup>2</sup> values were below the critical values (χ<sup>2</sup> 0.05,3 <sup>=</sup> 7.815; <sup>χ</sup><sup>2</sup> 0.05,2 = 5.991); thus, no differences emerged between the two samples. Similarly, comparing the Italian and the (partially available) Israeli distribution, no difference emerged.

Score Distributions and Cutoffs for ASD Risk. In Table 5, we summarized our scores and those obtained by the other international studies. It was not always possible to compare Italian findings with those of the American and Israeli studies since some data were not available in the papers.


**Table 5.** Comparison between the American, Israeli and Italian cutoffs.

\* Ben-Sasson and Carter reported that this value was from a personal communication by Reznick. \*\* This value was not reported in Reznick and colleagues (2007), it was calculated according to Ben-Sasson and Carter (2012). "-" Means the values were not reported in the paper.

Nevertheless, it is worth noting that the American data range was only theoretical and that the Israeli data range was higher than that in the Italian results. The modal value was 0 both in the American and the Italian data; this confirmed that, in the general population, the majority of FYI scores strived towards the lowest score that indicated typical development. With regard to the mean score, it was possible to compare Italian and Israeli data: the first score was lower than the second. The American mean score was not available.

Finally, with regards to the cross-cultural risk score comparison, it is worth noting, as few values were reported by the American authors, that the only two values reported in Table 6 were calculated according to Ben-Sasson and Carter's [31] suggestions. The American and Italian data comparison on risk score on the 95th and 98th percentile showed similar values. Figure 2 shows the distribution of risk score (skewness = 1.53; kurtosis = 4.03) for the Italian sample and the shape inflection corresponding to the score of 17, as found in Reznick and colleagues' [8] study. Comparing Israeli and Italian data, the Italian raw values corresponding to the 95th and 98th percentile were lower than the Israeli ones.


**Table 6.** Correlations among the FYI 8 constructs on the Italian sample. Between parentheses, the results of the correlations yielded in the American sample by Reznick and colleagues (2007).

Note: \* *p* < 0.05; \*\* *p* < 0.01; \*\*\* *p* < 0.001; df: 655.

**Figure 2.** Distribution of First Year Inventory (FYI) raw risk scores in the Italian sample according to the factor structure of Reznick et al. (2007).

According to the other two validation studies on a general population [8,31], scores equal or above the 95th percentile could be applied to detect children at risk for ASD. We decided to apply the mean score on the 95th percentile of the total score, which was 8.15; 32 children in our sample met this risk criterion (which corresponds to 4.87% of the sample). a similar result (4.88%) was found by Ben-Sasson and Carter [31] on the Israeli general population. The families with children under the risk condition were invited for a diagnostic assessment with gold standard measures. The evaluation is in progress, and the children have been followed over time.

### *3.3. Construct Validity*

To investigate the inter-correlations between the two domains and the eight constructs, Pearson r correlations were carried out. Table 6 reports the correlations between the eight constructs and also the correlations found in Reznick and colleagues' [8] study as a comparison. As expected, the four constructs of the Social–Communication domain correlated with each other, as did the four constructs of the Sensory–Regulatory Functions domain. Furthermore, results showed that the Expressive Communication construct (part of the Social–Communication domain) did not correlate with all constructs of the Sensory–Regulatory Functions domain; Social–Affective Engagement (part of the Social–Communication domain) did not correlate only with the Sensory Processing and Regulatory Pattern constructs (part of the Sensory–Regulatory Functions domain). The two domains are correlated as well, r = 0.13, *p* = 0.01.

### *3.4. Internal Consistency and Factorial Analyses*

The Hayes and Krippendorff' kalpha for Social–Communication and Sensory–Regulatory Functions domains were 0.91 and 0.88, respectively. These values were higher than those found by Reznick and colleagues [8] and suggested a moderate consistency among items.

Exploratory factor analyses (EFA) and confirmatory factor analyses (CFA) through SEM (Structural Equation Modelling) were carried out in Mplus v.8 applying WLSMV because the data were ordinal. Geomin rotation was applied to the EFAs. a first-order CFA was performed on the 52 items of the FYI to test the eight-factor structure corresponding to the constructs hypothesised by Reznick and colleagues [8] (Figure 3). The 10 items that did not load for any factor (see Appendix in Reznick et al.'s paper, [8]) and were not inserted into the analysis. a second-order CFA was tested based on the second-order factorial structure estimating the eight constructs (as first level latent factors) and the two domains (as second-level latent factors). For both CFAs, values of the χ2, the CFI (Comparative Fit Index), and the RMSEA (Root Mean Square Error of Approximation) were examined. The two CFAs showed several correlations between items or between constructs with values close or equal to 1, suggesting that the items or the factors should be collapsed.

**Figure 3.** A graphical reproduction of FYI Factor Structure by Reznick et al. (2007).

As both the CFAs failed to estimate acceptable factorial structures, we chose to go back to the EFA. Two factorial structures were tested on the original 61 items: the eight-factor structure, corresponding to the eight constructs, and the two-factor structure, corresponding to the Socio-Communication domain and the Sensory–Regulatory Functions domain. The χ2, the CFI, and the RMSEA were examined for both. The items were progressively excluded if the factor loadings loaded for two or more factors or none of them. The comparison between the eight-factor and the two-factor structure showed that the latter was the best fitted. Thus, we performed a further test via CFA.

The first order CFA on the 52 items yielded a moderate–low fit of the data, with a significant χ<sup>2</sup> (1196) = 2214.53, *p* < 0.001, and CFI = 0.83, RMSEA = 0.036 (LO90% = 0.034, HI90% = 0.038). Similarly, the second order CFA showed moderate-low fit of the data, with a significant χ<sup>2</sup> (1214) = 2238.99, *p* < 0.001, and CFI = 0.83, RMSEA = 0.036 (LO90% = 0.034, HI90% = 0.038).

The EFA on the 61 items estimating the eight-factor structure was well fitted, χ<sup>2</sup> (1318) = 1585.34, *p* < 0.001, CFI = 0.96, RMSEA = 0.018 (LO90% = 0.014, HI90% = 0.021). However, 43 items were excluded because the factor loadings loaded for two or more factors and the remaining items loaded for four factors instead of the hypothesised eight, and those four factors did not correspond with the theoretical model hypothesised by Reznick and colleagues [8].

For these reasons, the EFA estimating the two-factor structure was preferred and reached a moderate fit of the data, χ<sup>2</sup> (1651) = 2940.15, *p* < 0.001, CFI = 0.79, RMSEA = 0.034 (LO90% = 0.032, HI90% = 0.036). Nineteen items (Items 4, 5, 6, 16, 27–29, 31, 32, 39, 41, 49–56) were excluded from the subsequent analysis because the factor loadings loaded for two or more factors. After exclusion of those items, the subsequent fourth EFA reached moderate fit of the data, χ<sup>2</sup> (739) = 1185.47, *p* < 0.001, CFI = 0.92, RMSEA = 0.03 (LO90% = 0.027, HI90% = 0.033). Items 7, 14, 44, and 48 did not load for any factor and were subsequently deleted. The third EFA reached moderate fit of the data, χ<sup>2</sup> (593) = 996.58, *p* < 0.001, CFI = 0.92, RMSEA = 0.03 (LO90% = 0.029, HI90% = 0.036), and again, Items 11 and 57 did not load for any factor and were subsequently deleted. a final EFA was carried out with the remaining items (Factor 1: *n* = 15 items; Factor 2: *n* = 16 items), again showing moderate fit of the data, χ<sup>2</sup> (526) = 921.79, *p* < 0.001, CFI = 0.92, RMSEA = 0.034 (LO90% = 0.030, HI90% = 0.037). Table 7 shows the final EFA solution. Factor 1 contains items corresponding to the Social–Communication Domain, Factor 2 to the Sensory–Regulatory Functions Domain, so all the items loaded for the expected factor.


**Table 7.** Exploratory Factor Analysis (EFA) results (standard errors between parentheses).


**Table 7.** *Cont*.

The final EFA structure was tested via CFA. The two-factor structure showed moderate fit of the data, χ<sup>2</sup> (433) = 672.72, *p* < 0.001, CFI = 0.95, RMSEA = 0.029 (LO90% = 0.026, HI90% = 0.033). Item 9 had low factor loading with the factor and was subsequently deleted. The two factors were weakly correlated, r = 0.15, *p* = 0.045. The final CFA was carried out showing good fit of the data, χ<sup>2</sup> (404) = 617.699, *p* < 0.0001, CFI = 0.95, RMSEA = 0.028 (LO90% = 0.024, HI90% = 0.033). Figure 4 shows the factor structure obtained by the CFA.

**Figure 4.** FYI Structure according to Confirmatory Factor Analysis (CFA) run in this study.

After those analyses, we re-examined our data according to the new factorial structure. The total score ranged from 0 to 18.39, with a mean of 3.27 (SD = 3.04), a median of 2.17, and a distribution shaped as a chi-square (skewness = 1.49; kurtosis = 2.86). The t-tests showed a significant difference by children's gender, *t*(648) = 2.062, *p* = 0.040, with boys reaching a higher total score (M = 3.48; ds = 3.06) than the girls (M = 2.99; ds = 2.95). There were no significant differences by childbirth order or by parents' marital status. In contrast, the t-tests showed significant differences by educational level on the Sensory–Regulatory Functions domain, *t*(208.185) = 3.537, *p* < 0.0001 and on the total score, *t*(206.433) = 3.755, *p* < 0.0001. Specifically, mothers with a low educational level showed higher scores (Sensory–Regulatory Functions domain: M = 6.70; ds = 5.98; total score: M = 4.18; ds = 3.51) than mothers with a high educational level (Sensory–Regulatory Functions domain: M = 4.79; ds = 4.87; total score: M = 2.99; ds = 2.82). Finally, the risk cutoff on the 95th percentile of the total score corresponded to a score of 9.14.

### **4. Discussion**

The main purpose of this study was to conduct an early screening of the signs of risk of ASD, applying the FYI as part of pediatric surveillance on an Italian sample from the general population. We examined the cross-cultural generalisability of the screening measure, comparing the Italian scores with those of the two validation studies conducted on a general population [8,31]. The other two aims of the research were to test the construct validity of the FYI and to demonstrate its internal consistency and structural validity.

The combination of all the results mentioned represents a demonstration of the generalisability and stability of the measure across cultures. First of all, we considered the role played by the socio-demographic variables and compared the present findings with those found with the American and Israeli samples. Significant differences were found by children's gender, with boys showing higher scores than girls for the Reactivity construct (part of the Sensory–Regulatory Functions domain).

Considering the parental variables, it is worth noting that in the other two validation studies on the general population [8,31], among the socio-demographic variables considered, the authors examined whether maternal ethnicity influenced the scores of the FYI (Reznick et al., 200). Those differences were not tested on the Italian sample, because all the parents were European–White. Reznick and colleagues [8] and Ben-Sasson and Carter [31] also considered the educational level and marital status of the mothers.

Similarities between the Italian and American and Israel samples were also found for the maternal educational level and marital status. As Reznick and colleagues [8] and Ben-Sasson and Carter [31] found, a low educational level was associated with higher FYI scores compared to a high educational level. One possible explanation is that mothers with a low educational level may interpret several atypical behaviours as common because they misunderstood the meaning of the item [31]. In particular, the items of the Sensory–Regulatory Functions domain describe atypical behaviours, as they would be 'positive' (i.e., presence of a behaviour) instead of 'negative' (i.e., absence of a developmentally expected behaviour). For example, the item, 'Is your baby content to play alone for an hour or more at a time?' can be misleading because the mothers may interpret as positive the fact that child plays quietly alone for long periods (i.e., presence of a behaviour).

Moreover, we found that single mothers reported higher FYI scores on the Repetitive Behaviors construct (part of the Sensory–Regulatory Functions domain) than did married mothers. Ben-Sasson and Carter [31] found a similar result for the Sensory–Regulatory Functions domain. The explanations of these results may be twofold. First, the single mothers did not have a partner with whom they could discuss concerns about the child's development; thus, they could interpret the child's Sensory–Regulatory behaviours as atypically. Second, the child's self-regulation process may be affected by the absence of the father [38].

As a further demonstration of the cross-cultural generalisability of the FYI, we found similar patterns of response for each item, meaning that there were no differences across cultures in the way in which parents of children from 11 to 13 months of age replied to the questions. This result highlighted that targeted behaviours evaluated by the FYI were identifiable in a similar manner across different cultures. Thus, this property allows the detection of typical and atypical behaviours that appear to be cross-culturally invariant.

Finally, the Italian results were similar to the American findings for the total risk score calculated on the 95th and 98th percentile, and both were lower than the risk scores calculated on the Israeli sample. As Ben-Sasson and Carter [31] suggested, this could be due to the dysregulation [39] and the stress [40] endured by Israeli children growing-up in a stressed society faced with trauma and terror daily.

Nevertheless, the percentage (32%) of children detected at risk (with a total score equal or above the 95th percentile) in the Italian and the Israeli samples was similar (these data were not available in Reznick and colleagues' [8] study).

The second aim examined the FYI construct validity. The positive and significant correlations between the two domains of the instrument (Social–Communication and Sensory–Regulatory Functions) and between constructs highlighted a good construct validity of the measure, as found by Reznick and colleagues [8].

Since no previous studies on the FYI have validated its factorial structure, the purpose of the present study was to give insight on this property. In this vein, the Confirmatory Factor Analysis is a crucial and strategical analysis demonstrating the structural validity of a measure. Therefore, we firstly carried out a CFA on the theoretical structure hypothesized by Reznick and colleagues [8]. Our analyses did not confirm the structure of the scale organised on the eight hypothesised constructs. It should be noticed that Reznick and colleagues [8] also struggled to find a stable factorial solution for their data and decided to shape the final eight constructs through the item–total correlations and the expected thematic content of the items. As the second step in our study, two second-order latent factors, corresponding to the two main domains of the FYI, were estimated through CFA. Even in this case, the results did not support the hypothesised structure. Therefore, we decided to explore the structure of our data with a set of five nested EFAs in which several items were found as critical, because of loading more than one factor or because of not showing the expected factor loading (i.e., > 0.30), and deleted step by step. The final explorative factorial structure comprehended 30 items, which are coherently distributed in the Social Communication and Sensory–Regulatory Functions domains. a CFA confirmed this structure and allowed the estimation of a short version of the FYI, which was suggested by Turner-Brown and colleagues [34] as one point to be developed by future research after their study. The short version of the questionnaire makes its administration easier and faster and allows applying the questionnaire during systematic screening evaluations on the general population.

The short version of the FYI evaluated the two main core areas of risk for ASD, in which the main symptoms are included, as suggested by Reznick and colleagues [8] and the DSM-5 [1]. Most of the items of the short version assess the social and communicative deficit (Factor 1), focusing on the evaluation of receptive communication and social engagement. The others evaluate the first factor focussing on child's imitative capacity, and his/her expressive communication. Furthermore, the second factor estimated in the short version (Sensory–Regulatory Functions domain) evaluates the presence of repetitive behaviours and the hypo- or hypersensitivity of the child to sensory stimuli. The evidence on the FYI short version highlighted the expected results considering both the parental and the children's socio–demo variables, as found in the other validation studies [8,32] who applied the full version of FYI.

The total score calculated on the final structure of the scale showed significant difference by gender, with boys reaching higher total scores compared to the girls, confirming the American and Israeli findings and the gender ratio of ASD (4:1) [1]. Even with the total score calculated on the short version (FYI-30), low parental educational level was associated with higher total score compared to the opposite condition, whereas marital status was not significant. Therefore, the estimated short version seems to represent the two core symptoms of the ASD and, at the same time, maintains the impact of the socio-demographic variables on the total score as found by previous research.

### **5. Limitation**

The main limitation of the present study is the cross-sectional design. Longitudinal studies on the general population are required to demonstrate the accuracy of the FYI, its PPV (i.e., positive predictive value) and NPV (i.e., negative predictive value), and ability to detect signs of risk of ASD. Future studies, starting from our results on the FYI short version, should consider the diagnostic outcome evaluation, through gold-standard measures, and the convergent validity. Specifically, the evaluation should be focused on the severity of the autistic traits, the global child development, and characteristics of attention-selectivity processes [41]. a prospective study is currently ongoing with a longitudinal evaluation of children considered to be at risk at 11–13 months of life and evaluated one and two years later. Furthermore, other studies should further demonstrate the short version structure of the FYI developed in this study. The second limitation is related to the relatively low response rate of the professionals in our study, although it is similar to what was found by others [8,32]. It should be noticed that the low response rate of the professionals did not correspond to a similar low parental response rate. Indeed, when the paediatrician participation was obtained, on their side, parents easily agreed to be participants. It is highly likely that parental participation depends on the quality of their relationship with the paediatrician, as found by others [42,43]. This also means that a way to establish a continuous screening for children's mental health and speed up early diagnosis and intervention is increasing health professionals' awareness of that aspect.

### **6. Conclusions and Implication**

According to our results, the FYI is a valid and reliable screening tool for Italian children. Results for the current study stimulate further research in the field of cross-cultural validity and generalisability of the FYI and other measures for the early identification of signs of risk of ASD.

Our findings highlighted some positive features of the FYI and, at the same time, several others that should be further developed. On the one hand, the analyses have shown the cross-cultural stability and generalisability of the FYI as well as its construct validity. Therefore, the FYI is a reliable tool that may be administered in another cultural context from the American and Israeli ones.

On the other hand, modest demonstrations of internal consistency were found, as also confirmed by the factorial analyses. As for the latter, the hypothesised structure (see [8] for details) did not receive appropriate support, showing poor fit of the data with several correlations between items with values close or equal to 1. The alternative analyses carried out revealed a structure organised on the two main core symptoms of ASD, also identified by Reznick and colleagues [8] in their original version of the FYI, based on a short version of the questionnaire. Our analyses demonstrated that the factorial validity of the FYI requires further demonstration. This notwithstanding, the short version of the FYI may lead to a cost-effective and easy-to-administer instrument to be used by paediatricians during their pediatric surveillance on the general population. The early detection of atypical developmental trajectories may support medical decision-making on further steps for referral of the child to an early diagnostic assessment (which may enable early intervention when needed; [43–48].

**Author Contributions:** Conceptualization, F.L. and S.P.; methodology, A.L., F.L., and S.P.; software, A.L. and S.P.; formal analysis, A.L. and S.P.; investigation, A.L.; data curation, A.L. and S.P.; writing—original draft preparation, A.L., F.L., and S.P.; supervision, F.L. All authors have read and agreed to the published version of the manuscript.

#### **Funding:** This research received no external funding.

**Acknowledgments:** We are grateful to the two associations of paediatricians, the Italian Federation of Medical Pediatricians (FIMP) and the Italian Confederation of Pediatrics (CIPe), for their collaboration. We are also very grateful to all paediatricians and parents for their participation. We are grateful to the psychologists Filomena De Lumè, Luigia Duma, Valentina Garrapa, Emilia Perrone, Viviana Vetrugno, Annamaria Vizzi, Giuseppe Antonioli for their collaboration for data collection.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Inflammatory Biomarkers are Correlated with Some Forms of Regressive Autism Spectrum Disorder**

**Margherita Prosperi 1,2,**†**, Letizia Guiducci 3,**†**, Diego G. Peroni 2, Chiara Narducci 4, Melania Gaggini 3, Sara Calderoni 1,2, Ra**ff**aella Tancredi 1, Maria Aurora Morales 3, Amalia Gastaldelli 3,\*, Filippo Muratori 1,2 and Elisa Santocchi <sup>1</sup>**


Received: 12 November 2019; Accepted: 9 December 2019; Published: 11 December 2019

**Abstract:** *Background*: Several studies have tried to investigate the role of inflammatory biomarkers in Autism Spectrum Disorder (ASD), and their correlations with clinical phenotypes. Despite the growing research in this topic, existing data are mostly contradictory. *Methods*: Eighty-five ASD preschoolers were assessed for developmental level, adaptive functioning, gastrointestinal (GI), socio-communicative and psychopathological symptoms. Plasma levels of leptin, resistin, plasminogen activator inhibitor-1 (PAI-1), macrophage chemoattractant protein-1 (CCL2), tumor necrosis factor-alfa (TNF-α), and interleukin-6 (IL-6) were correlated with clinical scores and were compared among different ASD subgroups according to the presence or absence of: (i) GI symptoms, (ii) regressive onset of autism. *Results*: Proinflammatory cytokines (TNF-α, IL-6 and CCL2) were lower than those reported in previous studies in children with systemic inflammatory conditions. GI symptoms were not correlated with levels of inflammatory biomarkers except for resistin that was lower in ASD-GI children (*p* = 0.032). Resistin and PAI-1 levels were significantly higher in the group with "regression plus a developmental delay" onset (Reg+DD group) compared to groups without regression or with regression without a developmental delay (*p* < 0.01 for all). *Conclusions*: Our results did not highlight the presence of any systemic inflammatory state in ASD subjects neither disentangling children with/without GI symptoms. The Reg + DD group significantly differed from others in some plasmatic values, but these differences failed to discriminate the subgroups as possible distinct ASD endo-phenotypes.

**Keywords:** autism spectrum disorder; regression; cytokines; PAI-1; neuroinflammation; gastrointestinal

### **1. Introduction**

To date, the understanding of the underlying molecular mechanisms of some metabolic or neurological diseases and the deepening of knowledge on the role of inflammation in these disorders have radically changed our understanding of their etiology [1,2]. Alzheimer's (AD) and Parkinson's disease, type 1 and type 2 diabetes, and obesity are just some of the pathologies for which a well-defined role of inflammation has been identified, with consequent possible therapeutic implications [3,4]. For example, activated astrocytes and microglia are characteristically found in abundance near neurons and plaques in AD [5] and the block of the activation of insulin signaling receptors caused by the chronic exposure of pro-inflammatory mediators in β-cells of pancreatic islets has been evidenced in the pathogenesis of insulin resistance which underlies many metabolic diseases [6,7].

Recently, the contribution of immune dysregulation has been described as a common feature of the autism spectrum disorder (ASD), and alterations in circulating cytokine levels have been repeatedly reported [8,9]. ASD are neurodevelopmental disorders characterized by persistent social communication difficulties with concurrent restricted interests, repetitive activities and sensory abnormalities [10]. The etiopathogenesis of idiopathic ASD is complex and not yet fully elucidated, but it is widely recognized that genetic liability and environmental factors interact in producing early alteration of structural and functional brain development, responsible for ASD symptoms [11,12]. Despite a systematic review about proinflammatory markers in more than 3900 children and/or adolescents with neuropsychiatric disorders including ASD [13] found preliminary evidence for the role of inflammation and pro-inflammatory state in these conditions, until now conflicting and irreproducible findings have been detected in various studies.

Some authors have proposed interleukin (IL)-6, tumor necrosis factor-alpha (TNF)-α, and macrophage chemoattractant protein-1 (CCL2) as potentially involved in brain inflammation at least in a subgroup of subjects with ASD [14]. A recent meta-analysis of 25 studies revealed a higher concentration of pro-inflammatory cytokines interferon (IFN)-γ, IL-1β, IL-6, and TNF-α in children with ASD compared with controls [9]. Increased levels of IL-6 and IL-8 were found to be predictive biomarkers for ASD risk in a study analyzing circulating cytokine patterns from neonatal blood [15]. High levels of IL-6 in the brain could determine alterations of synapse formation, dendritic spine development, and neuronal circuit balance [16], while in plasma they have been associated with increased stereotypical behaviors and with regressive forms of ASD [17]. Conversely, TNF-α has a critical role in regulating synaptic strength and plasticity [18], and his levels have been positively correlated with ASD severity [19]. High CCL2 levels could be instead considered as a signal of microglia/astroglia activation [20], and have been associated with higher aberrant behavior scores and more impaired adaptive functioning [21].

Similarly, GI problems that frequently occur in ASD subjects seem to be caused by inappropriate immune activation and pro-inflammatory processes of the digestive tract [22]. It has been shown that the level of stress-responsive cytokines, like IL-6 and TNF-α, are increased both in ASD subjects [17] and in the general population in association to gastrointestinal (GI) symptoms [23,24], pointing to a link between peripheral inflammation and neuroinflammation. Particularly, high levels of TNF-α can influence the intestinal epithelial barrier possibly contributing to GI problems [25] and intestinal permeability, and also to ASD onset as recently suggests by the "leaky gut" hypothesis [26]. The myeloid dendritic cells, which produce among others TNF- α and IL-6, have been associated with increased GI symptoms in ASD as well as increased amygdalar volume and regressive autism [27]. More recently, other authors [22,28] did not confirm an association between the symptoms of the lower GI tract and levels of TNF-α or IL-6, however their levels were correlated with irritability, socialization and intelligence in ASD subjects.

Besides, a particular type of cytokines called adipokines seems to be implicated in the pathogenesis of inflammatory central nervous system (CNS) disorders and ASD [29] despite the findings obtained so far are mostly controversial. Adipokines, or adipocytokines, are active proteins secreted by white adipose tissue with functions similar to hormones in inter-organ communication [30] and their dysregulation has been implicated in obesity, type 2 diabetes, cardiovascular disease and recently, in peripheral tissue insulin resistance and inflammation [31]. Leptin, adiponectin and resistin are the only three molecules that belong exclusively to the class of adipokines and they have been studied in a limited number of researches concerning autism. Increased levels of leptin, decreased levels of resistin and a negative correlation between the levels of adiponectin and the severity of social impairment were

found in the plasma of ASD subjects vs. controls [29]. Previously, Blardi et al. [32,33] found higher levels of leptin in patients with Rett syndrome in comparison with healthy female subjects, as reported by Ashwood et al. [34] in patients with autism compared to typically developing controls. Leptin dysregulation has been proposed as a mechanism of psychopathology associated with mental health disorders [35], and elevated circulating leptin was consistently found in childhood neurodevelopmental disorders including ASD [34].

Resistin has been implicated in the pathogenesis of several inflammatory CNS disorders [36] and its levels are related to immune changes in autistic subjects: it has been shown that proinflammatory cytokines may increase the expression of messenger-RNA resistin [37] with a positive correlation between increasing resistin levels and inflammatory serum cytokines [38]. A recent case-control study [39] found that resistin levels were increased in ASD subjects compared to healthy controls. To date, no studies have investigated differences in adipokines' levels in ASD subjects with or without GI symptoms.

Distally regulated by some cytokines (i.e., IL-6, IL-1, and TNF-α), the plasminogen activator inhibitor-1 (PAI-1) seems to directly influence brain functions causing a neuronal dis-connectivity due to abnormal neuronal migration [40]. PAI-1 may regulate microglial migration and phagocytosis in an autocrine or paracrine manner playing an important role in the regulation of brain microglial activities in health and disease [41]. Moreover, his locus in human maps very close to or within a region in chromosome 7 linked to autism. No association was found between the presence of ASD and a particular polymorphism of the PAI-1 gene promoter that affects the PAI-1 plasma levels [40].

This pilot study aims (i) to investigate the plasmatic levels of several proinflammatory molecules (TNF-α, IL-6, CCL2, leptin, resistin, and PAI-1) in preschoolers with ASD; (ii) to explore the correlation between their plasmatic levels and behavioral profiles in preschoolers with ASD to detect possible specific subgroups within the ASD heterogeneity.

### **2. Materials and Methods**

### *2.1. Participants*

A total of 85 ASD preschoolers were included in the study and recruited from November 2015 to February 2018 at the ASD Unit of the IRCCS Stella Maris Foundation (Pisa, Italy), a tertiary care university hospital during a clinical trial on the efficacy of probiotic supplementation in ASD preschoolers [42]. In the present study baseline clinical and biochemical data of recruited subjects were investigated.

ASD diagnosis was performed according to Diagnostic and Statistical Manual of Mental Disorders (DSM)-5 [10] criteria by a multidisciplinary team. Exclusion criteria were brain anomalies; neurological syndromes/focal neurological signs; anamnesis of birth asphyxia, severe premature birth/perinatal injuries; epilepsy; significant sensory impairment; diagnosis of organic GI disorder or coeliac disease; special diets; recent any-known infections that could influence circulating cytokines.

All children had a comprehensive evaluation including Autism Diagnostic Observation Schedule-2 (ADOS-2) [43], Griffiths Mental Development Scales-Extended Revised (GMDS-ER) [44], Vineland Adaptive Behavior Scales-Second edition (VABS-II) [45], Child Behavior CheckList 1.5-5 (CBCL 1.5-5) [46], Repetitive Behavior Scale-Revised (RBS-R) [47], Social Communication Questionnaire (SCQ) [48]. The "Overall Level of Non-Echoed Spoken Language" item (A1 score) of the ADOS-2 was used to differentiate non-verbal (those with absent language or less than 5 words) from verbal children: 39 participants (46%) were verbal and 46 (54%) were non-verbal. Information about pharmacological treatments and food supplements in the previous 3 months were collected: parents reported an acute or episodic administration of antibiotics (28.2%), probiotics (8.2%), NSAIDs or paracetamol (14.1%), steroids (8.2%), other drugs without effects on GI symptoms (36.5%), and a chronic administration of osmotic laxatives (12.9%). None of the enrolled subjects used psychotropic drugs.

The demographic and clinical characteristics and a complete description of the tools of all participants and in no-verbal vs. verbal groups are reported in Table 1.


**Table 1.** Clinical characteristics of the total sample and in non-verbal vs. verbal group.

<sup>a</sup> ADOS-2 is a semi-structured assessment of communication, social interactions, play, imagination, and stereotyped or repetitive behaviors used as the gold standard tool for the diagnosis of ASD. Higher ADOS-2 CCS scores indicate greater severity of autism (range of possible scores for Total, Social Affect and Restricted and Repetitive Behavior is 1–10). <sup>b</sup> GMDS-ER are a developmental assessment procedure including five different subscales. We used the Performance subscale to measure the non-verbal skills of each child. Higher scores indicate greater non-verbal abilities. Scores around 100 indicate normal non-verbal skills; scores below 70 indicate a developmental delay of non-verbal skills. <sup>c</sup> VABS-II is a parent interview that assesses adaptive functioning in different daily skills. Higher scores indicate greater adaptive skills, scores around 100 indicate a normal adaptive functioning and scores below 70 indicate a delay with respect to age. <sup>d</sup> CBCL 1.5-5 is a parent-report questionnaire that includes 100 statements about the child's behaviors summarized into three summary scales (Internalizing, Externalizing and Total Problems). Besides, we have used the Aggressive Behavior, the Sleep Problems, the Attention Problems and the Attention Deficit/Hyperactivity (ADHD) Problems Scales of this tool as suggested by previous works on this argument. A T-score of 64 and above for summary scales, and a T-score of 70 and above for the other scales, are generally considered clinically significant. Values between 60 and 63 for summary scales, or between 65 and 69 for the other scales, identify a borderline clinical range. <sup>e</sup> RBS-R is a questionnaire completed by parents about the presence of a broad spectrum of repetitive behaviors. Higher scores indicate greater severity (range 0–114). A two-factor solution scoring of RBS-R was also adopted for this study: a Low-Level Index (composed of items pertaining to Stereotyped, and Self-Injurious subscales) and a High-Level Index (composed of items related to Compulsive, Ritualistic, Sameness and Restricted Interests Behaviors subscales). <sup>f</sup> SCQ is a 40-item parent-report screening measure evaluating the symptoms associated with ASD. We used the form "last three months", completed by parents concerning the child's last three months of life. Higher scores indicate greater severity (range 0–39) with a threshold of 15 compatible for a relevant impairment of social communication (some studies consider 9 in children younger than four years old). \* Age adjustment is not due for ADOS-2 CCS, GMDS-ER and VABS-II since they are already standardized to compare subjects with different chronological ages. Abbreviations (alphabetic order): ADOS-2 Autism Diagnostic Observation Schedule-2; BMI Body Mass Index; CBCL 1.5-5 Child Behavior Checklist 1.5-5; CSS Calibrated Severity Score; GMDS-ER Griffiths Mental Development Scales-Extended Revised; n.a. not applicable; NS not significant; RBS-R Repetitive Behaviors Scale Revised; SCQ Social Communication Questionnaire; SD standard deviation; VABS-II Vineland Adaptive Behavior Scales-II.

To evaluate the presence of GI symptoms we used a modified version of the GI Severity Index (GSI [49]) splitting the subjects into two groups (GI vs. No-GI). GSI is a score designed to identify signs and symptoms of GI distress commonly reported by parents of children with ASD including nine variables, the first six exploring specific GI symptoms (constipation, diarrhea, stool consistency, stool smell, flatulence, abdominal pain) and three exploring unexplained daytime irritability, night-time awakening, and abdominal tenderness. A total score of 4 and above (with at least 3 score points from the first six items) was considered clinically significant for the classification of a subject within the GI group.

Moreover, all preschoolers were divided into regressive or non-regressive (early-onset -EO-ASD-) autism based on the presence/absence of a history of loss of competences such as language or social competences [50]; children belonging to regressive group were further divided in those with regression plus a previous developmental delay (Reg + DD) and those without a previous developmental delay (Reg − DD). According to Kern et al. [51], "regression *plus* developmental delay" was defined as a significant lag in the appearance of normal developmental milestones with a later loss of previously acquired skills.

This study was carried out according to the standards for good ethical practice and with the guidelines of the Declaration of Helsinki. The study protocol was approved by the Pediatric Ethics Committee of the Tuscany Region (Approval Number: 126/2014). Written informed consent from a parent/guardian of each participant was obtained.

### *2.2. Blood Sample Collection*

A fasting blood sample (3 mL for each child) was collected in Ethylenediamine tetraacetic acid (EDTA) tube to perform the cytokines quantitative analysis. We didn't use pain patch before the sampling. Each tube was centrifuged for 10 min at 3500 rpm and all the plasma samples were stored at −80 ◦C until required the bio-humoral investigations

### *2.3. Cytokine Analysis*

The cytokines were measured directly in the plasma through specific immunometric tests (MILLIPLEX MAP, human-magnetic bead panel, Millipore Corporation, Billerica, MA, USA) using an integrated multi-analyte detection platform (high-throughput technology Magpix system, Luminex xMAP technology, Luminex, Austin, TX, USA)

Each sample was analyzed in duplicate. In each one, a sample was analyzed as a quality control. Inter-assay variability was evaluated using two samples at different concentrations and was <10%.

### *2.4. Statistical Analysis*

Descriptive statistics were computed for selected demographic variables across diagnostic groups. Contingency tables were used to perform the frequency analysis. Since the molecule's values were not normally distributed, we used log-transformed values with parametric statistic tests and non-parametric tests to compare GI vs. No-GI subjects (Mann-Whitney test) and to compare EO ASD vs. Reg-DD vs. Reg + DD (Kruskall-Wallis test) for all the selected molecules.

Correlation and regression analysis were computed to study the relationship between the molecules and the identified clinical parameters. Findings with *p* value <0.05 were considered significant. StatView software (version 5.0.1; SAS Institute, Abacus Concept Inc., Berkeley, CA, USA) was used for data analyses. To discriminate different subgroups of ASD children based on biomarker levels, we performed Principal Component Analysis (PCA) using as correlated variables: sex, BMI, age, and cytokine levels (TNFa, IL6, CCL2, leptin, resistin and PAI 1). After log transformation and auto scaling (e.g., mean-centered and divided by standard deviation of each variable) PCA was performed using MetaboAnalystR 1.0.3 (Xia Lab, McGill University, Montreal, Canada). We checked quality control of samples using PCA that allowed us to label the 85 samples as outlier so it was excluded from downstream analysis.

### **3. Results**

Thirty children (35%) were in the GI group and 55 (65%) in the No-GI group. Among the 30 GI subjects, 20 children (67%) were in the non-verbal group, whereas among the 55 No-GI, 26 children (47%) were in the non-verbal group. No statistically significant differences were found in the prevalence of GI subjects between verbal and non-verbal groups (*p* = 0.086). As concerns sex distribution, no differences were found in the prevalence of females in GI versus No-GI groups neither verbal versus non-verbal groups (*p* = 0.560 and *p* = 0.804, respectively).

As concerns clinical variables, there were no significant differences between the GI and the No-GI groups, with the exception of the Global Score of the RBS-R (60.24 ± 20.77 vs. 38.12 ± 27.06; *p* = 0.0016), the Internalizing and Total problem scores of the CBCL (all significantly higher in the GI group than in the No-GI group: 67.48 ± 7.80 vs. 62.06 ± 9.04, *p* = 0.0065 and 65.35 ± 10.02 vs. 60.62 ± 10.30, *p* = 0.0469, respectively), and of the Communication and Daily Living adaptive scores of the VABS (significantly higher in the No-GI group than in the GI group: 45.47 ± 15.22 vs. 54.46 ± 18.80 *p* = 0.0274 and 61.13 ± 14.29 vs. 69.07 ± 17.51 *p* = 0.0365, respectively).

As concerns proinflammatory cytokines levels, the single and the mean values in the total sample and in each subgroup are reported in Table 2. We did not find significant differences in the levels of plasmatic cytokines between GI and No-GI group except for resistin levels (*p* = 0.032). No difference in plasma biomarker levels was found between non-verbal and verbal groups.

Regarding the onset of autism, the mean values of cytokines were not statistically significant different between EO-ASD and regressive subgroups. Nevertheless, comparing cytokines levels in the EO-ASD subgroup with the two types of regressive preschoolers (with and without DD), resistin and PAI-1 levels were statistically significant higher in the Reg + DD group than in the other two groups, the EO-ASD and the Reg-DD ones (*p* < 0.01 for all).

Finally, after the correlation analysis between each molecule and all the clinical parameters, CCL2 levels negatively correlated with CBCL1.5-5 Internalizing and Total problems (*p* = 0.0003, *R* = 0.383 and *p* = 0.013, *R* = −0.272, respectively) and with RBS-R total scores (*p* = 0.05, *R* = 0.21), and positively correlated with VABS-II Motor Skills (*p* = 0.019, *R* = 0.25). TNF-α and PAI-1 levels negatively correlated with age (*p* = 0.0005, *R* = −0.37 and *p* = 0.024, *R* = −0.25, respectively); Leptin levels positively correlated with Body Mass Index (*p* = 0.002, *R* = 0.34) and negatively correlated with CBCL1.5-5 Internalizing problems (*p* = 0.0086, *R* = −0.29).

PCA analysis showed that the variability within the components explains the subdivision in clusters (No-GI vs. GI and EO-ASD vs. Reg − DD vs. Reg + DD) with a low percentage (PC1 = 21.3% and PC2 = 19.0%), indicating that the two and three groups respectively are not partially separated but overlapped (Figure 1).

**Figure 1.** In the left plot, the Principal Component Analysis in gastrointestinal (red) and non-gastrointestinal subjects (green) is presented; in the right plot the PCA based on the ASD onset is presented: subjects with early-onset in red, regression without a previous developmental delay in green, regression plus a previous developmental delay in blue.


 loss of competences; GI: gastrointestinal; IL-6: interleukin-6; PAI 1: Plasminogen Activator Inhibitor-1; Reg – DD: regression without a previous developmental delay; Reg + DD: regression with a previous developmental delay; SD: standard deviation; TNF-α: Tumor Necrosis Factor-alpha.

### *Brain Sci.* **2019**, *9*, 366

**Table 2.**

Comparisons

 between the cytokine levels in GI vs. No-GI groups, in EO ASD (a) vs. Reg-DD (b) vs. Reg+DD (c) subgroups and No-Verbal vs. Verbal

### **4. Discussion**

Our study fits within the complexity and the heterogeneity of studies that examine inflammation and immunity dysfunctions in ASD subjects, moving the field forward into the investigation of biological biomarkers to discriminate possible endophenotypes. The narrow age range considered, the detailed clinical characterization with specific and gold-standard tools for ASD evaluation, and an enough large sample represent the strengths of the study.

First, we found that the single and the mean values of our cytokines were lower than those expected in subjects with systemic inflammation [52–54]. These findings are in agreement with a part of the literature on this topic in which there is an absence of any atypical profile in the expression of relevant plasma cytokines both within ASD subjects and in comparison with TD children [55]. Regarding plasmatic cytokines, it should be highlighted that in literature the reference values and in particular those relating to the pediatric age, to date, are not definitively characterized. Despite our attempt to define specific subgroups based on cytokines levels and anthropometric measures using PCA, in our sample different endophenotypes were not identified. These results exclude the possibility that bringing all cases together in a single ASD group could have hidden significant results in one specific subgroup of preschoolers, as previously hypothesized [56,57]. Consequently, our findings do not support the use of anti-inflammatory therapies in ASD children, not even in a specific subgroup of ASD subjects as previously suggested [58].

Second, we did not observe significant differences in the levels of circulating cytokines between GI and No-GI ASD children, except for resistin. Notably, there is too scant relevant research on this topic in ASD subjects [29,39] to draw valid and accurate conclusions. Thus, the role of adipokines needs further studies, in particular, in correlation with GI symptomatology in ASD considering also the influence of fat mass in plasmatic levels of adipokines. These findings suggest that the frequently reported GI symptoms in ASD children seem to be independent from an inflammatory condition, confirming a not yet clarified meaning of these symptoms [59]. Previously, only a modest relationship between GI symptoms and TNF-α levels was detected [17,28], in one case [28] in significantly older subjects (school-aged children and adolescents) than ours. Specifically, when Ferguson et al. [28] considered only inferior GI symptoms (as we did) they did not identify any statistically significant correlations, in line with the findings that TNF-α levels are independent from the presence of GI symptoms [22,60]. Some authors [61–64] have measured the presence of cytokine-producing cells directly in the bowel of subjects with ASD, and found a local high level of these cells in patients with GI symptoms, supporting a local role of the inflammatory cytokines in altering intestinal epithelial barrier and thus in contributing to GI symptoms. Besides, we confirm our previous findings showing that ASD subjects with GI problems have worse clinical functioning than ASD subjects without GI problems, independently from the severity of autistic symptoms [65].

We did not find any significant correlations between the basal levels of TNF-α and IL-6 and the autistic features of the total sample, similarly to some investigations [56,66] and in contrast to others [17,28,67,68]. Moreover, we found a positive, though weak, correlation between CCL2 and better functioning of children, evaluated with the CBCL1.5-5, RBS-R and VABS-II, in contrast with studies reporting a significant correlation between higher CCL2 plasmatic levels and more severe impairment of the autistic condition [21,57,69]. Further studies are necessary to disentangle the controversial findings on the possible role of some cytokines as sensible markers of the impairment in ASD children.

Third, we found that the group with regression plus developmental delay prior to the onset of ASD (16.5% of the sample) was significantly different from the rest of the sample as far as the higher plasmatic levels of resistin and PAI-1. We could suggest that Reg + DD children represent a specific subgroup with a definite biological profile and a specific clinical feature. However, using the PCA method, we did not identify the Reg + DD group as a particular cluster of patients, making the individuation of a specific endophenotype unlikely in this sample. Future studies are needed to retest the robustness of these findings before we can consider them as reliable.

In addition, we did not identify any significant correlation between the levels of cytokines and the presence or absence of a regression of skills prior to the onset of autism. This result is in accordance with the majority of similar investigations, but in contrast with others where an association, although weak, between regressive autism and TNF-α [70], or lower plasma leptin levels [34] was found. Previous studies detected higher basal plasmatic levels of IL-1β [17,69], IL-5 [69], IL-17 [69] and higher levels of neural cell adhesion molecule (NCAM) [55]—a molecule playing a role in cell–cell adhesion, neurite outgrowth, synaptic plasticity, learning and memory—in subjects with a regression of skills prior to the onset of autism. More broadly, ASD subjects with regression have been repeatedly identified as different in pathophysiological findings from ASD subjects without regression both in terms of neuroanatomy [71], and EEG patterns [72]. However, there is an urgent need to study the clinical regression in ASD, since a clear understanding of the definition, prevalence, etiopathogenesis, age of onset, and outcome profiles of this complex phenomenon is far from being concluded [73,74].

### *Limitations*

We must consider this study as a pilot investigation with several limitations. Compared to other authors who have measured a series of pro-inflammatory cytokines in ASD subjects [22], we focused our analysis on six cytokines, so limiting the possible range of our results. The changes in the expression of cytokines due to subjects' age [75] have already been described, and we cannot exclude that our results on inflammatory markers could be age-specific; in addition, we have to consider that sex, sleep-wake cycle and the percentage of fat mass, which could increase that variability [76,77] representing possible interfering factors, have not been assessed in this study. Moreover, the low number of females within our sample of preschoolers with ASD did not allow us to accurately investigate possible sex differences in pro-inflammatory cytokine profiles.

### **5. Conclusions**

Despite the above-mentioned limitations and the existing controversies within the studies about the role of cytokines in ASD and the extreme variability of their findings, our study finds no evidence of the presence of inflammatory condition in ASD subjects, except for resistin. Our findings do not support the use of anti-inflammatory therapies in ASD children, and paves the way for the search of alternative hypotheses for the etiology of GI symptoms in subjects with ASD. Despite our findings showed a specific plasmatic cytokine profile in ASD children with a history of a regressive way of onset within a previous developmental delay, the specific endophenotype for these subjects has not been identified.

### *Ethics Approval and Consent to Participate*

The study protocol was approved by the Pediatric Ethics Committee of Tuscany Region (Protocol Number: 126/2014), with written informed consent obtained from a parent/guardian of each participant. The study was conducted following the 1964 Declaration of Helsinki and its later amendments, and the International Conference on Harmonisation Guidelines for Good Clinical Practice.

**Availability of Data and Material:** The datasets generated and/or analyzed during the current study are not publicly available due to the privacy policy (containing information that could compromise research participant privacy/consent) but are available from the corresponding author on reasonable request and with permission of parents of the involved children.

**Author Contributions:** Conceptualization and Methodology, F.M., L.G., M.A.M. and E.S.; Laboratory Analysis, L.G. and M.G.; Investigation, M.P., L.G. and E.S.; Formal Analysis, L.G. and C.N.; Resources, F.M., L.G., M.A.M. and E.S.; Data Curation, M.P. and C.N.; Writing – Original Draft Preparation, M.P.; Writing – Review & Editing, L.G., D.G.P., C.N., S.C., R.T., M.A.M., A.G., F.M. and E.S.; Supervision, D.G.P., M.A.M., A.G. and F.M; Funding Acquisition, F.M. All authors read and approved the final manuscript.

**Funding:** This trial is funded by the Italian Ministry of Health and by Tuscany Region with the grant 'GR-2011-02348280 . This work was also partially supported by grants from the IRCCS Stella Maris Foundation (Ricerca Corrente, and the "5 × 1000" voluntary contributions, Italian Ministry of Health to F.M., R.T., S.C. and E.S.). We are also grateful to Università di Pisa for supporting Dr. Prosperi with a research Grant (D.R. n. 33134 29/05/2018).

**Acknowledgments:** We gratefully acknowledge the families participating in our research.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Paternal—but Not Maternal—Autistic Traits Predict Frontal EEG Alpha Asymmetry in Infants with Later Symptoms of Autism**

**Valentina Riva 1,\*, Cecilia Marino 2, Caterina Piazza 3, Elena M Riboldi 1, Giulia Mornati 1, Massimo Molteni <sup>1</sup> and Chiara Cantiani <sup>1</sup>**


Received: 21 October 2019; Accepted: 24 November 2019; Published: 26 November 2019

**Abstract:** Previous research found that the parental autism phenotype is associated with child autism spectrum disorder (ASD), even if the pathway between autistic traits in parents and child ASD is still largely unknown. Several studies investigated frontal asymmetry in alpha oscillation (FAA) as an early marker for ASD. However, no study has examined the mediational effect of FAA between parental autistic traits and child ASD symptoms in the general population. We carried out a prospective study of 103 typically developing infants and measured FAA as a mediator between both maternal and paternal autistic traits and child ASD traits. We recorded infant baseline electroencephalogram (EEG) at 6 months of age. Child ASD symptoms were measured at age 24 months by the Child Behavior Checklist 11/2–5 Pervasive Developmental Problems Scale, and parental autistic traits were scored by the Autism spectrum Quotient questionnaire. The mediation model showed that paternal vs. maternal autistic traits are associated with greater left FAA which, in turn, is associated with more child ASD traits with a significant indirect effect only in female infants vs. male infants. Our findings show a potential cascade of effects whereby paternal autistic traits drive EEG markers contributing to ASD risk.

**Keywords:** autism spectrum disorder; infants; frontal EEG alpha asymmetry; early detection

### **1. Introduction**

Autism spectrum disorder (ASD) is a complex and heterogeneous condition characterized by social communication deficits and repetitive patterns of behavior [1]. Twin studies show that ASD is a heritable condition, with heritability estimates ranging between 64% and 91% [2]. Genetic susceptibility appears to be expressed in relatives of individuals with ASD through an independent segregation of a broader range of subclinical features (autistic traits) in social communication and atypical patterns that are referred to as representing the broader autism phenotype (BAP). Several studies demonstrated that autistic traits are distributed normally in the general population and are heritable [3]. In particular, parental autistic traits have been found associated with child ASD symptoms in both ASD samples and the general population, suggesting that broader autistic traits are important to identify both clinical and subclinical conditions [3]. However, even if BAP and ASD seem to exist on a continuum, it is still unknown whether and how maternal and paternal autistic traits are differentially associated with

child social communication development. Several studies examined sex-specific associations between parental autistic traits and child ASD symptoms [4–6]. For example, Schwichtenberg et al. [5] found that paternal autistic traits predicted child ASD severity, while this relationship was not found for mothers. Klusek et al. [7] showed that paternal autistic traits (rigid and untactful traits) were associated with more social deficits in their ASD children. Conversely, a significant association emerged between child performance on a facial identity recognition task and maternal autistic traits, with no relationship between fathers and their children's scores [6]. Overall, the idea that paternal characteristics are more strongly associated with child ASD phenotype than maternal characteristics is more consistent with the literature, but it has not been well replicated.

Parental autistic traits are also relevant for the evaluation of endophenotypes [8]. A study using an eye-tracker system in 8-month-old infants found that paternal autistic traits were associated with infants' attentional functioning, suggesting that early impairments in low-level attentional systems may affect high-level social impairment [9].

Although it has been clearly established that child ASD traits can be influenced by parental autistic traits, the underlying mechanisms and the pathway between parental autistic traits and ASD symptoms in children are still largely unknown. Previous electroencephalogram (EEG) studies showed that anomalous oscillatory organization in multiple frequency bands was strongly associated with ASD, with many of these studies emphasizing the crucial role of individual changes in frontal EEG alpha power [10]. Alpha-band oscillations are associated with precise timing of sensory and cognitive inhibition [11]. Interestingly, significant differences in baseline alpha power were identified in infants at high risk for ASD (siblings of children with ASD) compared to typically developing infants, whereby high-risk infants at age 6 months showed lower alpha power as compared to controls [11].

In addition to spectral power differences, changes have been reported in hemispheric asymmetry of the frontal alpha band. Frontal alpha asymmetry (FAA) refers to the difference in EEG power between the frontal right hemisphere and the frontal left hemisphere [12]. Differences in alpha activity between the left and right hemispheres have been used to measure neural activity as a metrics for frontal lobe organization. In particular, because alpha power is inversely related to cortical activation (meaning that decreased alpha power reflects greater brain activity), right FAA indicates higher cortical activation in the right hemisphere and left FAA indicate higher cortical activation in the left hemisphere. In other words, positive values are associated to left FAA (left hemisphere activation) and negative values are associated to right FAA (right hemisphere activation).

FAA has been found associated with ASD [13,14]. Sutton and colleagues [14] examined the relationship between FAA, social impairment, and social anxiety in a sample of high-functioning ASD children compared to controls. The groups significantly differed on FAA, social impairment, and anxiety symptoms. ASD children with right FAA displayed more social deficits and ASD symptoms, whereas greater left FAA was associated with less social deficits and more anxiety symptoms.

Interestingly, research with typically developing infants has provided evidence that FAA changes during early years of life. Typically developing infants gradually shift from right FAA at age 6 months to left FAA at age 18 months [15–18]. Interestingly, recent evidence indicates that 6-month-old infants at high risk for ASD show an opposite developmental trajectory in FAA, shifting from left FAA at age 6 months to right FAA by age 18 months [17]. Overall, these data demonstrated a different hemispheric organization in infants at high risk for ASD, whereby frontal asymmetry may represent one of the earliest potential endophenotypes for ASD. Despite the fact that FAA has been found associated with ASD [16,17] and parental autistic traits are associated with child ASD [8], no study has explored the role of autistic traits in mothers and fathers on FAA and later child ASD symptoms concurrently.

In addition, recent growing evidence suggests that the pattern of frontal EEG asymmetry might be associated with psychological processes in female infants to a greater extent than male infants. Indeed, there is strong evidence that the link between FAA and psychosocial difficulties may be moderated by sex, with stronger associations in female infants than male infants [19,20]. Even if there are no studies

in the ASD field, it may be interesting to explore the relationship between FAA and ASD symptoms in female infants and male infants separately.

Based on this, we conducted a prospective longitudinal study on 103 typically developing infants (general population) and investigated FAA at age six months as a possible mediator of the impact of parental autistic traits on child ASD symptoms. The study aimed to determine 1) whether parental autistic traits are associated with FAA at age six months and child ASD symptoms at age 24 months 2) whether six-month FAA may reflect a potential neural mechanism predicting child ASD symptoms at age 24 months, 3) whether 6-month FAA is a mediator of the contribution of parental autistic traits to ASD-related symptoms, and 4) whether child sex moderates the associations between parental autistic traits, FAA, and ASD symptoms. Our assumption was that FAA significantly predicts child ASD traits and that FAA would serve as a potential neural mediator between parental autistic traits and child ASD-outcome. Based on previous research indicating the influence of sex on the FAA link to developmental psychopathology, we hypothesized that the tested mediation model is moderated by child sex, with a stronger effect in female infants than male infants.

### **2. Materials and Methods**

### *2.1. Sample*

At 6 months of age, 103 typically developing infants took part in the study (female-to-male ratio = 0.51). The sample was recruited from two hospitals in Northern Italy [21,22]. Inclusion criteria were (a) gestational age ≥36, (b) birth weight ≥2500 g, (c) Bayley Cognitive Score at 6 months ≥7 [23], and (d) no certified intellectual disabilities among first-degree relatives. Families with a diagnosis of ASD among first-degree relatives were also excluded, since we decided to focus on a broad autism phenotype in the general population. Descriptive statistics of demographics and clinical characteristics are shown in Table 1. Parents of all children had given their informed consent for inclusion before participation in the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of IRCCS Medea (Ricerca Corrente "2016, 2017, 2018, 2019", Ricerca Finalizzata "NET-2013-02355263-2" and "5 per mille" funds for biomedical research).


**Table 1.** Descriptive statistics of demographics and clinical characteristics.

<sup>a</sup> The educational level of mothers and fathers was scored on a 9-point ordinal scale created ad-hoc and based on the Italian school system. <sup>b</sup> Socioeconomic status was scored according to Hollingshead 9-point scale, whereby a score ranging from 10 to 90 was assigned to each parental job and the higher of two scores was considered when both parents were employed [24]. <sup>c</sup> Age-standardized (mean = 10; SD = 3) score on the Bayley cognitive subscale [23]. <sup>d</sup> Age-standardized *z*-scores (mean = 0; SD = 1) for the total Autism Spectrum Quotient (AQ) score [25]. <sup>e</sup> Age-standardized T-scores (mean = 50; SD = 10) for the Child Behavior Check List 11/2–5 (CBCL 11/2–5) [26].

### *2.2. Frontal EEG Alpha Asymmetry*

### 2.2.1. EEG Data Acquisition

Four minutes of baseline EEG at age 6 months (M = 6.46 months; SD = 0.49) were used to compute alpha asymmetry scores. Baseline EEGs were recorded after an experimental session (i.e., a passive oddball paradigm intended to test auditory processing skills; see [27]. During EEG recording, infants were looking at an experimenter blowing soap bubbles.

### 2.2.2. EEG Data Processing and Analysis

EEGs were recorded from 60 scalp electrodes using HydroCel Geodesic sensor nets (Electrical Geodesics, Inc., Eugene, Oregon, USA). The vertex electrode was the online reference. EEGs were recorded with a sampling rate of 250 Hz and an online band-pass filter (0.1–100 Hz). After recording, EEGs were exported and further processed using lab-internal MATLAB (Mathworks, Natick, MA, USA) routines and the EEGLAB toolbox [28]. Data were band-pass filtered at 1–47 Hz. Bad channels were identified by means of the EEGLAB "TrimOutlier" plugin (identification criteria: 5 < all channels SD < 100) and interpolated with a spherical spline (a maximum of 12 out of 60 channels were interpolated, M = 3.3, SD = 2.7). Data were then re-referenced offline to an average reference and segmented in one-second non-overlapping epochs. Bad EEG epochs were identified and rejected using two EEGLAB functions: (1) "find abnormal values", marking for rejection epochs in which EEG values exceeded ±150 μV) and (2) "find abnormal trends", marking for rejection epochs corrupted by linear drift (setting parameters: R = 0.3, max slope = 150 μV). Additional manual artifact inspection was computed. A minimum of 60 artifact-free segments (M = 119.8; SD = 39.9) was used for subsequent power analysis. Power spectral density (PSD) was estimated by Welch's method [26,29] with non-overlapping 0.5 s windows. PSD values were calculated for each channel in each epoch and then averaged across segments. Following previous literature [17,30], the mean power in the infant alpha frequency band (6–9 Hz) was computed. We selected two clusters of electrodes (based on and adapted from [17], frontal left hemisphere: 9, 11, 12, 13, 14 and frontal right hemisphere: 2, 3, 57, 59, 60 (see Supplementary Materials) and power values were averaged across electrodes within each cluster. In full-spectrum data, we focused on frontal alpha asymmetry (FAA) that has been well characterized in infants. Frontal asymmetry scores were calculated from log-transformed PSD values in selected clusters as follows: (right − left)/(right + left). This formula has been used in most studies to summarize the relative activity at homologous right and left sites [31]. Use of this formula to calculate FAA offers the advantage of minimizing bias due to individual differences in skull thickness that might influence the power spectrum amplitude. In addition, it approaches a normal distribution and shows good stability and reliability [32]. Positive values indicate left FAA and negative values indicate right FAA (M = 0.03; SD = 0.13).

### 2.2.3. Autistic Traits in Parents: The Autism Spectrum Quotient

The Autism Spectrum Quotient (AQ) is a self-administered questionnaire to quantify autistic traits in the general population [25], with the Italian version by [33]. The AQ questionnaire offers several advantages, including subscales tapping both social and non-social aspects of behavior and cognition and a brief, self-administered, and forced-choice format [34]. Subjects are instructed to respond to each of the 50 items using a 4-point Likert scale as follows: "definitely agree", "slightly agree", "slightly disagree", and "definitely disagree". The maximum score on the AQ is 50 points, with higher scores meaning higher presence of autistic traits. Two cut-offs were previously described/reported [25]: clinical threshold (raw scores ≥ 32) and screening cut-off (raw scores ≥ 26). Reflecting the non-clinical nature of our sample, only 3 parents (all fathers) reached the clinical threshold and only 13 parents (9 fathers and 4 mothers) reached the screening cut-off. The AQ questionnaire was completed by both parents upon their children's inclusion in the study. Total AQ scores were transformed into *z*-scores (see Table 1) and were used in further analysis.

### 2.2.4. Autistic Traits in Infants: The CBCL 11/2–5 Pervasive Developmental Problems Scale

The Child Behavior Check List 11/2–5 (CBCL 11/2–5) consists of 99 items designed to rate emotional and behavioral problems in toddlers. Items are scored by parents on a 3–point Likert scale (0 = not true; 1 = sometimes true; 2 = very true) and they refer to a time span of 6 months before the questionnaire completion. In our sample, 73 questionnaires were filled by mothers (80%), 5 by fathers (5%), and 14 by both parents (15%). This measure with strong psychometric properties across cultures has been translated into, and validated in Italian [35,36]. For the purpose of this study, the Pervasive Developmental Problems scale (PDP) was used as child ASD symptoms and PDP T-scores (mean = 50; SD = 10) were used in the analysis. Reflecting the non-clinical nature of our sample, only 7 children (4 female infants and 3 male infants) reached the clinical threshold (T ≥ 65).

### *2.3. Statistical Analysis*

Descriptive statistics and Pearson's bivariate correlations to examine the associations among study variables were run using SPSS, Version 25.0 (IBM Corp., Armonk, NY, USA). The association between FAA and ASD-related traits was assessed by linear regression analysis: the CBCL 11/2–5 PDP score was entered as the dependent variable and FAA was set as the predictor.

To investigate the contribution of paternal and maternal AQ to ASD-related traits in children and the potential role of FAA as a mediator, we used Structural Equation Modeling (SEM) [36], as implemented in the MPLUS software (Los Angeles, CA) [37]. SEM simultaneously models all paths, providing a more accurate estimation of mediation effects [38,39] than more traditional tests based on sequential regressions.

The mediation model tested the hypothesis that ASD-related traits in children would be explained by a sequence of potentially associated effects involving parental autistic traits and FAA. Specifically, the following model was proposed: maternal and paternal AQ → FAA → child ASD traits. We then assessed the mediation model which best described the associations between the measured variables [39]. Finally, moderation by child sex was examined to assess whether relations between study variables differed by sex (male vs. female).

The bias-corrected 5000 bootstrap technique was used to test mediation effects [36]. Confidence intervals (95% CI) that do not contain zero indicate significant indirect effects [40–43]. Several fit indices are used to assess the best fitting model: a) Chi-Square assessing the difference in magnitude between the model estimated variance/covariance matrices, b) the RMSEA (root mean square error of approximation) considering the complexity of the model [42], c) the SRMR (standardized root mean square residual) indicating the average residual value from the model fit covariance matrix to the sample covariance matrix, and d) the CFI (comparative fit index) indicating the improvement in overall model fit by comparing the hypothesized model with a more restricted one, which specifies no relations among variables. RMSEA ≤ 0.05, SRMR ≤ 0.08, and CFI ≥ 0.95 indicate a good model fit [42–45]. To allow for the use of all available data with inclusion of subjects with missing data, we considered the full information maximum likelihood estimation. Significant effects were set to *p*-values ≤ 0.05.

### **3. Results**

We first examined the correlations between maternal and paternal AQ (*z*-scores), CBCL 11/2–5 Pervasive Developmental Problems (T-scores), and FAA. We found a significant correlation between FAA and CBCL 11/2–5 PDP scores (*r* = 0.42, *p* < 0.001), with greater left FAA being associated with more child ASD-related symptoms. Correlations between paternal AQ scores and FAA (*r* = −0.29, *p* = 0.003) and paternal AQ scores and CBCL 11/2–5 PDP scores were low-to-moderate (*r* = −0.39, *p* < 0.001). Higher paternal autistic traits (negative z-scores for AQ values) were associated with greater left FAA at age 6 months and more child ASD-related symptoms at age 24 months. No significant correlation emerged between maternal AQ scores and FAA (*r* = −0.11, *p* = 0.244) and a low—although not significant—correlation was found between maternal and paternal AQ scores (*r* = 0.14, *p* = 0.157). Finally, no significant correlation was found between maternal AQ scores and CBCL 11/2–5 PDP scores (*r* = −0.09, *p* = 0.394).

### *3.1. Testing a Mediation Model: FAA as a Mediator between Parental AQ Scores and Child ASD-Related Traits*

After carrying out descriptive and correlational statistics, we used SEM to test the mediation model shown in Figure 1, which assumes that maternal and paternal AQ scores contribute to FAA. FAA, in turn, affects child ASD-related traits. The model provided a good fit to the data (χ<sup>2</sup> (5) = 37.31, *p* < 0.001; RMSEA = 0.000, CI (90%) = 0.000–0.000; CFI = 1.00; SRMR = 0.000) and accounted for 26.3% of the variance in CBCL 11/2-5 PDP scores. Figure 1 shows standardized coefficient estimates. The mediation model yielded several significant direct effects. There was a significant path coefficient from paternal AQ to CBCL 11/2–5 PDP scores (β = −0.30, *p* = 0.004). Children with higher paternal autistic traits showed higher PDP scores at age 24 months. No significant association from maternal AQ to CBCL 11/2–5 PDP scores emerged (β = −0.03, *p* = 0.719). Significant effects were found from paternal AQ scores to FAA (β = −0.28, *p* = 0.014), and from FAA to CBCL 11/2–5 PDP scores (β = 0.33, *p* = 0.028): Higher paternal autistic traits predict greater left FAA and greater left FAA predicts higher infant ASD-related traits. However, 5000 bootstrap estimates (95% CI) showed that the indirect effect from paternal AQ to CBCL 11/2–5 PDP scores via FAA was not significant (β = −0.092; SE = 0.256; 95% CI (−0.193; 0.010), *p* = 0.112).

**Figure 1.** Frontal asymmetry in alpha oscillation (FAA) as a mediator between maternal and paternal autism spectrum quotient (AQ) scores and child autism spectrum disorder (ASD)-related traits. PDP—Pervasive Developmental Problems.

### *3.2. Testing a Moderated Mediation Model: FAA as a Mediator between Parental AQ and Child ASD-Related Traits Moderated by Sex*

Moderated mediation was also applied to examine whether child sex moderated the associations between maternal and parental AQ scores, FAA, and child ASD-related outcome via multigroup analyses in the SEM framework (see Figure 2 for group-specific parameter estimates).The model provided a good fit to the data (χ<sup>2</sup> (10) = 44.74, *p* < 0.001; RMSEA = 0.000, CI (90%) = 0.000–0.000; CFI = 1.00; SRMR = 0.000) and accounted for 15.5% of the variance in the CBCL 11/2–5 PDP scores in male infants and 40.5% of the variance in the CBCL 11/2–5 PDP scores in female infants.

**Figure 2.** FAA as a mediator between maternal and paternal AQ and child ASD-related traits moderated by sex.

Standardized estimates of path coefficients in each group are depicted in Figure 2. In male infants, there was only a direct effect of paternal AQ to CBCL 11/2–5 PDP scores (β = -0.38, *p* = 0.027), with higher paternal autistic traits predicting higher CBCL 11/2–5 PDP scores. No other direct or indirect effect was found.

In female infants, significant associations were found from paternal AQ scores and FAA (β = −0.40, *p* = 0.010), from paternal AQ scores and CBCL 11/2–5 PDP (β = −0.27, *p* = 0.038) and from FAA to CBCL 11/2–5 PDP scores (β = 0.44, *p* = 0.014): Higher paternal autistic traits predicted greater left FAA which, in turn, predicted higher CBCL 11/2–5 PDP scores. A significant correlation between maternal and paternal AQ scores (*r* = 0.29; *p* = 0.041) was also found.

Interestingly, 5000 bootstrap estimates (CI 95%) showed that the indirect effect from paternal AQ to CBCL 11/2–5 PDP scores via FAA was significant for female infants (β = −0.176; SE = 0.579; 95% CI (−0.334; −0.016), *p* = 0.041) but not for male infants (β = −0.008; SE = 0.117; 95% CI (−0.251; 0.246), *p* = 0.815). In female infants, paternal AQ scores were associated with greater left FAA at age 6 months, which affected child ASD-related traits at age 24 months. In other words, FAA mediates the association between paternal AQ scores and child ASD-related symptoms, and this link was moderated by child sex in female infants but not in male infants (for graphical purposes only, see Figure 3).

**Figure 3.** FAA differences between paternal AQ scores, CBCL 11/2−5 PDP scores and child sex. Legend: Paternal AQ scores (−): low paternal autistic traits; Paternal AQ scores (+): high paternal autistic traits; PDP scores (−): low CBCL 11/2–5 PDP scores; PDP scores (+): high CBCL 11/2−5 PDP scores.

### **4. Discussion**

This is the first general population study looking at frontal asymmetry in EEG alpha oscillation at age 6 months as a potential mediator in the developmental pathway from maternal and paternal autistic traits to child ASD-related traits at age 24 months.

### *4.1. Parental Autistic Traits and Child ASD Symptoms*

Not surprisingly, paternal autistic traits were associated with more child ASD-related symptoms, supporting the assumption that broader autistic traits in parents may be useful in identifying both clinical and subclinical conditions [3]. This finding is well replicated in previous studies. The association between higher autistic traits in parents and their children strongly supports an underlying genetic mechanism [46]: Shared genetic variability may be a plausible pathway for familial transmission of common factors between parental autistic traits and their child ASD-related traits. Consistent with the literature, we found that paternal characteristics are associated with child ASD phenotype rather than maternal characteristics [4–6], supporting greater patrilineal effects within families [7].

### *4.2. Parental Autistic Traits and Frontal Asymmetry in Alpha Oscillation*

Our findings also showed that higher paternal autistic traits were associated with greater left FAA at age 6 months. Typically developing infants shifted from greater right FAA (right frontal activation) at age 6 months to relative greater left FAA (left frontal activation) at age 18 months [15,16]. Interestingly, an opposite pattern was found in infants at high risk for ASD (siblings of children with ASD) from greater left FAA at age 6 months to greater right FAA at age 18 months [17,18,47]. In line with this piece

of evidence, we found that at-risk infants (having a parent with higher autistic traits) aged 6 months showed greater left FAA, suggesting an atypical organization and lateralization of oscillatory processes. We might speculate that hemispheric organization follows a different developmental shift in infants with higher paternal autistic traits. However, since EEG measures at age 18 months were not obtained, further studies are needed to confirm this hypothesis.

### *4.3. Frontal Asymmetry in Alpha Oscillation and Child ASD Symptoms*

Greater left FAA at age 6 months was associated with increased child ASD-related traits at age 24 months. Past research has linked FAA to different cognitive and behavioral processes, supporting the role of FAA as a biomarker for psychopathology [48]. FAA is a measure of the propensity to adopt an "approach or withdrawal" behavior [48], with greater left frontal activity associated with an increased tendency to approach, and greater right frontal activity associated with an increased tendency to withdraw. Taken together, frontal alpha asymmetry can be interpreted with respect to the amount of motivation towards (approach) or away from (withdraw) something or someone. Relating to ASD, greater right frontal asymmetry is associated with social impairment and earlier onset of ASD symptoms [13,14], whereas less social impairment has been observed in children with greater left frontal asymmetry.

However, the direction of the effect that we found in the present study (i.e., greater left FAA at age 6 months associated with earlier ASD-related symptoms) was different from what was expected based on previous literature. This discrepancy could be due to different population characteristics (infants vs. children/adolescents; typically developing infants vs. high-functioning ASD children). Since frontal asymmetry tends to change over the first two years of life, with a shift in lateralization from right to left FAA in typically developing infants and from left to right FAA in infants at risk for ASD [15], the different direction of the reported effects between the present and previous studies might be due to maturation effects. This needs to be confirmed in more overlapping study populations.

### *4.4. Frontal Asymmetry in Alpha Oscillation as a Mediator between Paternal Autistic Traits and Child ASD Symptoms*

Perhaps more importantly, we found that paternal, but not maternal, autistic traits are directly associated with FAA and FAA directly affects child ASD-related traits. This different association may reflect different biological mechanisms based on parent-of-origin effects, namely the genetic effects on the (endo)phenotype of an offspring that are dependent on the parental origin of the associated genetic variants. Several studies [49,50] found that parents may transmit genes or epigenetic dysregulation affecting ASD through sex-specific pathways and, in line with our own results, parent-of-origin effects were found in ASD, with a paternal over-transmission of risk alleles for ASD [49]. Even if further research is needed, our results may support the importance to explore the role of epigenetic modulators in the etiology of ASD. If parent-of-origin effects are proven, more homogeneous ASD-related phenotypes could be identified by grouping according to parental ASD traits.

Although in an exploratory manner, we found a significant indirect effect and provided the first evidence that FAA at age 6 months significantly mediates the contribution of paternal autistic traits to ASD-related traits in their children, while this is seen at age 24 months in female infants but not in male infants. It is well replicated that male infants are more frequently diagnosed with ASD than female infants, with a reported sex ratio of 4:1 [51]. Sex differences may reflect the distinctive sexual dimorphism of the brain, including hormonal and structural factors as well as genetic and epigenetic influences, which emerge during development. For example, effects of serotonin genotypes on EEG activity were found to vary as a function of sex. The 5-HTTLPR polymorphism was associated with modulation of the EEG activity at different EEG frequencies only in female infants and not in male infants [52], suggesting that baseline EEG frontal activity marks different neurobiological processes in female infants and male infants. Understanding the mechanisms underlying the sex difference in ASD is not only fundamental per se, but it might crucially contribute to unravelling the well-known sex differences in prevalence, age of onset, and severity that we observe in many psychiatric diseases, including depression and anxiety disorder and ADHD, in which a role of FAA has been reported [20,53]. Even if replication studies are necessary, it is conceivable that FAA involved in cortical development—if combined with higher parental autistic traits—could potentiate different genetic vulnerabilities in male infants and female infants, specifically ASD-related problems.

This study presents some limitations. First, ASD traits were assessed solely by parental report. Although no evidence for report bias regarding parent–offspring autistic traits emerged in previous studies [54], in our study we cannot exclude that parental ASD traits might have an effect on parental perception of their child's behavior. Therefore, we suggest that future studies should focus on direct assessment of autism-related symptoms. Second, we measured EEG frontal alpha asymmetry only at 6 months of age. As reported by previous studies [15–17], FAA seems to be developmentally sensitive from age 6 to 24 months. Future longitudinal studies with larger samples of typically developing and at-risk infants are important to increase our confidence on (a) typical EEG asymmetry trajectories and how such EEG trajectories relate to different broader autism phenotype domains. A further point to be highlighted concerns the specificity of the relationship between FAA and ASD traits. Previous studies reported that oscillations in different frequency bands are related to the development of other cognitive skills (i.e., oscillations in the gamma frequency bands have been reported to be predictive of language skills) [55]. In our study, we tried to disentangle a possible connection between frontal alpha asymmetry and language by means of exploratory correlations with language measures at 24 months and found—as expected—no significant correlations, thus supporting the assumption of a specific pathway between FAA and ASD traits. Further studies are needed in this direction.

### **5. Conclusions**

These findings support the use of objective measurements of EEG frontal alpha asymmetry to delineate specific pathophysiological mechanisms in ASD. Notably, this study reports a prediction of ASD symptoms at age 24 months. However, our longitudinal data collection is ongoing, and we are prospectively following our current cohort to identify children who will or will not receive a diagnosis of ASD. Characterization of reliable biomarkers will guide the detection of the most vulnerable infants that will benefit from early intervention and rehabilitation, with the long-term aim of substantially reducing the heavy impact of ASD on the National Health System.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2076-3425/9/12/342/s1, Figure S1: Sensor layout of the 60-channel Hydro-Cel Geodesic Sensor Net used in the study. Red and blue squares represent the electrodes included, respectively, in the Left and Right frontal clusters and entered into statistical analyses (see Gabard-Durnam et al., 2015).

**Author Contributions:** V.R. planned the study, performed data analyses and prepared the manuscript; C.M. performed data analyses, supervised the manuscript; C.P. supervised the experiment, performed data analysis; E.M.R. and G.M. were involved in participant recruitment, data collection and processing; M.M. supervised the manuscript; C.C. supervised the experiment, performed data analyses and prepared the manuscript.

**Funding:** This study was funded by the Italian Ministry of Health (Ricerca Corrente "2016, 2017, 2018, 2019" to Chiara Cantiani, Ricerca Finalizzata "NET-2013-02355263-2") and by "5 per mille" funds for biomedical research to Valentina Riva.

**Acknowledgments:** The authors wish to thank all participating infants and their parents.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Cannabinoids for People with ASD: A Systematic Review of Published and Ongoing Studies**

### **Laura Fusar-Poli 1, Vito Cavone 1, Silvia Tinacci 1, Ilaria Concas 1, Antonino Petralia 1, Maria Salvina Signorelli 1, Covadonga M. Díaz-Caneja <sup>2</sup> and Eugenio Aguglia 1,\***


Received: 15 July 2020; Accepted: 17 August 2020; Published: 20 August 2020

**Abstract:** The etiopathogenesis of autism spectrum disorder (ASD) remains largely unclear. Among other biological hypotheses, researchers have evidenced an imbalance in the endocannabinoid (eCB) system, which regulates some functions typically impaired in ASD, such as emotional responses and social interaction. Additionally, cannabidiol (CBD), the non-intoxicating component of *Cannabis sativa*, was recently approved for treatment-resistant epilepsy. Epilepsy represents a common medical condition in people with ASD. Additionally, the two conditions share some neuropathological mechanisms, particularly GABAergic dysfunctions. Hence, it was hypothesized that cannabinoids could be useful in improving ASD symptoms. Our systematic review was conducted according to the PRISMA guidelines and aimed to summarize the literature regarding the use of cannabinoids in ASD. After searching in Web of KnowledgeTM, PsycINFO, and Embase, we included ten studies (eight papers and two abstracts). Four ongoing trials were retrieved in ClinicalTrials.gov. The findings were promising, as cannabinoids appeared to improve some ASD-associated symptoms, such as problem behaviors, sleep problems, and hyperactivity, with limited cardiac and metabolic side effects. Conversely, the knowledge of their effects on ASD core symptoms is scarce. Interestingly, cannabinoids generally allowed to reduce the number of prescribed medications and decreased the frequency of seizures in patients with comorbid epilepsy. Mechanisms of action could be linked to the excitatory/inhibitory imbalance found in people with ASD. However, further trials with better characterization and homogenization of samples, and well-defined outcomes should be implemented.

**Keywords:** autism spectrum disorder; cannabinoids; cannabidiol; cannabidivarin; THC; problem behaviors; sleep; epilepsy; hyperactivity; side effects

### **1. Introduction**

Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by deficits in communication and social interaction and by a pattern of restricted interests and repetitive behaviors that might vary in severity [1]. It was estimated that around 1.5% of the general population might belong to the autism spectrum [2]. Along with core symptoms, ASD might present with several associated problems, such as irritability, challenging behaviors [3], and self-injury [4], especially in the presence of associated intellectual disability (ID), a condition that seemed to regard at least one-third of the autistic population [5]. Conversely, individuals with higher cognitive abilities are more frequently burdened by psychiatric comorbidities, such as depression, anxiety, attention deficit-hyperactivity disorder (ADHD), or sleep problems [6–8]. Medical comorbidities are also highly prevalent among the ASD population [9–11]. In particular, epilepsy represents the most frequent co-occurring neurological condition, affecting 5 to 30% of individuals with ASD [12–15]. Even in absence of frank seizures, people with ASD seem to present subclinical electrical discharges with abnormalities in EEG patterns [16,17].

The etiopathogenesis of ASD still needs to be clarified. Several genetic [18], perinatal [19,20], and environmental factors [21,22] seem to be involved. Research has also evidenced an imbalance in some endogenous neurotransmission systems [23], such as the serotoninergic [24], γ-aminobutyric acid (GABA)-ergic [17,25], and endocannabinoid (eCB) system [26–28].

Imbalances in the eCB neurotransmission system were found in animal models of ASD [29]. Additionally, lower serum levels of eCB were detected in children with ASD compared to typically developing peers [30,31]. Notably, the eCB system is relevant, as it seems to regulate some of the functions typically impaired in ASD, such as the form of emotional responses and social interaction [32].

Given the alterations in the eCB systems, researchers started to hypothesize that phytocannabinoids, which are naturally present in the plant of *Cannabis sativa*, might exert beneficial effects on the core and associated symptoms of ASD. First, multiple experimental studies conducted on mouse models showed that cannabidiol (CBD), the non-intoxicating component of cannabis, affects social interaction [33,34], which is severely impaired in ASD. Although CBD does not exert psych mimetic properties or the ability to induce addiction, it indirectly affects the transmission of the cannabinoid-related signal, the degradation of the endocannabinoid anandamide, and thus act on autistic-like symptoms in rats [35].

Interestingly, in June 2019, the US Food and Drug Administration (FDA) approved the Epidyolex, a CBD-based oral solution, for the treatment of seizures in Dravet and Lennox-Gastaut syndrome, two rare forms of epilepsy, in children older than two years of age [36]. As mentioned above, epilepsy is one of the most frequent co-occurring conditions of ASD, and the presence of seizures or non-epileptic abnormalities in EEG patterns might be partially responsible for the challenging behaviors or aggression in people with ASD. Thus, the correction of these abnormalities could improve, at least in part, the behavioral problems [37]. Moreover, the common co-existence of ASD and epilepsy suggests the presence of shared neuropathological mechanisms. Of note, both conditions are associated with abnormalities in the inhibitory GABA neurotransmission, including reduced GABAA and GABAB subunit expression. These abnormalities can elevate the excitatory/inhibitory balance, resulting in a hyper-excitability of the cortex, with an increased risk of seizures [38]. The literature showed that CBD seems to act similarly to antiepileptic drugs, as it increases the GABA transmission, thus reducing neuronal excitability [39,40].

Additionally, CBD exerts an agonist activity on the 5-HT1a receptors (i.e., serotoninergic system), which could mediate its pharmacological antidepressant, anxiolytic, and pro-cognitive properties [41,42]. In fact, the therapeutic effects of CBD were tested in patients suffering from anxiety disorder [43], a psychiatric comorbidity affecting at least 20% of people with ASD [8]. Possible benefits of CBD, due to its potential effects on the dopaminergic system, were also studied on subjects suffering from psychosis, [44], which could also represent a mental health issue for autistic individuals [8].

The effects of other cannabinoids were scarcely explored in clinical research. Cannabidivarin (CBDV) improved neurological and social deficits in early symptomatic Mecp2 mutant mice, a model of the Rett syndrome [45]. Moreover, it was proven to be an effective anticonvulsant in several models of epilepsy [46]. Delta-9-tetrahydrocannabinol (THC), the psychoactive component of cannabis, might increase sleep duration [47], thus being a potential candidate for a sedative effect. Additionally, it seems to reduce locomotor activity, which is indicative of a decrease in anxiety-like behavior [48]. According to a recent pilot randomized trial [49], a cannabinoid compound containing a 1:1 ratio of THC:CBD, significantly improved symptoms of hyperactivity, impulsivity, and inhibition measures in adults with ADHD, a condition that seemed to affect around 28% of autistic subjects [8].

As mentioned above, ASD presents serious deficits in social interaction and communication, as well as repetitive behaviors. However, till date, no effective pharmacological treatment exists for ASD core symptoms; only two atypical antipsychotics (i.e., risperidone and aripiprazole) were approved by the FDA for the treatment of irritability in children and adolescents with ASD [50]. Nevertheless, psychotropic medications are frequently prescribed in everyday clinical practice, with the frequent onset of side effects [51]. Given their properties, cannabinoids were proposed as candidate therapeutic options in people with ASD. Two recent narrative reviews were conducted on the topic [52,53]. However, to the best of our knowledge, no systematic reviews have comprehensively summarized the effects of cannabinoids for the treatment of individuals with ASD. The present paper aimed to describe the current state-of-the-art regarding the use of cannabinoids in individuals with ASD, focusing on both published and ongoing trials.

### **2. Materials and Methods**

### *2.1. Search Strategy*

We followed the PRISMA Statement guidelines to perform a systematic search [54]. First, we searched the following databases from inception up to 26 May 2020: Web of KnowledgeTM (including Web of Science, MEDLINE®, KCI—Korean Journal Database, Russian Science Citation Index, and SciELO Citation Index), PsycINFO, Embase, and ClinicalTrials.gov, without any time or language restriction. We used the following search strategy: *(cannab \*) AND (autis \* OR asperger OR kanner OR "neurodevelop \* disorder \*").* Second, we reviewed all references of relevant reviews and meta-analyses to find any additional eligible study.

### *2.2. Eligibility Criteria*

Two review authors (LF and VC) screened all retrieved papers, independently and in duplicate. Any doubt was solved by consensus. The authors included all original studies written in English, published as full papers or abstracts in peer-reviewed journals, and met the following criteria:

(1) Participants: Individuals with a diagnosis of autism spectrum disorder (ASD), according to international valid criteria or measured by a validated scale, regardless of age.

(2) Intervention: *Cannabis sativa* or cannabinoids, such as, cannabidiol (CBD), cannabidivarin (CBDV), delta-9-tetrahydrocannabinol (THC) and others, administered at any dosage and any form.

(3) Comparison: Studies with or without a comparison group (placebo or other forms of treatment).

(4) Outcomes: Any outcome.

(5) Study design: Case report, case series, retrospective, observational longitudinal, randomized or controlled clinical trials, both parallel and crossover.

### *2.3. Data Extraction*

Data were extracted by two authors (S.T. and I.C.) who worked independently and in duplicate. Any doubt was solved by consensus. A standardized form was used to extract data from the included studies. We extracted information about study characteristics (authors, year, study design, country), characteristics of the ASD sample (sample size, age, presence of ID, presence of epilepsy, concomitant medications), type and duration of the intervention and the comparison, outcomes and outcome measures, findings, and side effects. We also reported data regarding ongoing studies, as retrieved in ClinicalTrials.gov. Results of the study were reported in a narrative summary that was organized around the study characteristics.

### **3. Results**

### *3.1. Search Results*

Our search yielded a total of 758 studies, while four additional articles were found through other sources. After removing duplicates, we screened 604 titles and abstracts. After reading the full texts of 24 papers, we finally included 10 published works (eight full articles and two conference abstracts) in our systematic review. Additionally, nine ongoing trials were found in ClinicalTrial.gov, of which four met the eligibility criteria. The selection procedure of the included studies was reported in Figure 1.

**Figure 1.** PRISMA flow chart of the study selection process.

### *3.2. Characteristics of Studies and Participants*

We included three retrospective studies, three prospective studies, one case report, and three randomized placebo-controlled crossover trials. Apart from the case report [55], all papers were published within the last three years. Studies were conducted in Israel (*n* = 3), United Kingdom (*n* = 3), Brazil, Chile, Austria, and United States (*n* = 1 each). Sample sizes ranged from one [55] to 188 [56]. Participants were mainly children, although in two studies there were mixed samples [57,58]. The three studies conducted by Pretzsch and colleagues [59–61] included only adults with normal cognitive abilities (IQ > 70). Interestingly, only another study [62] specified the level of functioning, which was not reported in the remaining papers. Many participants were taking concomitant medications, and part of the samples had epilepsy. However, this information was not specified in two studies [55,57]. Study characteristics are reported in Table 1.




*Brain Sci.* **2020**, *10*, 572

### *3.3. Characteristics of Treatment*

The treatment was represented by a cannabinoid oil solution with a CBD:THC ratio of 20:1 in two studies [57,62] and with a 30:1.5 ratio in one study [56]. Fleury-Teixeira et al. [63] and Kuester et al. [58], instead used *Cannabis sativa* extracts with different compositions. Kurz and Blaas [55] reported the use of dronabinol (delta-9-THC) dissolved in sesame oil. McVige et al. [64] documented the use of medical cannabis with unspecified composition. Finally, Pretzsch and colleagues administered single doses of 600 mg of CBD or CBDV [59–61]. Only the studies by Pretzsch et al. used a control treatment (placebo). The duration of treatment was extremely variable, ranging from single administrations [59–61] to six months [55,56]. Of note, many studies were naturalistic and treatment duration was different among participants. Characteristics of treatment with cannabinoids are reported in Table 1.

### *3.4. Outcomes and Findings*

The results of the included studies are reported in Table 2. It could be observed that studies typically had multiple outcomes. The most investigated were global impression, sleep problems, hyperactivity, problem behaviors, use of concomitant medications, and seizures. Parenting stress was measured in two studies [58,62]. Anxiety, mood, and quality of life were evaluated in the context of global impression. Only one study [63] specifically evaluated socio-communication impairments, reporting a median perceived improvement of 25%. However, the authors did not use standardized tools to measure the changes in the communication and social interaction domain. Surprisingly, none of the included studies aimed to evaluate changes in stereotypies. Of note, the three studies conducted by Pretzsch et al. [59–61] investigated the acute effects of cannabinoids using neuroimaging techniques (magnetic resonance spectroscopy [MRS] and functional Magnetic Resonance Imaging [fMRI]). Outcomes and results are reported in Table 2.

### *3.5. Ongoing Trials*

We retrieved four ongoing studies from ClinicalTrials.gov, of which two were randomized controlled trials and two open label trials. Three of these studies are being conducted in the United States, and one in Israel. Researchers mainly planned to recruit children (except for the trial NCT02956226, which planned to extend the administration of treatment up to the age of 21 years). Two studies are testing the effects of CBDV, one study is examining the effects of CBD at different dosages, and one is looking at the effects of a combination of CBD and THC (ratio 20:1). Duration of trials are from 6 to 52 weeks. All trials planned to administer multiple outcome measures to both patients and caregivers. Interestingly, specific tools measuring changes in ASD core symptoms were inserted, including the evaluation of repetitive behaviors and stereotypies. Adaptive abilities, aberrant behaviors, and sleep disturbances are other target symptoms of the studies. The characteristics of the ongoing trials are summarized in Table 3.


*Brain Sci.* **2020**, *10*, 572



functional Magnetic Resonance Imaging; *GABA*+: Gamma aminobutyric acid; *Glx*: glutamate + glutamine; *HSQ*: Home situation Questionnaire; *MRS*: Magnetic Resonance Spectroscopy.


Characteristics of ongoing trials testing cannabinoids in people with ASD.

> **Table 3.**

Severity; *PedsQL*: Pediatric Quality of Life

Anxiety Related Disorders; *SDSC*: Sleep Disturbance Scale for Children; *SRS-2*: Social

Treatment-Emergent

 Adverse Events; *THC*:

Inventory—Family

 Impact Module; *RBS-R*: Repetitive Behavior

delta-9-tetrahydrocannabinol;

 *Vineland 3*: Vineland Adaptive Behavior Scale-3.

Scale-Revised; *RCT*: randomized controlled trial; *SCARED*: Screen for Child

Responsiveness

 Scale, 2nd Edition; *TEAEs*: Number of Participants Who Experienced Severe

### **4. Discussion**

In the present systematic review, we found preliminary evidence showing that cannabinoids (compounds with different ratios of CBD and THC), might exert beneficial effects on some ASD-associated symptoms, such as behavioral problems, hyperactivity, and sleep disorders, with a lower number of metabolic and neurological side effects than medications. Importantly, treatment with cannabinoids allowed to reduce the number of prescribed medication and significantly reduced the frequency of seizures in participants with comorbid epilepsy. We will now reflect in-depth on some critical points related to the main findings, mechanisms of action of cannabinoids, and methodology of the included studies.

### *4.1. E*ffi*cacy and Safety of Cannabinoids in ASD*

The majority of available interventions for ASD are based on behavioral, psychoeducational, and pharmacological therapies [65]. To date, the FDA has approved only two medications for the treatment of children and adolescents with ASD—risperidone and aripiprazole. Such medications are mostly used for irritability, aggressiveness, and self-injurious behaviors, but, unfortunately, there is no evidence of efficacy on core symptoms [66]. However, many drugs, such as antipsychotics, mood stabilizers, antidepressants, and stimulants, are prescribed off-label in clinical practice [51,67].

The findings of the studies included in the present systematic review are promising, as cannabinoids seem to improve some associated symptoms in many individuals with ASD, such as behavioral problems, hyperactivity, and sleep disorders. On the contrary, changes in core symptoms were scarcely explored—only one study [63] reported some improvements in communication and social interaction in a small sample of Brazilian children with ASD. No studies specifically investigated the effect of cannabinoids on repetitive behaviors or restricted interests. Of note, in individuals with comorbid epilepsy, the use of cannabinoids significantly reduced the frequency and intensity of seizures. Additionally, the number and dosage of used medications were reduced after the treatment with cannabinoids. This is a secondary, but extremely important finding. In fact, pharmacological therapies commonly prescribed to individuals with ASD are frequently burdened by side effects, such as weight gain, dyslipidemia, diabetes, and metabolic syndrome. These adverse events are also frequent in children, given the sensory difficulties, food selectivity, and rigidity in eating behaviors, which can lead to an increased risk for weight gain and poor nutritional habits [68–71]. For this reason, the correct management of pharmacological treatment should try to prevent the onset of side effects, through reviewing and identifying the risk factors, monitoring metabolic markers, and promoting potential modifiers of the course of metabolic syndrome (i.e., lifestyle, polypharmacy) [72]. For example, patients with a history of weight or diabetes might avoid medications that are known to increase the risk of these side effects, such as risperidone and olanzapine [73,74]. Some cardiovascular risk factors (QTc prolongation, diabetes, and weight gain) also seem to have dose-dependent side effect profiles that might require monitoring at dose changes [68,74,75].

We found that the most common side effects of cannabinoids were somnolence, increased appetite, and irritability. As many patients were taking concomitant medications, it is not possible to determine if these adverse events were caused by the cannabinoids or by other drugs. Only Aran et al. [62] reported a severe adverse event (a psychotic episode) that resolved after stopping the cannabinoid oil solution and treating the patient with an antipsychotic (i.e., ziprasidone). None of the included studies reported cardiac adverse events (i.e., QTc prolongation) or severe metabolic side effects (i.e., hyperlipidemia, diabetes, hyperprolactinemia) that could depose for a better safety profile in cannabinoids than antipsychotics.

### *4.2. Mechanisms of Action: The Role of Excitatory*/*Inhibitory System*

The three papers published by Pretzsch et al. [59–61] primarily investigated the modulation of the brain's excitatory and inhibitory systems in adults with ASD and neurotypical controls, after a single dose of 600 mg of cannabinoids (CBD and CBDV). The findings evidenced a CBD-related increase of glutamate (excitatory system) in subcortical regions (i.e., basal ganglia) and a decrease in cortical regions (i.e., dorso-medial prefrontal cortex), both in subjects with and without ASD. Conversely, CBD increased GABA transmission (inhibitory system) in critical and subcortical regions of neurotypical subjects, while decreased it in the same areas of the ASD group. This confirmed the hypothesis that GABA transmission could be altered in people with ASD [17,76,77]. Moreover, CBD modulated low-frequency activity, used as a measure of spontaneous regional brain activity, and functional connectivity in the brain of adults with ASD [61]. The experiment with CBDV replicated the findings of the CBD study for glutamate transmission, but not for GABA [60].

Such findings might further explain the link between autism and seizures. About 25% of children with treatment-resistant epilepsy are comorbid with ASD and often present other severe comorbidities, such as sleep disturbances, intellectual disability, or other psychiatric conditions [78]. Additionally, as mentioned above, epilepsy is one of the most frequent medical comorbidities in people with ASD [12–15], and is also found to be more common in those patients with autism-like behaviors as part of phenotypes of genetic syndromes (i.e., Angelman, Rett syndrome, etc.) [79]. This overlap might be explained by common biological mechanisms. Like ASD, in fact, epilepsy is characterized by an imbalance between excitatory and inhibitory transmission in the central nervous system [80]. The presence of seizure in ASD could also be responsible for the onset of challenging behaviors [81]. Therefore, it could be hypothesized that treating seizures with cannabinoids might also exert a significant impact on externalizing symptoms.

Unfortunately, the action of cannabinoid administration on other neurotransmission systems was not investigated in autistic individuals. As mentioned in the introduction, studies on animal models provided evidence for the role of serotoninergic [42,82,83] and dopaminergic systems [84]. However, their role in the etiology of ASD still needs to be clarified.

### *4.3. Limitation: Heterogeneity of Studies*

The present systematic review included ten published studies (of which two conference abstracts) and four ongoing trials. Looking at Table 1, which summarizes the characteristics of the studies, it is possible to notice that the works conducted to the present date are highly mixed in terms of study design and participants. Some studies included both children and adults, other participants with and without epilepsy (which is not irrelevant, as cannabinoids act on the excitatory/inhibitory system, altered in epilepsy). Additionally, the intake of concomitant medications acting on the GABAergic system might represent a bias. Finally, the level of functioning or the intelligence quotient (IQ) was specified only in four studies [59–62]. The characterization of samples is fundamental as target symptoms might vary. Individuals with associated intellectual disability (ID) typically present more severe behavioral problems that could benefit from the use of cannabinoids. People with higher levels of functioning, instead, could present more frequently concurrent anxiety disorders. This is important because different target symptoms need different outcome measures.

Other caveats rely, in fact, on the heterogeneity of outcomes and administered treatment. It seems evident that the studies were mainly explorative and did not report a differentiation between primary and secondary outcomes. Moreover, measures were often multiple and combined both core and associated ASD symptoms (e.g., global impression). Standardized measures were used only in a few studies, and in some cases, the authors reported only the proportion of improvement for each symptom. This important issue confirms the findings of a recent systematic review of 406 clinical trials [85], which pointed out that the tools used in autism research are heterogeneous and non-specific. This fragmentation might significantly hamper the comparison between studies and the understanding of the real effectiveness of cannabinoids in the ASD population. In addition, the majority of studies used combinations of CBD and THC in different concentrations and ratios, even in the same study sample. It is indisputable that the dosage of cannabinoids needs to be calibrated on individual characteristics

(e.g., weight), but again, the use of different concentrations/ratios does not allow to compare studies and find the optimal therapeutic range.

Importantly, seven of the included studies did not have a control group. Only the three studies conducted by Pretzsch et al. [59–61] administered a control treatment (placebo), while also using a control group (healthy subjects). However, these studies principally aimed to explore the neural modifications induced by the assumption of CBD or CBDV in individuals with ASD, while also evaluating the differences with neurotypical subjects. Even if not directly designed to study the efficacy and safety of cannabinoids in ASD, the completion of similar studies appears fundamental as they might elucidate the neurochemical functioning of the autistic brain.

### **5. Conclusions**

Our systematic review was the first to critically summarize the published and ongoing studies investigating the use of cannabinoids in the ASD population. Despite cannabinoids having shown promising effects on some ASD-associated problems (e.g., aberrant behaviors, sleep disorders, hyperactivity, seizures), their efficacy on core symptoms (i.e., socio-communication impairments, restricted interests, and stereotypies) remains largely unknown. The main limitation of the present paper is the absence of a statistical analysis of results that was hampered by the heterogeneity of study design, populations, type of cannabinoid, and particularly, outcomes, and measures. Future studies investigating the acute effects of cannabinoids in people with ASD on neurotransmitters levels could clarify the mechanisms of action of cannabinoids. Moreover, the comparison with healthy samples might clarify at least some aspects of the etiopathology of ASD and lay the ground for potential treatments for core and associated symptoms. Even if some clinical trials are ongoing, there is the need for further long-term studies, with homogeneous samples in terms of age, medication use, level of functioning, and presence/absence of seizures. Of great importance would be the choice of specific primary and secondary outcomes, focused on the cluster of symptoms that could benefit from the use of cannabinoids.

**Author Contributions:** Conceptualization, L.F.-P., V.C., S.T., and I.C.; methodology, L.F.-P.; investigation, L.F.-P., V.C., S.T., and I.C.; data curation, L.F.-P., V.C., S.T., and I.C.; writing—original draft preparation, L.F.-P., V.C., S.T., and I.C.; writing—review and editing, A.P., M.S.S., C.M.D.-C., and E.A.; supervision, A.P., M.S.S., C.M.D.-C., and E.A.; All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Autistic-Like Features in Visually Impaired Children: A Review of Literature and Directions for Future Research**

**Anna Molinaro 1,\*, Serena Micheletti 1, Andrea Rossi 1, Filippo Gitti 1, Jessica Galli 1,2, Lotfi B. Merabet <sup>3</sup> and Elisa Maria Fazzi 1,2**


Received: 9 July 2020; Accepted: 24 July 2020; Published: 1 August 2020

**Abstract:** There remains great interest in understanding the relationship between visual impairment (VI) and autism spectrum disorder (ASD) due to the extraordinarily high prevalence of ASD in blind and visually impaired children. The broad variability across individuals and assessment methodologies have made it difficult to understand whether autistic-like symptoms shown by some children with VI might reflect the influence of the visual deficit, or represent a primary neurodevelopmental condition that occurs independently of the VI itself. In the absence of a valid methodology adapted for the visually impaired population, diagnosis of ASD in children with VI is often based on non-objective clinical impression, with inconclusive prevalence data. In this review, we discuss the current state of knowledge and suggest directions for future research.

**Keywords:** autism spectrum disorder; autistic-like features; social-cognitive development; stereotypical behaviors; visual impairment

### **1. Introduction**

Research into the presence of autistic-like features among blind children has a long history. Starting from a series of publications appearing in the 1960s and 1970s [1–7] (which considered that autistic-like behaviors showed by blind children were a possible consequence of the lack of visual experience on the development of self-image and self-representation), researchers and clinicians have increasingly reported commonalities between children with autism spectrum disorder (ASD) and those with visual impairment (VI), particularly with regard to social interaction and communication skills [8–12]. Restricted symbolic play, difficulties in social interaction with peers and imitation, echolalic speech, and increased stereotyped behavior have all been frequently reported in blind children [9,10,13,14]. Indeed, these behaviors resemble subjects with ASD and are often termed "blindisms" since they are explainable in the context of VI [15]. However, the similarity between these "blindisms" and "autistic-like" behaviors, coupled with the lack of ASD assessment tools specifically designed for blind and visually impaired children, complicates the diagnosis of ASD in these individuals. Finally, while the estimated prevalence of ASD among sighted children is between 1 and 2% in Europe [16], determining the prevalence in the visually impaired population still varies greatly, ranging from 2 up to 50% [12,17–19]. The underlying mechanisms related to autistic-like symptoms shown by some children with VI, as well as how certain visually impaired children are able to overcome these developmental challenges, remains poorly understood.

Reviewing available literature, it remains to be established whether ASD in VI is primarily a neurodevelopmental condition that occurs independently of the visual disorder (possibly with a common causal agent such as a genetic defect), or is secondary to the VI, and is more closely associated with the disruption of vision on early interactive experiences, or represents a combination of the two [20,21]. For many of the children who are blind and also display features of ASD, it is possible that their characteristics (while being representative of ASD), actually follow a different developmental pathway than those who have ASD and normal vision.

Referring to papers published between 1958 and 2020, the purpose of this review is to provide a discussion of these important, yet still controversial issues. Using two electronic databases (PubMed and Google Scholar), we included combinations of the following search terms: "autism", "autism spectrum disorder", "blindness", "sight loss", and "visual impairment". Citations identified from the automated search were manually verified for appropriateness.

The original search yielded 1613 documents, that were reduced to 921 following duplicate removal. Independent screening (by the first and second authors of this review) of the study titles and abstracts was carried out to identify studies that were most relevant to the aims of this review.

Articles were included for further inspection if they satisfied the following inclusion criteria: (1) explored the mechanisms that may explain the observed relationship between VI and ASD, taking into account the nature and role of contributing risk factors such as the severity of VI, type of blindness, age at onset, and other associated impairments; (2) discussed specific behavioral and neurocognitive traits in visually impaired compared to ASD children such as: joint attention, language, verbal and non-verbal communication, theory of mind, stereotypical behaviors; (3) described the approaches employed to assess ASD in visually impaired children, with specific attention given to the fact that the most common methods used for scoring autistic behaviors include several items which are directly dependent on visual abilities; (4) included participants between 0 and 18 years of age.

Articles were included (irrespective of the age range) if they added relevant information, as judged by the authors. Articles were excluded if they were focused on the prevalence and/or the type of ophthalmic problems in the ASD population or the characteristic of visual deficit in specific genetic/metabolic conditions which also presented autistic-like traits. This resulted in the exclusion of 821 papers that did not meet these inclusion criteria and lead to a final sample of 100 studies for the purposes of qualitative synthesis.

### **2. The Observed Relationship between Visual Impairment and ASD: Possible Underlying Mechanisms and Contributing Risk Factors**

Since the first reported description by Keeler [22] of a co-occurrence between blindness and autistic behavior, various studies have focused on identifying specific types of ophthalmological disorders as potential organic etiological factors, suggesting the presence of common causal agent potentially independent from the VI itself [22–24]. Keeler [22] hypothesized that autistic behavior in children with "retrolental fibroplasia" (i.e., retinopathy of prematurity) resulted from a combination of brain damage, blindness, and emotional deprivation. Wing [25] listed several similarities between children with ASD and children with partial blindness and partial deafness caused by maternal rubella. Chase [24] found a gradient of autistic-like features in a group of 246 individuals with "retrolental fibroplasia", but none had a clear diagnosis of infantile ASD. The author also reported a strong relationship between autistic-like symptoms and neurological findings. Chess [23] assessed the behavioral data of 243 children with congenital rubella and reported that the common component accounting for ASD in these observed cases was brain damage. Rogers and Newhart-Larson [26] reported the presence of ASD in 5 preschool children with Leber's congenital amaurosis and compared these children to a control group with congenital blindness due to other causes and typical development, suggesting that cerebellar deficit in some patients with Leber's congenital amaurosis could provide a neurobiological basis for the behavioral similarities observed between these patients and sighted autistic individuals. Ek and colleagues [27] studied the relationship between blindness due to retinopathy of prematurity (ROP) and

ASD and concluded that an ASD diagnosis in blind children is likely to be mediated by brain damage or dysfunction. Fazzi and colleagues [28] submitted 24 children affected by Leber's congenital amaurosis to a modified version of the Childhood Autism Rating Scale (CARS) by excluding item VII about visual responsiveness [29]. According to their results, 20 children were found to be non-autistic, 4 presented with mild/moderate ASD, and no child was found to be severely autistic. Nearly every child presented some degree of restricted and stereotyped patterns of interest, adherence to specific routines or rituals, difficulties in adapting to environmental changes and showed dysfunctional relationships with other people or in their social and emotional responsiveness. Impaired verbal communication, a tendency for passiveness, and difficulties in using their bodies were also observed. In a prospective study, Garcia-Filion and colleagues [30] demonstrated that autistic-like features occurred with high frequency in children with mild to severe optic nerve hypoplasia (OHN). Since the study included children with various degrees of VI (including those with unilateral ONH), this supported the hypothesis that the autistic component could have a neurological basis, rather than being connected to the visual impairment itself. Similarly, Parr and colleagues [31] assessed the prevalence of social, communicative, and repetitive or restricted behavioral (SCRR) difficulties and defined clinical ASD in 83 children with ONH and/or septo-optic dysplasia (SOD), finding the presence of at least one SCRR difficulty in 58%. Thirty-four percent of the sample was clinically diagnosed with ASD. Moreover, SCRR difficulties and ASD were statistically higher in children with significant cognitive impairment and profound VI and there was no evidence that additional neuro-anatomical abnormalities were a further risk factor in the development of ASD. These data suggested the authors that ASD in children with OHN and/or SOD may arise through different mechanisms compared to the idiopathic ASD population.

Jutley-Neilson and colleagues [32] evaluated the occurrence of ASDs in 28 children with SOD and 14 with ONH. According to the previous study of Parr et al. [31], 33% of children with SOD and ONH received a clinical diagnosis of ASD. Using the Social Communication Questionnaire, 55% of the children met the cut-off threshold for further investigation to differentiate between ASDs and non-ASD (row scores ≥15) and 21% met the cut-off for further investigation to differentiate between ASD and autism (row scores ≥22). The authors identified the degree of visual loss and the severity of intellectual disability as good predictors for ASD, and recommended that children with SOD/ONH would benefit from routine ASDs screening. De Verdier et al. [33] described neurodevelopmental impairments in children with congenital or early infancy blindness born over a decade in Sweden; they found that ASD was one of the most common additional impairment (38% of these population) and that the prevalence was higher in children with ONH (70%), in children with ROP (58%), in children with microphthalmia/anophthalmia (44%), and in children with LCA (36%).

In a different perspective, some researchers [8,10] have suggested that focusing on the cause of blindness is irrelevant, emphasizing rather on the role of sensory deprivation and environmental risk factors in the emergence of autistic-like behaviors. Goodman and Minne [34] assessed 17 congenitally blind children (aged 4 to 11 years) without any additional impairment using the Autism Behavior Checklist [35]. The prevalence of ASD in this sample was 23.5% using a critical cut-off number of symptoms to determine diagnosis. In a study by Brown et al. [36], a prevalence of 20.8% was determined investigating 24 congenitally blind children without any neurological damage (aged 3 to 9 years) using the CARS [29]. Hobson and colleagues [8] found that nine congenitally blind children were similar in their range of clinical features with nine sighted autistic children (age- and verbal IQ-matched).

Regardless of the ophthalmological diagnosis, the potential vulnerability may partially be caused by early blindness and may not only be limited to a lack of vision, but also to severe and early damage to the visual system, threatening the development of mental and emotional processes that allow children to organize experiences and develop different areas of learning [37].

Data from healthy populations suggests that mutual influences between vision and emotion start at very early stages of information processing [38]. The brain regions involved in mental and emotional states include the prefrontal cortex, limbic structures, and the insula as well as visual areas [39]. In particular, enhanced activation of the occipitoparietal regions (corresponding to the dorsal visual processing stream) has been reported during the emotional processing of visual stimuli [39,40]. Abnormal neuronal responses of these cortical regions, such as what could be expected in cerebral visual impairment, may contribute to an impairment in emotional recognition [41].

Recently, Fazzi et al. [19], among 214 children with cerebral causes and 59 with peripheral causes of vision impairment, found that ASD was more prevalent compared to a general population, and that the prevalence varied according to the type of visual disorder (2.8% for cerebral and 8.4% for peripheral visual impairment). Moreover, the presence of autistic symptoms was consistent with the diagnosis of ASD only in subjects with cerebral visual impairment, while in those with peripheral visual impairment, many symptoms related to visual loss overlapped with the clinical features of ASD, making clinical diagnosis more challenging.

Moreover, it is not clear why some children fail to progress, or even regress their communicative and cognitive skills. Mukaddes et al. [17] showed that individuals with blindness and ASD have greater neurological impairment and more severe visual impairment with respect to individuals with blindness only. This suggests that, regardless of the cause of blindness, brain damage remains an important contributing factor for the development of ASD. Certain investigators [42,43] have also described a phenomenon of serious developmental disruption or "setback" which seems to occur between the 15th and 27th month of age. An explanation for this setback occurring in children with profound visual impairment relates to the notion of a sensitive or critical period of brain development within the first to second year of life that relies on normal visual experience occurring within this period [43]. Finally, in a retrospective study by Waugh and colleagues [44], a higher proportion of brain lesions detected with magnetic resonance imaging (MRI) was associated with greater developmental setback in children with visual impairment, which may be an early manifestation of clinical ASD [42,43].

More recently, Vervoled et al. [45] reviewed the literature associated with developmental setback in blind and visually impaired children. Although the authors recognized the period around the second year of life as most vulnerable in these children (particularly in those with neurological abnormalities), they pointed out that the individual variability in development and the wide variability in the methodological aspect make it difficult to draw conclusions on the occurrence of developmental setback in blind and visually impaired children.

It is crucial for professionals who are in contact with these children to recognize these developmental risk signs, namely the presence, persistence, and entrenchment of a whole series of behaviors which are expressions of considerable social isolation. These behaviors include remaining in a lying down position, lack of attention towards environmental stimuli, absence of smiling (or problems eliciting smiling), poor adaptive use of the hands to explore and recognize objects, absent or poor "reach on sound" after the fourth trimester of life, and persistence of excessive and non-functional use of the mouth as the main interface with the environment [34].

### **3. Behavioral and Neurocognitive Traits in Visually Impaired Compared to ASD Children**

Although visually impaired children do not present a typical personality profile, it is possible to recognize certain frequently occurring traits, namely high levels of anxiety, some difficulties in social interactions, an excessive production of speech (with declarative rather than communicative intent) serving to fill an emotional void, behavioral rigidities [19], that need to be early detected and constantly monitored. There is a remarkable risk that a blind child's personality can be limited to body sensations and that the bridge between the self and the outside world can become unstable or even non-existent. If this issue becomes a source of excessive self-restraint, then the onset of problematic behaviors, such as stereotypes, becomes more common in these children [10,46,47]. A presentation of the most representative behavioral and neurocognitive traits that lead to consider the presence of overlapping symptoms between VI and ASD is listed below.

### *3.1. Joint Attention*

Sighted babies and young children use visual behaviors like eye contact, gaze following, and joint attention to set up and sustain communication and to learn about the behavior and intentions of others, especially during the pre-linguistic stage [48]. These early visual behaviors and associated interactions appear to lay the foundation for developing emotionally secure attachments, language, and achieving knowledge about self and others [48]. Joint attention is a triadic relationship that arises in the first months of life, based on mutual gaze between the child, an object, and a social partner, in which both the child and the partner are aware of one another's attention towards an object or event [49]. Visual perception is crucial in this interaction [50].

Joint attention occurs in blind children as well, even if they can acquire it later and differently with respect to sighted children [51]. Infants with VI can be less engaged in joint attention: they usually tend to respond to social interaction with decreased visual attention, pointing [49], or smiling [52]. They are reported to tend to turn head/body away from caregivers and to initiate play interactions with their mothers less often than their sighted peers [13,53]. These behaviors can be interpreted by caregivers as simply a lack of interest, decreasing positive social exchanges [54]. Dale and Salt [48] found that less than a third of the children with profound VI aged 28–40 months were able to share interests and experience with a toy or share interest in an event, in contrast to the great majority (over 80–90%) of the severely visually impaired and sighted children. In a longitudinal study, Urqueta Alfaro and colleagues [54] showed that, in 12-month-old visually impaired infants, the reduction of contrast sensitivity predicted the percentage of time spent in joint engagement. Caregivers of infants/children with VI can learn to interpret and sensibly respond to their baby's signals through non-visual means [55,56]. Rattray and Zeedyk [57] identified touch, vocalizations, and facial orientation as alternative means to maintain the quality of communicative interactions between mothers and their infants with VI, even if it was not explicitly explained. In their study, infants with VI used active touch during shared attention as a tactile form of communication and made use of facial orientation to a lesser degree than touch and vocalizations, indicating that facial orientation is not as important as an alternative communication means [57].

The atypical development of joint attention in infants with VI, compared to their sighted peers' developmental patterns, is considered by some authors as a typical sign of ASD [58]. The emergence of joint attention may in fact be disrupted by ASD [59,60]. However, as recently outlined by Urqueta Alfaro et al. [54], the mechanisms and timelines of joint attention development in infants with VI is obviously different from what is expected in infants with typical development, as described above. Failing to recognize this may put VI children at risk of being wrongly labeled as autistic [54]. However, if in ASD the absence/reduction of interest in shared objects and people is a typical feature, as defined in the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) [61], alternative means beyond visual attention is shown in infants with VI to maintain the quality of communicative interactions [62,63].

### *3.2. Language and Communication Skills*

Vision is implicated in general language development, as visually driven joint attention experiences in early childhood provide a framework within which language learning occurs [64]. Despite marked variability in visual profiles, children with both peripheral and central VI may exhibit the presence of language and communication disorders [37,64]. This can be a reflection of the visual deficit itself on early interactive experiences, or represent an associated neurodevelopmental condition that occurs independently of the VI or, more frequently a consequence of the two conditions [19].

Communicating with other people can be a challenge both for children with ASD and children with VI, especially as the pragmatic component of language is concerned [64]. As with individuals with ASD [65], children with VI have unique methods of communication (relying instead on non-verbal communication techniques, echolalia, moving from topic to topic, speaking with no eye contact) that may be important in overcoming social barriers.

In children with VI due to central origin, language disorders have been described [37], and may be influenced by both the degree of visual loss and by widespread brain damage that impacts brain network organization and consequently, the development of general neurocognitive functions including language [19].

Language skills have also been widely detailed in children with VI due to peripheral origin and have been considered in the past as the most promising indicator of peripheral VI children's ability to compensate for early deficits in developing inter-subjectivity [66].

Differently from children with ASD, language may be a developmental domain which provides blind children with alternative non-visual strategies for social development [67] but, similar to children with ASD, adverse outcomes in social communication may be also present in children with both peripheral visual impairment (PVI) and cerebral visual impairment (CVI), probably given to disruptions in visually guided experiences and visual behaviors, which are seen as precursor milestones for subsequent social development [62].

Language includes shared understanding of what words mean (lexicon/semantics); the capacity to change words in systematic ways (morphology); and rules that govern word order in a sentence (syntax). Speech and phonology are the oral means of communicating language. The use of language as a social tool (pragmatics) involves a complex set of rules about using eye contact, interpreting nonverbal messages together with words that may have a different literal meaning. In blind children due to peripheral origin, structural language skills, namely phonology, morphology, and syntax, may allow for fluent conversation and have been described as typically developed, differently from most of the autistic children, in which language impairment is reported [61]. On the other hand, semantic and pragmatic skills, that are required for successful socio-communicative functioning, have been described by Tadic and colleagues [64] as being poorer in both VI and ASD.

Mills [68] outlined that children with VI due to peripheral origin usually develop fully intelligible speech within the same time frame as sighted children. In a recent study, Feng et al. [69] showed that they have enhanced attentional sensitivity to "non-visual" components of language such as phonetic-phonological components. Roder and colleague [70] showed that blind participants were more efficient than sighted children in terms of phonological processing. They score consistently higher than their sighted peers on tests of verbal working memory [71–73] as well which, on the contrary, is usually impaired function in children with ASD [74].

With regards to the lexical component, Vinter and colleagues [75] showed that blind children tended to define words denoting concrete animate or inanimate familiar objects evoking their close perceptual experiences of touch, taste, and smell. It was different from what sighted children, who relied their definition on visual perception, and produced more visually oriented verbalism. They also may exhibit atypical conceptual and semantic development [76,77] and demonstrate specific deficits in understanding visual concepts that they have learned through language and not through direct experience. Given fewer opportunities to benefit from traditional classroom education, blind children, due to peripheral disorders, have shown that they may score below their sighted peers on comprehension, similarity, and vocabulary subtest [70,71,78]. Similar to those with ASD [79], young blind children have a limited capacity for generalizing a given word for other items in the category, and use a word for the original referent or only very few items in the category [79].

No significant difficulties with syntactic development have been described in children with PVI [68]. If complexity of structures is analyzed, blind children show similar performance to that of sighted children not only during the first steps of grammatical development, but also taking into account the acquisition of complex sentences [80]. Blind children's morphological development, with the exception of personal and possessive pronouns usage, has not been described as delayed nor impaired in comparison to the one of sighted children [80]. Dunlea and Andersen [81] have suggested that young blind children use few morphemes such as plural, 3rd person of present indicative, and locative prepositions in organizing structures and imitations. Blind children seem to start to productively use pronouns very late (around age 4), and they produce a great proportion of reversal errors (1st person for 2nd person pronouns and vice versa) [7,81]. On the other hand, language can be delayed in children with CVI [64], whose ability to respond to stimuli has been described by parents as altered [37].

Considering pragmatic aspects, the tendency to use words whose concrete referent is unknown to the speaker, a behavior named verbalism, is another common language behavior of both children with peripheral VI and ASD [82], as is the tendency to use self-oriented language instead of externally oriented language or the tendency to produce a lesser proportion of verbal expressions to offer, show or draw another person's attention [80].

Echolalia represents one of the peculiar ways of communicating found in children with peripheral VI and ASD. However, learning and using whole phrases or formulas for specific contexts and activities allows to participate in social interactions and share activities with other people [83], while the social role of echolalia in ASD is controversial [84]. Like children with ASD, blind children may ask many questions, sometimes inappropriately, and may make 'off-the-wall' comments [83]. They also tend to refer more often to their personal experiences than sighted children when evoking familiar objects [75]. Mothers of children who present severe peripheral VI seem to take more frequent and longer turns at speaking or with other forms of communication than do mothers of sighted children, resulting in an asymmetry between relative dyads' experiences [67,85]. Parents of blind children also tend to use more response control, more test questions instead of real questions, more requests and more repetitions [56], use more imperatives and requests, and were more likely to introduce the topic of conversation [86] than do mothers of sighted children.

### *3.3. Stereotypical Behaviours*

The presence of stereotypical behaviors in children with VI has also been observed and extensively reported in several studies [47,87–91]. Although stereotyped movements are a defining characteristic of ASD, there is also some evidence of a distinct pattern in the visually impaired group. Gal and colleagues [91] assessed self-injurious and other stereotyped movements in children with ASD, vision impairment, intellectual disability, or hearing impairment and in typical children. The group with visual impairment had the second greatest prevalence of manneristic behaviors, but it is also engaged in forms of stereotyped movements sufficiently distinctive and rarely present in other groups. Particularly, visual self-stimulatory behaviors, including eye poking, eye pressing, eye rubbing (which may lead to a number of ocular complications including infections, keratoconus, and corneal scarring), light gazing, and staring, form a large portion of the stereotyped exhibited behaviors by visually impaired children [47,88,92–95]. These behaviors are generally exclusive to children with VI and are especially present in children with peripheral visual impairment: Jan and colleagues [96] found that those children with a retinal disorder such as Leber's congenital amaurosis or retinopathy of prematurity were the most intense eye pressers. Other stereotypical behaviors typically observed in visually impaired children are motor stereotypes. These include repetitive head/body rocking, thumb sucking, jumping, swirling, and repetitive hand/finger movements [89,92,93,97–99]. However, in a study by McHugh and Lieberman [94], it has been suggested that body rocking often occurs also in children with retinopathy of prematurity and severe VI. This behavior is most likely to occur in those with a CVI, perhaps because of poor motor development in these subjects [97,100]. Similarly, flickering fingers in front of the eyes while staring at light is common in children with CVI and has been interpreted as an extension of light gazing behaviors [101,102].

Various interpretations of stereotyped behaviors have been reported in the literature [92]. For example, some authors considered eye-digital signs as a means to self-stimulate the sensation of light, producing phosphenes (light sensation that result from mechanical pressure on the eyeball that stimulates photoreceptors and activates intact visual pathways) [47]. Other authors have suggested that these behaviors may be caused by an imbalance of neurotransmitters, especially dopamine and serotonin, due to a damage in the central nervous system [100]. Theoretical approaches have been used to explain stereotypical behaviors from a behaviorist, developmental, and functional perspectives [100]. Specifically, children with VI might acquire and maintain stereotyped behaviors because they are

reinforced by their consequences (e.g., avoiding an unpleasant situation, or drawing attention), because of a delayed motor development (as an expression of neuromuscular maturation processes), or because these behaviors can act as modulators of arousal state, increasing or decreasing the level of stimulation (e.g., thumb sucking in situations of under-stimulation, repetitive hand movements, and jumping in situation of overstimulation) [91,93,100,103]. According to these hypotheses, the frequency of stereotypic behaviors in visually impaired children seems to decrease with age [93,97] and children affected by isolated visual deficits present stereotyped behaviors which are generally more reversible than the ones found in children with additional disabilities [47]. Further, studies have supported the view that the prevalence and the type of stereotyped behaviors are directly related to the severity of visual impairment [91,92,97]. Early intervention is very important in order to stop stereotyped behaviors from becoming established, entrenched, and irreversible [92]. The purpose of this intervention is to provide support, but also to promote opportunities and situations which will allow children to re-establish contact and communication with the world around them. The way VI impacts children's development does not solely depend on the sensory limitation itself, but also on the degree caregivers and society accommodate to these children's needs and strengths [54]. Sensitive parenting in which parents are vocally and tactually responsive to their children's actions facilitates many blind infants' ability to learn their interpersonal effectiveness in the social world.

Instead of focusing mainly on visual attention and facial expressions, parents can be encouraged to become more sensitive to their children's unique inviting signs, pay more attention to the use of movement, touching, tickling, vocalizing, and speech in eliciting physical-tactile and vocal interaction routines [67,104] and to look at body pointing and other unique nonvisual referring signs to create good levels of communication and shared affective meaning about objects and events in the immediate environment [63]. Moreover, the possibility to refer to autobiographical memory is very important in blind children because it is the way they can understand the world. Consequently, unexpected changes in their environment can disturb them and parents should pay attention to guarantee coherence in the personal environment of these children [75].

### *3.4. Theory of Mind*

Baron-Cohen [51] has argued that an individual's eye movements and relationship with a "shared visual attention mechanism" play a key role in establishing a theory of mind module in the developing infant. Hobson [62,105] described foundations of theory of mind and interpersonal understanding in terms of a child taking part in triadic interactions that involve both the child's and the partner's awareness of the other's mutual focus of attention to a third object or event (joint attention). Through joint attention, the child can understand the other person's attitude towards an object [49], and this behavior is usually carried out via the visual modality [105].

Deficits of theory of mind (ToM) in ASD have been related to a lack of inter-subjectivity in ASD children [106]. In other words, an inability to understand and anticipate the thoughts and emotions of others has been associated with a lack of shared social understanding [107]. Children with VI may have difficulties in understanding thoughts and emotions of others as well since, as Bedny et al. [108] highlighted, congenital blindness can alter two important sources of information that can be considered as building blocks of ToM. At first, it does not permit blind children to learn about other people's minds via visual observation of other people's facial expressions or body movements. Secondly, it alters first person experiences of mental life. Specifically, children with VI can understand and share abstract features of other's experience, but could not have the same experience [108]. It is interesting to note that, differently to individuals with ASD, whose ToM disrumption is debated since the Baron Cohen's study on 1985 [106], children with congenital VI may present with a delay, but not a deficit in the ToM construction [109], despite not having access to some (visual) information about the mind during development. Eventually, as adults, they can develop a functional and effective ToM, including an understanding of other people's experience of sight [110].

Evaluating ToM in children with VI can be challenging because many tests used rely on visual capacities. This can help explain 4–7 years delay previously described in developing ToM in congenitally blind children [21,111–114]. False belief tasks have been particularly used in the evaluation of ToM in children [109]. The first type of false-belief tasks, in which children are expected to predict or explain another agent's behavior in terms of the agent's mental states (e.g., Baron-Cohen and colleagues' "Sally-Anne" task), have been used in assessing ToM in children with VI [113]. Sometimes, they have been based on tasks in which visual experience has a significant role [112–114]. Because VI can affect the development of ToM, purely due to visual and perceptual deficits, different tasks from the first-order FB have been needed to evaluate ToM in blind children. Second-order FB tasks were introduced later to examine people's belief about others' belief (i.e., "John thinks that Mary thinks that ... " [115]), with positive performance provided by children with VI [109,116]. As a matter of fact, in a recent study, the introduction and use of more reliable tools has identified a similar development of ToM capacities in blind children as compared to sighted peers [109]. In Bartoli et al.'s study [109], 17 children with PVI or blindness underwent an adapted version of the ToM Storybooks and performed similarly to the ones of matched typically developing children, matched on chronological age and gender. Pijnacker et al. [116] administered to blind children several first-order and second-order auditory tasks, showing that the visually impaired children's performances did not differ from sighted children, matched on gender, age, and verbal IQ. These data suggest that the visual nature of the tests or the stimuli should systematically be considered.

Different performance on ToM tasks seem to be related to the type of VI as well. In children with PVI, a delay in ToM development was described in the first studies [21,111–114], not found in the more recent ones [109,116]. Children with CVI may present a more compromised neurocognitive profile than what is usually expected in children with PVI [117]. Begeer et al. [118] found that ToM performances in children whose blindness involved the optic neural pathways were delayed, compared to the performances of children whose blindness did not involve any neural damage. The detected difficulties in interpreting others' intentions and reactions that children with CVI showed, could have reflected the deleterious effect of CVI on the understanding of the social context and facial expressions [37]. These difficulties may also be a consequence of the low IQ levels that children with CVI may present [19] and that are in relation to ToM tasks [111]. As suggested by Bartoli et al. [109], a possible future area of research could compare VI children and children with autism matched on verbal IQ, age, and gender, in order to further understand the role of visual experiences on ToM development.

### **4. Methods Used to Assess ASD in Visually Impaired Children**

Since there are no consistent results in terms of the relationship between specific types of ophthalmological problems, severity of VI, and the role of associated handicaps (such as hearing deficits, cerebral palsy, epilepsy, and other intellectual disabilities), and their relationship with ASD, it seems necessary to find a new approach when explaining autistic symptoms in the blind and in the sighted population [18]. ASD is known to be highly heterogeneous, and this has made it hard to define a clear phenotype. Although biologically based and with an evident genetic component [119], ASD is defined and diagnosed based on behavioral difficulties, concerning social interaction and the development of communication skills, and repetitive behaviors and restricted interests. Since ASD is defined by a common set of behaviors, it is best represented as a single diagnostic category that is tailored upon the individual's clinical presentation including clinical characteristics and associated features [120]. Assessing ASD in blind and visually impaired children is a very delicate process in which most of the common methods used to score autistic behavior, including several items linked to vision [121,122] are applied. Therefore, in clinical practice, these standard assessment tools may not be appropriate for specific VI populations [123]. Some authors have designed checklists and/or questionnaires as screening tools to guide further clinical evaluations. Hobson and colleagues [8,20] suggested a checklist containing some clinical features typically found in ASD (derived from DSM-III-R) and used it to interview the children's teachers. Jutley-Neilson and colleagues [32] used the Social Communication

Questionnaire (SCQ), a standardized parent report measure to evaluate communication skills and social functioning in children. Many of the items in the questionnaire involved situations that can only be experienced by sighted children, and the authors highlighted that the SCQ was not as sensitive and specific for visually impaired children. Hoevenaars-van den Boom and colleagues [123] aimed to identify ASD-specific behaviors in deaf-mute people. For this purpose, authors have developed the "observation of characteristics of ASD in persons with deaf-blindness (O-ADB)", an originally semi-standardized observation tool based on the Autism Diagnostic Observation Schedule [124], the Autism Screening Instrument for Educational Planning [35], the Autism Diagnostic Interview Revised [125], and on the Van Dijk Approach to Assessment [121].

The absence of a valid methodology for this population has often led to the conclusion that diagnosing ASD in children with visual impairment should be based on clinical judgment [122]. However, more recent efforts have been made to adjust or modify the assessment tools used to assist with the clinical diagnosis of ASD in VI children. For example, most authors administer the modified CARS and exclude Item VII on visual responsiveness in order to identify children at risk of developing pervasive developmental disorders [8,20,26,28]

Recently, Williams and colleagues [126] have started applying systematic modifications to the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview, Revised (ADI-R) in order to assess symptoms of ASD in visually impaired children (the majority of whom have ONH). This pilot study has provided preliminary evidence regarding how to modify ASD measures which are now more useful in the diagnostic evaluation of visually impaired children and both these tools have shown a good agreement with clinical diagnoses. Authors have concluded that additional research is needed to validate the modified measures in larger samples which may include different diagnoses and levels of visual impairment, and also to follow visually impaired children over time to identify common developmental paths and outline whether specific symptoms change over time [126].

In this direction, a recent study by Fazzi et al. [19] employed systematic modifications (i.e., materials and scoring procedures) to the ADOS 2 [127] (second edition) to assess symptoms of ASD in visually impaired children, taking into account the specificity of type of visual disorder (cerebral vs. peripheral visual impairment). In children with CVI, the use of the modified assessment tool (M-ADOS 2) did not modify the diagnostic category, and the clinical diagnosis matched the ADOS 2 classification and the M-ADOS 2 classification in almost all patients. Conversely, among participants with PVI, 16.9% were classified as autism/autism spectrum in accordance to the ADOS-2 scale but only 10% were confirmed using the M-ADOS 2, exhibiting good concordance with the clinical evaluation result. Although preliminary due to the small sample size, the study suggested that autistic-like finding in children with PVI are more influenced by the degree of VI, and specific symptoms may be more reliable than others in discriminating ASD in VI children. The authors point out the importance of using appropriate adapted tools in PVI subjects to avoid overestimation of ASD that may be confounded by the presence of VI and symptoms and habilitation strategies associated with ASD should take into account possible differences in the context of impaired visual abilities.

The utilization of modified assessment tools, specific not only for ASD but also for VI, matched with a careful clinical observation, is needed in order to ensure a correct diagnoses. As clinicians have independently modified existing autism measures to assess children with VI, future challenges associated with improving the diagnostic precision of ASD in VI will be the development of specific assessment based on visual neutral tasks, detailing modifications so that findings can be replicated, and the validation of these tools on larger sample.

### **5. Conclusions**

The relationship between VI and ASD is a controversial issue and it is well expressed by the still controversial estimated prevalence of ASD among visually impaired population.

The current review suggests that some evidences can help us in understanding autistic-like behaviors in VI. ASD among visually impaired children can be a neurodevelopmental condition that occurs independently of the visual disorder. This seemed to be particularly true for those described subjects who present potential common causal factors, such as genetic defects, prematurity, pathologies that interest the central nervous system. These conditions cause a combination of blindness and brain damage, which is an important contributing factor for the development of ASD.

Autistic-like symptoms can also be secondary to the VI and related to sensory deprivation and environmental risk factors. This is typical of those children who present only severe VI or blindness, without other disorders that involve the central nervous system. In these cases, the underlying pathway of autistic-like features in VI is distinctive of that of individuals with ASD. Peculiar differences can be found, starting from the great interest in shared objects in blind, but not in ASD individuals; good structural language skills that allow for social participation and shared activities in blind, but not in ASD individuals; evidence of potential reversibility of autistic signs as a transient phenomenon in blind but not in ASD individuals.

According to Brambring [128], in these individuals, autistic-like symptoms may reflect blind-specific developmental problems in the acquisition of social-cognitive abilities rather than a psychopathological disorder. In other words, sighted autistic children and blind children may reveal similar symptoms, but for different reasons.

In visually impaired individuals who present associated problems with potential common causal agent, a detailed analysis of autistic-like symptoms is necessary, in order to avoid an overestimation of the co-occurrence of ASD.

Diagnosing ASD in VI children should be done very carefully in clinical practice and assessment tools that take into account the type and level of VI are needed. The future challenge will be to apply new tests involving alternative nonvisual tasks (e.g., based on tactile or auditory experiences) and to improve our understanding of the alternative developmental pathways and adaptive-compensatory approaches in children with VI and autistic-like symptoms.

**Author Contributions:** Conceptualization, A.M., E.L.F.; investigation, A.M., S.M.; writing—original draft preparation, A.M., S.M., A.R., J.G., L.B.M.; writing—review and editing, A.M., S.M., L.B.M., E.M.F.; supervision, F.G., L.B.M., E.M.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **Glossary**


### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Morphofunctional Alterations of the Hypothalamus and Social Behavior in Autism Spectrum Disorders**

### **Andrea Caria \*, Luciana Ciringione and Simona de Falco**

Department of Psychology and Cognitive Sciences, University of Trento, 38068 Rovereto, Italy; luciana.ciringione@unitn.it (L.C.); simona.defalco@unitn.it (S.d.F.)

**\*** Correspondence: andrea.caria@unitn.it

Received: 28 May 2020; Accepted: 3 July 2020; Published: 8 July 2020

**Abstract:** An accumulating body of evidence indicates a tight relationship between the endocrine system and abnormal social behavior. Two evolutionarily conserved hypothalamic peptides, oxytocin and arginine-vasopressin, because of their extensively documented function in supporting and regulating affiliative and socio-emotional responses, have attracted great interest for their critical implications for autism spectrum disorders (ASD). A large number of controlled trials demonstrated that exogenous oxytocin or arginine-vasopressin administration can mitigate social behavior impairment in ASD. Furthermore, there exists long-standing evidence of severe socioemotional dysfunctions after hypothalamic lesions in animals and humans. However, despite the major role of the hypothalamus for the synthesis and release of oxytocin and vasopressin, and the evident hypothalamic implication in affiliative behavior in animals and humans, a rather small number of neuroimaging studies showed an association between this region and socioemotional responses in ASD. This review aims to provide a critical synthesis of evidences linking alterations of the hypothalamus with impaired social cognition and behavior in ASD by integrating results of both anatomical and functional studies in individuals with ASD as well as in healthy carriers of oxytocin receptor (OXTR) genetic risk variant for ASD. Current findings, although limited, indicate that morphofunctional anomalies are implicated in the pathophysiology of ASD and call for further investigations aiming to elucidate anatomical and functional properties of hypothalamic nuclei underlying atypical socioemotional behavior in ASD.

**Keywords:** autism spectrum disorders; hypothalamus; amygdala; oxytocin; social cognition; social interaction; affiliative behavior; neuroimaging

### **1. Introduction**

Autism spectrum disorders (ASD) are neurodevelopmental disorders with complex and diversified pathogenesis characterized by dramatic impairment of social communication, social interaction and empathy with an estimated prevalence in the general population ranging from 1 in 100 to 1 in 54 children [1]. ASD are heterogeneous disorders with multisystem and multigenic origin, where even identical genetic variations may lead to divergent phenotypic characteristics [2]. Neuroimaging studies suggested widespread abnormalities involving distributed brain networks [3–7], but convincing evidences of systematic differences in brain network dynamics underlying the cognitive and behavioral symptoms of ASD are still lacking. On the other hand, an accumulating body of evidence indicates a tight relationship between the modulatory functions of the endocrine system and typical and atypical social behavior [8–12]. In particular, two evolutionarily conserved hypothalamic peptides, the oxytocin (OT) and arginine-vasopressin (AVP), because of their extensively documented role in supporting and regulating affiliative and socio-emotional responses [13–17] have attracted great interest for their critical implications in ASD.

Animal studies revealed that OT and AVP critically mediate social and affiliative behavior [18–20]. In addition, administration of OT has been shown to facilitate protective and nursing behavior toward pups [21]. In non-human mammals, OT is generally observed to facilitate approach behavior by decreasing avoidance of proximity and reducing defensive behavior, whereas AVP appears to modulate aggressive responses in relation to pair bonding and mating behavior, especially in males [22,23]. In humans, the effects of intranasal OT administration indicate a reduction of social stress and anxiety facilitating positive social approach and interaction, and affiliative behavior [24–27]. Moreover, intranasal AVP administration in humans, similarly to the effects observed in animals, has been shown to differentially influence social behavior in males and females, with increasing aggressive and agonistic responses in men and facilitation of pair bonding in women [28]. Several investigations also reported an association of the levels of peripheral OT and oxytocin receptor (OXTR) polymorphisms with the diagnosis and severity of ASD [29]. Genomic and epigenetic evidences for OXTR deficiency have been also observed in individuals with ASD [30]. Remarkably, a large number of controlled trials indicated that intranasal OT and AVP administration can ameliorate social abilities in autism [31–36].

Altered OT and AVP synthesis and release appear to be among the core dysfunctions underpinning the impairments in social and communication behavior of individuals with ASD [9,11,37], although it remains unclear whether OT neuropeptide can be used as biochemical marker for ASD [38].

OT and AVP are synthesized by magnocellular neurons of the supraoptic and paraventricular nuclei of the hypothalamus that secret them into the peripheral blood circulation through the posterior pituitary gland. Importantly, these peptides also act as neurotransmitters through the dendritic terminals of magnocellular neurons that release them into the hypothalamic extracellular fluid [39], and through parvocellular neurons projections to brainstem and subcortical regions, such as the amygdala, nucleus accumbens and hippocampus [40,41]. In addition, besides passive diffusion in brain circuits following dendritic release [42,43], OT transmission is also mediated by widespread long-range axonal projections of hypothalamic OT neurons [14] permitting direct modulation of the amygdala and other forebrain regions [20]. Correspondingly, OT and AVP receptors have been localized in various brain regions including the hypothalamus and the limbic system [30,44,45]. Notably, differential OT release mechanisms through dendritic and axonal terminals characterize hypothalamic activity. In fact, dendritic OT release can occur with no spiking activity, and thus, no secretion into the peripheral circulation; vice versa, electrical activity of the cell bodies can induce OT release from axon terminals without central OT release from the dendrites [46,47]. Moreover, dendritic release can lead to a very large disproportion between the concentration of OT in the extracellular fluid of the hypothalamic supraoptic nucleus and that in the periphery by over 100-fold greater [41].

Furthermore, there exists long-standing evidence of severe socioemotional impairment after hypothalamic lesions, involving in particular the ventromedial nuclei [48]. Rage has been observed after ventromedial hypothalamic lesions in both animals and humans (Wheatley 1944 [49]; Reeves & Plum 1969 [50]). Separation-induced distress vocalization can be elicited by electrical stimulation of the medial hypothalamus in guinea pigs (Herman & Panksepp 1981 [51]). Stereotactic stimulation studies in humans showed altered sexual behavior triggered by accidental focal lesions of rostromedial basal forebrain structures including the septo-hypothalamic area [52]. In addition, hypothalamic stimulation can also induce pleasurable experiences and prosocial behavior in humans [53,54]. For instance, several investigations demonstrated reduced aggressive behavior and increased social interactions after deep brain stimulation of the posteromedial hypothalamus [55].

Nonetheless, despite the unquestionable key role of the hypothalamus in the production of the OT and AVP (Swanson and Sawchenko, 1983), the severe socioemotional dysfunctions caused by hypothalamic lesions, and the apparent association between hypothalamic neuropeptides and socio-affective responses in ASD and neurotypical population (NT), hypothalamic involvement remains elusive in most of neuroimaging investigations exploring the neural correlates of normal and abnormal human socioemotional behavior [56–62]. In particular, a surprisingly limited number of studies analyzed the implication of the hypothalamus in the social impairment of individuals with ASD.

Building on the above mentioned evidences, this review aims at providing a synthesis of neuroimaging investigations reporting morphofunctional alterations of the hypothalamus in ASD through the examination of data from individuals with ASD as well as from healthy carriers of genetic risk variation in OT receptors, as several polymorphisms of OT receptor genes have been associated with modulation of socioemotional responses and ASD [63–71]. A description of MR-based anatomical studies reporting abnormal morphology of the hypothalamic region will be followed by a survey of the few existing task-based and resting state functional MRI studies reporting hypothalamic alterations in individuals with ASD and in healthy carriers of genetic risk variation in OT receptors. A critical discussion integrating anatomical and functional findings will then attempt to provide some interpretations of the possible role of the hypothalamus, and its functional exchanges with cortical and subcortical networks, in the atypical socioemotional responses of ASD individuals. In conclusion, some fundamental open questions aiming at elucidating the morphological and functional hypothalamic anomalies and their impact on social cognition and behavior in ASD will be proposed.

### **2. Literature Search**

This review is based on a Pubmed and Scopus search aiming to comparatively analyze the current literature until April 2020 using the following keywords "autism" AND "hypothalamus" AND "social." In total, 236 papers were obtained from Scopus, whereas only 22 papers from Pubmed. After refining the search by limiting articles that included the term "MRI," 42 documents remained in Scopus and just one in Pubmed. The remaining publications were then further screened for articles reporting original research studies. Careful inspection of papers, aiming to identify anatomical and functional investigations related to ASD, led to additional rejections of few unrelated papers as well as inclusion of some others missing in the initial literature search, and surprisingly resulted in only 10 relevant scientific publications for our qualitative analysis.

### *2.1. Structural MRI Studies*

One of the first direct evidence linking anatomical alterations of the hypothalamus with ASD was provided by a study assessing structural MRI based measures of brain morphometry in children and adolescents with ASD (*n* = 52) [72]. ASD individuals with respect to typically developing controls showed significant decrease of gray matter (GM) volume in the hypothalamic region including the supraoptic and paraventricular nuclei, independently of age, IQ or gender. No differences were observed in global volumes of GM, white matter and cerebrospinal fluid.

In another study, hypothalamic atrophy was measured in young male adults with ASD (*n* = 10) with respect to neurotypical participants using two complementary structural analysis approaches [73]. First, an ROI-based voxel-based morphometry (VBM) analysis applied to the hypothalamic region, delineated through manual segmentation and including voxels in the third ventricular space between the left and right hypothalamus, revealed reduced GM density of the hypothalamus and increased cerebrospinal fluid density in the third ventricle proximal to paraventricular nucleus. Second, an automatic method was applied to a larger cohort of male ASD individuals (*n* = 41) to estimate ventricular volume of the third ventricle. This method aimed to indirectly validate previous results on the basis of the assumption that relative increase of third ventricle would imply volume reduction of the adjoining tissues. This analysis demonstrated an increase of third ventricle volume that was independent of the lateral ventricles (used as covariate), and thus excluded global brain volume increase.

Recently, decreased volume in the bilateral hypothalamus along with increased volume in the left amygdala and left hippocampus was observed in young children with ASD (*n* = 14, mean age = 4.5) compared to typically developing children (*n* = 14, mean age = 4.1) [74]. In addition, the authors observed that the hypothalamic volume was positively correlated with plasma AVP concentration.

In parallel, several indirect evidences of abnormal hypothalamic structure and function in ASD emerged from studies of healthy OXTR risk allele carriers, in particular with the OXTR variant rs53576 that appears to be associated with lowered socioemotional responses [63,75] and is often observed in individuals with ASD [76–80].

One of the first demonstrations in this direction was a multimodal neuroimaging genetics approach that permitted to identify several neural alterations in a large sample (*n* = 212) of healthy Caucasian OXTR risk allele carriers [64]. Tost et al., using VBM, revealed a significant decrease of hypothalamic GM volume in rs53576 risk allele carriers that correlated with the degree of allele risk. Notably, decreased hypothalamic volume was predictive of a lower prosocial temperament trait in males. Structural correlation analysis, information that has been shown to mirror anatomical connectivity, showed allele-dependent increase of coupling between the hypothalamus and higher-order limbic processing areas, such as the dorsal anterior cingulate cortex, including the paracingulate cortex and amygdala (encompassing high density OT receptors), in rs53576A allele carriers.

In a consecutive study using VBM methods, reduction of GM volume in the dorsal anterior cingulate gyrus and hypothalamus was also associated in carriers of OXTR rs2254298A, another identified genetic risk variant for ASD; this result was mainly related to male carriers [81]. Structural covariance analysis revealed a significant increase in the structural connectivity between hypothalamus and dACG in rs2254298A carriers, similar to that observed in rs53576A carriers. The observed increase of anatomical coupling in healthy carriers of genetic risk variants for ASD may suggest abnormal connectivity related to alterations of several white matter morphological properties as well as atypical functional interactions [82,83].

Additional studies examining brain morphology in individuals with single nucleotide polymorphisms in the OXTR gene related to ASD indicated other alterations of locale brain volumes including the hypothalamus. Inoue et al. [84], adopting a manual tracing methodology for measuring regional brain volume, observed larger bilateral amygdala volume in Japanese adult carriers of OXTR rs2254298A, proportional to the dose of this allele. No significant association of this genotype was instead observed with hypothalamus as well as with global brain volume. In a subsequent analysis on the same data using VBM, stimulated by result of Tost et al. (2011), the same authors reported that rs2254298A was also associated with reduced GM volume in the dACG but not in the hypothalamus and amygdala [85]. However, they observed an interaction effect between gender and rs2254298A genotype in the right hypothalamus, reflecting smaller right hypothalamus volume in females only.

### *2.2. Functional MRI Studies*

Aoki et al. in a focused metanalysis of 13 fMRI studies in ASD individuals during emotional-face processing (considering both emotional-face vs non-emotional-face and emotional-face vs non-face contrasts) observed abnormal functioning of several subcortical regions [86] among which hypothalamic hypoactivity was prominent. In particular, individuals with ASD (*n* = 226, age ranging from 9 to 37 years) in comparison to NT controls (*n* = 251, age ranging from 9.2 to 28.6) showed significant hypoactivation of the hypothalamus, and hyperactivation of the bilateral thalamus, bilateral caudate, left cingulate and right precuneus. The comparison of emotional-face to non-face conditions showed a similar activation pattern but hypoactivity was also observed in the parahippocampal gyrus and amygdala, in addition to the hypothalamus. In line with behavioral studies demonstrating impaired emotional-face processing in ASD [56], the observed alteration of subcortical rather than cortical regions during face perception suggested dysfunctional unconscious processes in relation to social cognition. Notably, reduced hypothalamic activity was not observed in each of single studies included in the metanalysis, possibly because of their limited statistical power [87].

Preliminary evidence of a direct association between hypothalamic dysfunction and social interaction was also shown by Chaminade et al. that measured fMRI-based brain responses in ASD individuals (*n* = 10, mean age 21) during a more realistic and entertaining social behavior consisting of an interactive videogame of the popular "stone-paper-scissors" game [88]. ASD and NT participants played against three different agents: a human being, a humanoid robot endowed with artificial intelligence attempting to win the games by considering previous games' results, and a computer that randomly generated the three possible responses. A significant interaction effect between Agent (Human, Robot) and Group (ASD, NT) delineated an activation cluster in the left and right hypothalamus, attributed to the paraventricular nucleus, resulting from decreased activity when ASD participants played against the human as compared to the artificial agent, with respect to NT. In addition, functional connectivity analysis of the left hypothalamus revealed a single cluster in the left temporoparietal junction resulting from the interaction effect of Group and Agent. Specifically, a significant negative coupling between the left hypothalamus and left temporoparietal junction (lTPJ) was measured when NT played against the robot and when ASD participants played against the human. Moreover, the coupling observed when ASD participants played against the human, but not against the robot and computer, was negatively correlated with the severity of autistic symptoms measured with Autistic Spectrum Quotient [89]. Interestingly, the decreased modulation of hypothalamic nuclei activity along with negative functional connectivity between hypothalamus and lTPJ, a region associated with anthropomorphization—which is the tendency to attribute human traits to artificial agents—was observed when ASD individuals interacted with a human player, and similarly when NT individuals played against the robot. The anticorrelation between lTPJ and hypothalamus might reflect inhibitory activity exerted by the lTPJ on hypothalamic nuclei that would then result in reduced social motivation and reward for human interactions in ASD.

In line with hypothalamic functional alterations in ASD during processing of emotional expressions [86], reduced hypothalamic activation was also observed in adult carriers of risk OXTR gene mutation for autism [64,81]. Tost et al. (2010), besides abnormal anatomy of the hypothalamus, reported functional alteration of hypothalamic activity during perception of facial expressions (using a Face-Matching Task). In particular, they observed increased fMRI-based connectivity (measured with cross correlation) between hypothalamus and amygdala, and decreased amygdala activation in adult carriers of rs53576A (*n* = 228) with respect to individuals with the GG genotype [64]. In a subsequent analysis the same authors observed reduced deactivation of the dorsal anterior cingulate and paracingulate cortex associated with healthy carriers of another OXTR gene polymorphism, the rs2254298A [81]. Moreover, differential functional brain connectivity was revealed by genotype-by-sex interaction effect associated with negative coupling of the hypothalamus with dACG and amygdala in male rs2254298A carriers, and positive coupling in females.

Likewise, Wang et al. (2013) demonstrated gender dependent effects of OXTR rs53576 gene variation on hypothalamic functional connectivity in healthy individuals. Specifically, whole brain analyses of local functional connectivity density (FCD) during resting-state fMRI data (*n* = 270) revealed a main effect of genotype on the local FCD in the hypothalamus and no gender-by-genotype interaction effect, although local FCD in male AA homozygotes was significantly lower than in male G-allele carriers [90]. Additional analysis of gender-by-genotype interaction considering resting-state functional connectivity of the hypothalamic region only showed significantly weaker coupling between the hypothalamic region and the left dorsolateral prefrontal cortex in male AA homozygotes with respect to male G-allele carriers.

### **3. Discussion**

Building on the well-recognized role of the hypothalamus in the production of the OT and AVP, and the emerging evidences of an association between activity of hypothalamic neuropeptides and socioaffective responses in ASD and NT population, we here aimed to inspect the current neuroimaging literature in humans in search for evidences of hypothalamic alterations in relation to the core social deficits in ASD. Examination of current structural and functional MRI studies reporting alterations of the hypothalamus in ASD, although rather limited, revealed quite consistent morphofunctional abnormalities. Specifically, two main findings emerged from VBM and fMRI analyses, in both adults and children: anatomical hypothalamic atrophy and functional hypoactivation during face processing and social interaction, respectively.

### *3.1. Hypothalamic Morphological Alterations*

Anatomical hypothalamic atrophy was mainly related to smaller hypothalamic volume in both ASD individuals [72] and healthy carriers of OXTR genetic risk variant for ASD, and to reduced GM density observed in ASD [73]. Notably, in line with gender-dependent differences in the expression of the OXTR gene [91,92], hypothalamic structural abnormalities in healthy carriers of OXTR genetic risk variant for ASD appear not equivalent in males and females and dependent on OXTR variants [64,85].

Sexual dimorphisms of the hypothalamus might follow similar gender-related differences observed in other brain regions including the amygdala, as well as in interhemispheric connectivity, along with differences in hormone-related personal traits, cognition, behavior and psychiatric disorders manifestation [93], ultimately mirroring ASD prevalence that appears larger in males than in females with a male-to-female ratio closer to 3:1 [94].

The observed anatomical abnormality of the hypothalamus is in line with several neuroimaging observations that, although not always congruently, reported morphological changes in ASD in multiple brain regions [95,96], including reduced volume in the social brain network [97–100].

However, it remains difficult to infer the exact neuronal mechanisms leading to hypothalamic atrophy, since variations of multiple properties of GM can equally affect VBM signal. Changes at the level of neuronal cell bodies, glia or neuropils might all contribute to hypothalamic grey matter reduction and differentially affect regulation of central neuropeptides and peripheral hormonal regulation through abnormal synthesis and release. Indeed, postmortem brain analysis in ASD highlighted various anatomical anomalies related to neuronal density and size, dendritic spine density, glia and cerebral vasculature [101]. In particular, lower neuronal density has been measured in human brain regions involved in social behavior such as the fusiform gyrus and amygdala [102–104], as well as in specific layers of ACC [105], possibly reflecting specific hypoactivation of these same regions in ASD.

Moreover, some insights about neural mechanisms underlying hypothalamic atrophy might arise from animal models of ASD. Genetically modified animal models such as the Black and Tan Brachyury (BTBR) mouse model [106,107] and the copy number variants mouse model simulating the 15q11-13 duplication in human (15q dup) [108] were also associated with decreased GM volume of the hypothalamus. In addition, mice carriers of neurexin gene mutations have been associated with fewer oxytocin-expressing neurons in the hypothalamic paraventricular nucleus [109]. Similarly, mice with missense heterozygous mutation in the contactin-associated protein-like 2 (CNTNAP2) were characterized by specific reduction in the number of OT expressing cells in the paraventricular nucleus in association with reduced OT concentrations in brain extracts [110]. In humans, reduced plasma concentration of OT has been indeed measured in ASD [111] and predicted social impairment [29], but no clear evidences of alterations at central level emerged. Some indications suggest a possible correlation between plasma and CNS OT concentrations, but this correspondence seems particularly dependent upon the assessing methods employed [112]. Thus, there are currently no demonstrations of the specific impact of hypothalamic atrophy on OT transmission to brain circuits in humans.

### *3.2. Hypothalamic Functional Alterations*

FMRI studies in ASD revealed hypoactivation of the hypothalamus in relation to face processing, and during interactive play with humans. Likewise, reduced hypothalamic activity during face perception was also observed in healthy carriers of risk genetic mutations for ASD [64].

As for morphometric anomalies, no direct interpretation of the neuronal processes underlying hypothalamic fMRI hypoactivation is yet possible. Decreased BOLD response does not necessarily imply reduced OT/AVP release. Although dendritic and axonal neuropeptides release is generally enhanced by increased action potential frequency, the BOLD signal neither directly nor exclusively reflects neuronal spiking activity but correlates more strongly with local field potentials, which represent postsynaptic activity and integrative soma-dendritic processes [113]. Considering the observed possible uncoupling between hypothalamic spiking activity and dendritic oxytocin release, which can be locally mediated by intracellular calcium stores independently of action potentials [114], decreased BOLD signal in the hypothalamus might indeed reflect reduced dendritic release of oxytocin.

Animal models of ASD indicate some convergent evidences in relation to the hypothalamic activity. For instance, 15q dup mice showed no hypothalamic activation in response to odor stimulation and resting-state functional hypoconnectivity in a widespread brain network including the hypothalamus [115]. Conversely, hypothalamic activity positively correlated with measures of typical social behavior in rats not responding to exposure to valproic acid in utero, another animal model of ASD [116].

In humans, hypoactivation of the hypothalamus and reduced oxytocin secretion has been observed in eating disorders [117,118]. However, to date, a clear demonstration of any relationship between hypothalamic activity and OT release at central and peripheral level in humans is still lacking. Furthermore, the limited spatial resolution of the considered studies does not permit to correctly attribute hypoactivation to single hypothalamic nuclei, all having diverse functions in the autonomic and central nervous system.

Reduced activation of the hypothalamus was also frequently associated with decreased amygdala activity in both ASD and carriers of OXTR rs53576A allele, in particular during face processing [64,86]. These findings are consistent with several previous studies reporting decreased amygdala activation during face perception in ASD [119,120]; nevertheless, opposite findings were also reported, but they were supposedly ascribable to longer gaze fixation time and higher anxiety level of individuals with ASD [58,121,122].

Hypothalamic nuclei can be controlled directly by the amygdala through the amygdalofugal pathway and the stria terminalis, and indirectly through the bed nucleus of the stria terminalis, which mediates stimulation of the hypothalamic-pituitary-adrenal axis. Projections of the central amygdala to the hypothalamus and brainstem can directly trigger autonomic fear responses [123]. Stimulation of amygdaloid OXT receptors is then assumed to inhibit these efferences' activity so as to decrease aversive responses to socially-relevant stimuli [20,124,125], which would be increased in case of diminished amygdala activation. Accordingly, increased hypothalamic activity and amygdala deactivation appear to mediate initiation and consolidation of social relationship in healthy individuals. Such a reverse activation pattern of the hypothalamus and amygdala has been associated with several social behaviors such as other people's trust and trustworthiness [126] and mother–infant and pair bonding [127–130], as well as visual processing of personally known faces including same-sex sibling and best friend with respect to unknown faces [131].

Prosocial behavior can be enhanced by hypothalamic through the modulation of two complementary responses: enhancement of social stimuli saliency processing and reduction of fear and avoidance behavior, both mechanisms being strictly dependent on amygdala activity [132,133]. Notably, AVP and OT have opposite modulatory effects on fear and anxiety-related behavior: the former by enhancing sympathetic responses such as stress level, anxiety, aggressiveness and boosting fear memory consolidation, the latter by acting on complementary parasympathetic responses that facilitate prosocial attitude and interactions as well as extinction of conditioned avoidance responses. These opposite regulatory neurophysiological processes result from activation of distinct elements of an inhibitory network within the medial part of the amygdala, and consecutive integration of different afferences to the central amygdala into a modulatory output to the hypothalamus and brainstem for appropriate anxiety and fear responses [134].

In addition, the hypothalamus can significantly influence socioemotional responses through a complex network that includes widely distributed, mostly bi-directional, neural connections to other brain regions. The hypothalamus is interconnected with basal forebrain areas such as the periamygdaloid region and the septal nuclei and other brainstem nuclei through the medial forebrain bundle, which mediates top-down modulation of both somatic and visceral activity by the forebrain and limbic system, as well as bottom-up influences of higher brain activity by internal organs and bodily interoceptive signals.

Previous studies reporting hypothalamic activation concurrent to other socioemotional-related brain regions indeed indicated widespread interactions of the hypothalamus with emotional, motivational and social brain centers [132,135–138]. However, the still scarce evidence of functional connectivity of the hypothalamus in both NT and ASD individuals, which might also be partly dependent of the variable association between hypothalamic spiking activity and oxytocin release, do not permit to clarify how this region interacts during typical and atypical socioemotional behavior.

Indirect indications about alterations of functional connectivity emerge from studies on healthy carriers of risk genetic mutations for ASD. Tost et al. (2010) measured increased structural and functional connectivity during face perception between hypothalamus and amygdala in OXTR risk allele carriers, suggesting a dysfunctional coupling underlying inappropriate responses to socially-relevant stimuli, although the actual nature of their interactions remains unknown. In addition, the same authors observed a negative coupling between hypothalamus and dorsal anterior cingulate and paracingulate cortex resulting from respectively decreased and increased activity [64]. Direct projections of the anterior cingulate cortex (ACC) to the hypothalamus have been demonstrated in both animals and humans [139,140]. Interestingly, concurrent increased activity in the paracingulate cortex and in the septal area, including the hypothalamus, has been associated with unconditional trust towards other people [126]. Maladaptive changes in trusting behavior, for instance after repeated violations of trust, consequent to exogenous administration of oxytocin, have been associated with increased ACC activity and decreased amygdala and midbrain activation [141]. The anticorrelation between the hypothalamus and ACC might then result from exaggerated ACC inhibitory activity of the hypothalamic nuclei preventing adaptive social behavior. The ACC is an important regulatory center that, through direct projections to the amygdala, insula, ventral striatum, hypothalamus and brainstem [140] can control socioemotional responses. Remarkably, it has been proposed that the medial prefrontal cortex, including ACC, would encode abstract representation of social experiences [142] permitting to predict and guide social goal-directed behavior based on social prediction error [143,144]. In line with this assumption, ACC connections with brain regions related to emotion and reward such as OFC, ventral and dorsal striatum, amygdala, insula and hypothalamus would then support generation of active inferences of affective, interoceptive and reward values [145,146] of socioemotional responses, as well as the minimization of prediction error between expected and actual behavioral outcome, the latter seemingly compromised in ASD [9,147].

### *3.3. Relevance of Hypothalamic Alterations in Healthy Carriers of Genetic Risk Variation in OT Receptors*

Structural and functional alterations of the hypothalamus in individuals with polymorphisms of the OXTR gene are intriguing considering the increasing evidence indicating their relationship with ASD [79]. For instance, the OXTR rs53576A and recently the rs2268498 were associated with ASDs in both Asian and Caucasian populations [76,77,148,149]. Despite some inconsistency of the studies linking OXTR rs53576 variant with impaired socioemotional traits and behavior [150], the rs53576 and rs2254298 OXTR single nucleotide polymorphisms were shown to correlate with increased severity of social deficits in ASD, and less with social deficit in ADHD, thus indicating a differential relationship between this neuropeptide receptor gene allele and the social phenotype [80]. A metanalysis on the relationship between the OXTR rs53576 variant and human sociality indicated a clear influence of this OXTR polymorphism on individual psychological traits related to social responses to other people (for instance extraversion, empathy, and social loneliness) [151]. In short, neuroimaging findings in healthy carriers of OXTR rs53576A and OXTR rs2254298A genotypes indicate that alterations in the expression, and possibly function, of OXTR gene might be related to abnormal morphofunctional characteristics of the hypothalamus in ASD. However, it is conceivable that other OT signaling genes, such as the structural gene for OT (OT/neurophysin-I) [152] and gene for OT secretion (CD38) [153] that along with the OXTR have been frequently linked to human social behavior [154], might also contribute to structural and functional brain alterations in ASD.

### **4. Conclusions**

The few studies that have thus far observed, directly or indirectly, a relationship between the hypothalamus and ASD indicate both structural and functional alterations. However, considering the paucity of current investigations, further well-defined studies are strongly needed to clarify morphological and functional properties of hypothalamic nuclei and their complex functional exchanges with cortical and subcortical networks during socioemotional behavior in ASD (Figure 1).


**Figure 1.** Questions for future research.

Current neurophysiological investigations on the role of the hypothalamus in typical and atypical human social behavior have been likely hindered by several limitations related to the experimental methodology and MR signal acquisition techniques, resulting in a surprising disregard of its essential contribution. Designing protocols that permit to assess neural activity during realistic and entraining social scenarios, with rigorous control of experimental variables, for both NT and ASD individuals, is particularly challenging. In addition, MRI acquisition schemes adopted in most of previous anatomical and functional brain investigations in ASD were not specifically tailored for the hypothalamus. Neuroimaging of the small hypothalamic nuclei is certainly arduous as needs very high spatial resolution to clearly delineate their functional subdivisions and at the same time it requires prevention of potential partial volume effects, compensation for signal-dropouts occurring in ventromedial subcortical regions and correction for distortions generated by neighboring ventricles and blood vessels. Nevertheless, extraordinary progresses in high-field and ultra-high-field MRI techniques indicate feasibility of high-resolution structural [155,156] and functional [157,158] imaging of the human hypothalamus, and might then valuably support the elucidation of morphological and functional properties of this region in typical and atypical socioemotional behavior. Ultimately, greater understanding of the human hypothalamic morphology and functions is essential not only for the comprehension of socioemotional behavior but also in relation to the direct implication of hypothalamic neuropeptides in synaptic activity and plasticity [37], and neurogenesis [159], that may considerably impact the still obscure pathophysiology of ASD.

**Author Contributions:** A.C. conceived the study, reviewed the literature and wrote the paper; L.C. reviewed the literature and wrote the paper; S.d.F. wrote the paper. All authors have read and agreed to the published version of the manuscript

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Theory of Mind Deficits and Neurophysiological Operations in Autism Spectrum Disorders: A Review**

### **Maria Andreou \* and Vasileia Skrimpa**

Department of English, School of Arts and Humanities, University of Cologne, 50923 Cologne, Germany; vskrimp1@uni-koeln.de

**\*** Correspondence: mandreou@uni-koeln.de

Received: 1 June 2020; Accepted: 18 June 2020; Published: 20 June 2020

**Abstract:** Theory of Mind (ToM) is a multifaceted skill set which encompasses a variety of cognitive and neurobiological aspects. ToM deficits have long been regarded as one of the most disabling features in individuals with Autism Spectrum Disorder. One of the theories that attempts to account for these impairments is that of "broken mirror neurons". The aim of this review is to present the most recent available studies with respect to the connection between the function of mirror neurons in individuals with ASD and ToM-reflecting sensorimotor, social and attentional stimuli. The majority of these studies approach the theory of broken mirror neurons critically. Only studies from the last 15 years have been taken into consideration. Findings from electroencephalography (EEG) studies so far indicate that further research is necessary to shed more light on the mechanisms underlying the connection(s) between ToM and neurophysiological operations.

**Keywords:** EEG; autism; theory of mind; adults and adolescents

### **1. Introduction**

Autism Spectrum Disorder (ASD) is a complex pervasive neurodevelopmental disorder, presenting great heterogeneity with respect to symptomatology and traits. With regards to the degree of severity, ASD reveals impairments in many domains such as social interaction, verbal and non-verbal communication, and restricted and repetitive behaviours [1]. Regarding cognitive and social abilities, it has been shown that individuals with ASD [2] present great variability. The spectrum roughly ranges from high-functioning autism to low-functioning autism associated with learning impairments and disabilities [3].

In most of the cases, ASD is connected with mental disability, difficulties in movement coordination, attention deficits, sleep disturbances and gastrointestinal disorders [4]. However, it is not uncommon for some individuals within the spectrum to achieve high performance skills in visual abilities, music, art and mathematics [5]. Research so far has shown that the appearance of the disorder is estimated at 1–2%, is 4.5 times more frequent in males than in females and could emerge in all national and socio-economic strata [6]. It should be noted that impairment in social skills is one of the fundamental characteristics of the disorder, accompanying the individual throughout his or her lifespan [7].

As advancements in cognitive neurosciences have drawn attention to neurobiological features of ASD, there is a great need to understand the disorder mechanisms. Various theories have been proposed; however, the most prominent to consider for the majority of the social dysfunction traits has been Theory of Mind (ToM), which relates to the ability of individuals to evaluate the behaviour of others on the basis of their own mental states, such as goals, feelings and beliefs [8] and enables the identification of others' intentions, emotions, as well as self-awareness [9].

### **2. Theory of Mind**

ToM is the ability to interpret the mental states of oneself and others [10] and allows individuals to make considerations as well as reasonable explanations regarding the behavioural patterns of others [11]. However, in the case of individuals with ASD, asymmetry between their own knowledge and that of others is often detected [12]. This is why poor performance in ToM tasks is observed in individuals with ASD [13].

Although ToM is unfounded as an exclusive explanation for the characteristics of ASD, the influence of ToM on social skills is paramount [14]. Individuals with ASD show impairments in the reasoning of intentions and emotions that highlight social conventions [15]. The performance of children with ASD in advanced ToM tasks correlates with their social competence; however, the practice of ToM skills in everyday life is often diminished [16]. It is therefore evident that, in spite of the ability of some children with ASD to generate thoughts, beliefs and intentions in ToM tasks, they are unable to implement these skills in social situations [17,18].

Impairments in ToM abilities in children with ASD lead to social, behavioural and communication deficits, as well as discrepancies in social interaction, due to the inability of individuals with ASD to perceive that behaviour is driven by mental states [19,20]. Social dysfunction can be therefore attributed to the delayed or incomplete acquisition of ToM in ASD; however, individuals within the spectrum show individual differences with regards to the acquisition of those skills. In fact, children with ASD who succeed in ToM tasks are considered to be better socially integrated compared to their ASD peers who fail in those tasks [21,22].

Furthermore, factors in ASD such as social and communication experiences, interaction with parents, inability to process information, weak central coherence and lack of motivation, as well as perception problems prohibit the development of ToM [15,22]. Spontaneity in relation to ToM stimuli and reciprocal socio-psychological cues is totally absent in individuals with ASD, even in the case of high-functioning autism. That being said, individuals with ASD exhibit significant deficits in the process of basic emotion recognition judged from information acquired just from the eye gaze of other people. High-functioning individuals within the spectrum are, however, able to interpret mental states based on the whole facial expression [23]. In all cases of ASD adults, though, there is a lack of spontaneous capacity to attribute mental states.

### **3. Mirror Neurons–Mu Suppression**

### *3.1. Mirror Neurons and ToM*

The "broken mirror neurons" theory has received attention in literature with respect to possible connections between autistic traits and discrepancies in the function of mirror neurons (MN); it is hypothesised to constitute one of the factors responsible for ToM attenuation in individuals with autistic traits, and is linked to the interpretative neurocognitive theory of social and communication impairments in ASD [24].

MN delineate a functional set of neurons located in the cerebral cortex, activated both during the performance of an action, as well as during the observation of the performed action [25]. They were designated as such due to their ability of mirroring behavioural patterns, enabling the observers to encode the intentions behind the observed action sequences and to be in a position to further imitate them [26]. This set of neurons is mainly found in the inferior frontal cortex, the premotor cortex, the supplementary motor area, the primary somatosensory cortex and the inferior parietal cortex [27] and is hypothesised to be directly related to social abilities and skills in primates and humans, including imitation, empathy, ToM and language development [28–30]. Due to the fact that individuals with ASD demonstrate impairments in all the aforementioned domains, it is suggested that the system of MN is dysfunctional in ASD [31,32].

The mechanism underlying the activation of MN is strongly linked to imitation ability and imitation-based learning [33], more precisely the imitation of gestural movements and facial expressions [34]. The inferior frontal cortex and the ventral premotor cortex play a compelling role in the action of facial imitation and mimicry, which is essential for empathy to emerge on a neurobiological level [34,35] and mirrors the synchronous coupling of behavioural and emotional development through non-verbal communication [36]. The inferior frontal cortex has a specific significance in the process of defining intentions or goals by delegating those intentions to representations of internal states, as well as to the transmission and perception of emotional states that are connected to the imitation of facial expressions [37].

Imitation processes depend on the perception of action of the sensorimotor system. The prerequisite of these processes is the elicitation of imitation through external motor stimuli that are identified and executed as action that initiates imitation as response to those stimuli [38]. One of the theories that attempt to provide an explanation for the initiation of imitation is the ideomotor theory of action, which suggests that it is not the motor property of action that triggers a reaction, but rather the goal and intention that defines this action. Iacoboni (2009) mentions that "the coactivation of the intended goal and the motor plan required to achieve it—according to the ideomotor framework—is the result of our experience. We have learned the effects of our own actions, and we expect certain effects when we perform certain acts. This previous learning makes it possible that just thinking about the intended goal automatically activates the representation of the action necessary to obtain it." [39] (p. 655).

MN are theorised to be in the centre of the process of perceiving the intentions behind an act, which further facilitates the emergence and establishment of empathy [40], and plays a significant role for the foundation of common objectives and motives among individuals [41]. Dysfunction in the system of MN in ASD has an impact on the comprehension of action and intention. In particular, individuals with ASD present deficits in perceiving the motor action and the reasoning of an action [42]. It is hypothesised that MN are the substratum of human cognition and social understanding and that their operation facilitates the process of accessing and perceiving the emotional state of others as the result of one being able to reflect one's own individual internal states and experiences [41,43–46].

### *3.2. Mu Suppression in Literature*

A method to investigate the activation of MN in humans is through measuring *mu* suppression. *Mu* is a type of rhythm that can be described as the frequency band 8–13 Hz and can be detected in an EEG test. The modulations of the power of the *mu* frequency band can provide evidence for the specific functionality of the MN in terms of the comparison between an active condition and a baseline condition [47,48]. It is still under question whether the suppression of the *mu* rhythm is a sufficient method for measuring the operational activation of the Mirror Neuron System (MNS), mostly due to small sample sizes in the research studies (especially when examining atypical populations, such as ASD) or the fact that it is mainly the power modulations of the central electrodes that are taken into consideration [49].

Despite the fact that the theory of broken MN in individuals with ASD has attracted attention, it has also created a debate with regards to its plausibility and application. The hypothesis of dysfunctions in MN accounts for deficits in ToM and imitation, but nevertheless, literature has critically approached the theory, suggesting that the aberrant operation of the system of MN does not provide efficient reasoning for the aforementioned deficits, but that it is sensorimotor impairments in ASD that have an impact on the interpretation of actions. This theory derives from the observation of animal behavioural patterns that indicate the ability of comprehension of action without being in a position to reproduce or imitate it [50]. As a consequence, only a small body of literature has investigated abnormalities in mirror neuron functioning in individuals with ASD, especially in the cases of adolescents and adults. The majority of the studies using brain activity screening techniques focus on the activation of MN in very young populations (children or infants), and their findings demonstrate impairments in the function of these neurons during ToM and imitation tasks [51,52] or gestural movements [53,54]. Although there is an adequate body of literature investigating the involvement of MS in the performance and observation

of mentalising tasks in neurotypical adults [55,56], the studies observing this performance in adult individuals with ASD are limited.

Cole et al. [57] examined the activation of MN in young adults with ASD during intention mentalising tasks, in order to detect a possible link between aberrant mirror neuron activity and autistic traits. They recruited 43 participants that matched in terms of age and verbal IQ level, dividing them into three groups according to their level of autistic traits as evaluated by the Autistic Spectrum Quotient (AQ; Baron-Cohen et al., 2001): low AQ (n = 15, mean age = 23.40), high AQ (n = 15, mean age = 24.13) and ASD (n = 13, mean age = 28.30). The participants were required to watch short videos demonstrating motor actions performed by actors that were divided into two categories: a mentalising task and a non-mentalising task. After the end of the videos, they had the task of deciding upon either the intention of the performed action (mentalising) or its success (non-mentalising). The data derived from an EEG screening test performed during the viewing of the tasks, in combination with an eye-tracking test and a Transcranial Magnetic Simulation (TMS)-induced electromyography (EMG). Their EEG findings demonstrated a lower level of *mu* suppression in the right hemisphere in participants within the group with high autistic traits during the mentalising task, which did not, however, correlate with the quality in mentalising performance. On the other hand, a lower performance in the mentalising task positively correlated with a poorer activation of MN in the left hemisphere; nevertheless, this was not linked to the level of autistic traits of the participants. Data derived from TMS revealed no variation between the groups in terms of the activation of MN and its link to performance in mentalising tasks. The hierarchical categorisation of autistic traits was a predictive factor for *mu* suppression in the 8–10 Hz band for the mentalising task and therefore for poorer mirror neuron firing in the right hemisphere. During the non-mentalising task, however, no low level *mu* suppression was detected. The authors attribute the poorer activation of MN in the right hemisphere in individuals with ASD to an abnormal connectivity among MN and the process of mentalising intentions deriving from actions, which is in accordance with the theory of impaired ToM in ASD.

The observation or mentalising of movement, as well as the imitation or execution of the movement, has been associated with the suppression of the *mu* rhythm [58] and has attracted the interest of research, so as to disentangle the relations that underlie the deficits in imitation and reduced *mu* suppression in ASD. In an earlier study, Bernier et al. [59] conducted a study aiming to investigate this link, hypothesising that individuals with ASD will present an impaired imitation ability in correlation with a low suppression of the *mu* wave. They conducted an EEG imitation experiment examining 14 male adults diagnosed with ASD and 15 neurotypical controls, matched in age (18–44 years), gender and intelligence quotient. The experiment included four condition states: (a) resting state, where the participants were required to just sit still, positioning their hands on their lap, (b) observation state, where the participants had to observe a person grasping a manipulandum, (c) execution state, where the participants were instructed to grasp the manipulandum in the exact same way they saw the person doing, and (d) imitation state, where the participants were required to imitate the instructor grasping the manipulandum (experiment adapted from Muthukumaraswamy et al. [60]). The findings with regards to the resting and execution state conditions did not show differences in *mu* suppression between the two groups (reduced in the resting state and increased in the execution state). Nevertheless, the ASD group demonstrated a significantly reduced *mu* suppression in the observation state condition in contrast to the neurotypical controls, which further supported the hypothesis of an impaired execution/observation system in ASD and therefore identified deficits in imitation abilities. The authors, however, observed that this system could not be totally impaired, since individuals with ASD do not entirely lack the ability to imitate but rather demonstrate poor imitation performance. These findings are in alignment with discrepancies in ToM abilities and attenuations in social integration. The study concluded that the execution and imitation of human movement was connected with impairments in *mu* suppression in ASD, further implying a possible involvement of a dysfunctional MS.

Fan et al.'s [61] study dealt with the index of *mu* suppression in relation to the observation/imitation mechanisms, with the intention to challenge the theory of "broken mirror neurons" in ASD. They conducted an EEG study focusing on *mu* suppression as a factor to measure resonance in the sensorimotor system during the observation and imitation of gestural movements. The researchers recruited 20 male adolescents and young adults with ASD (11–26 years) and 20 neurotypical individuals matching the ASD group in terms of age and intellectual abilities. The experiment included an eye-tracking recording and an EEG recording during the conducting of a test containing four conditions: *baseline* condition (observation of a static object on a screen), *hand* condition (observation of a video-recorded gesturing action), *dot* condition (observation of a video with a white dot), *execution* condition (manipulation of an object in the same manner as in the *hand* condition). Their findings constitute strong evidence against the theory of broken mirror neurons in ASD. More in particular, the *mu* suppression occurring from the EEG monitoring under the conditions of observation and imitation of a gestural action did not show significant variation among the two groups. Predominantly, the results did not reveal any correlation between imitation performance and *mu* suppression, contradicting the most up-to-date research findings [32]. The activation of MN in the ASD participants was evaluated as being preserved, and the *mu* rhythm was very close to that of neurotypical individuals, despite the fact that the performance in the imitation condition was significantly lower in the ASD group. The findings also reveal a relation between attenuated communication capacities and a weak *mu* rhythm, indicating a variation in the symptomatology of ASD. Age progression was not found to influence the results in either the ASD or the control group.

When dealing with impaired ToM, empathy is one of the most prominent attenuated social cues, and it is a hallmark of ToM deficits and social discrepancies in ASD. The study of Fan et al. [62] aimed at investigating the empathic and social understanding abilities of neurotypical individuals and individuals with ASD, in order to disentangle the different factors that contribute to the variances with regards to empathic arousal and the perception of social cues in ASD. Their participants consisted of 24 ASD and 21 controls who participated in an fMRI experiment evaluating pain empathy, and 20 adolescents and young adults (16–29 years of age) and 20 age-matched neurotypical controls who participated in an EEG/ERP experiment. A set of 48 images illustrating injured and non-injured body parts were presented, distinguishing between intentional and unintentional injuries as well as individual pain vs. dyad pain situations; these had to be evaluated by the participants with respect to the level of pain. The results of the study demonstrated lower pain thresholds detected in individuals with ASD in comparison to their neurotypical peers, who presented increased hemodynamic responses in SI/SII, stronger N2 but weak responses in the anterior mid-cingulate and anterior insula, and preserved LPP in the view of unintentional body harm, whereas in the case of intentional injuring, they presented reduced LPP and decreased hemodynamic responses in the medial prefrontal cortex. *Mu* suppression and MN activation in view of injuries appeared to be preserved in ASD individuals, similar to the control group, which in combination with an elevated hemodynamic response in the area of the amygdala and higher PPT indicated that individuals with ASD evaluated the pain of others lower due to an impaired perception of social cues. Their emphatic engagement, however, appears to be high, which contradicts the hypothesis of discrepancies in empathy in individuals with ASD.

Another study that examined the hypothesis of attenuated MN activation and *mu* suppression in adults with ASD in terms of decoding the intentions deriving from motion observations and execution is that of Dumas et al. [49]. The aim of the study was to investigate the validity of this hypothesis for the totality of the brain, focusing on two sub-bands of alpha-*mu* bands (8–10 Hz and 10–12/13 Hz), in contrast to other studies that accept a homogeneity of *mu* suppression in terms of frequency (8–13 Hz) and take into account only the electrodes C3/C4, which are located in the centre of the scalp. They examined ten high-functioning adults with ASD and thirty neurotypical individuals matched in terms of age (20–39 years of age) in a three-condition experiment: simple observation of gestures, free imitation of gestures and imitation of a pre-recorded video. Their findings revealed variations in the *mu* response once two bandwidths of alpha-*mu* were considered. More particularly, the differentiation was detected in the upper sub-band of the sensorimotor region in the ASD group under the condition of a gestural observation, whereas the two groups did not show significant variations in the response of the lower sub-band. The increase in the lower *mu* rhythm band was found atypical, whereas in the higher alpha frequency band it appeared to be normal for the observation condition. On the other hand, the responses to the condition of execution were found normal. The study questions the hypothesis of global impairments in the function of MN in ASD, dissociating attenuations in the process of intention perception from them.

As mentioned above, MN are hypothesised to fire during the observation of an act and could possibly be involved in facial-recognition processing as well as mimicry and imitation processes, reflecting emotional states and ToM abilities [63]. Deschrijver et al. [64] questioned the efficacy of dysfunctional MN as the reason behind deficits in motion observation and imitation abilities in individuals with ASD. Their study aimed to shed light on the cognitive processes that underlie imitation control and imitation impairments in ASD, giving emphasis on three EPR components, the P3, the N190 and the RP in terms of congruency in the stimulus conditions. They tested 23 adults diagnosed with High-Functioning Autism and 23 neurotypical controls matched in terms of age (22–46), handedness and gender. The participants were part of an EEG experiment, during which they were required to observe a videotaped hand performing gestures and to execute a pre-instructed hand gesture right after, under three different conditions: a congruent condition, when the action they were required to execute matched the gesture shown in the video, an incongruent condition, when the gesture they were required to execute did not match the gesture shown in the video and a baseline condition, when the hand shown in the video was in a resting state. The study was the first to conduct a neuroimaging experiment examining the imitation–inhibition task. The findings with respect to the P3 ERP component did not confirm the original hypothesis, demonstrating no significant variation between the ASD group and the neurotypical controls. In that respect, both the individuals with ASD and neurotypical participants generated larger numbers of P3 component during the observation of the congruent gestural action with the gesture they intended to perform. The ASD participants showed the ability to distinguish between compatible and incompatible observed gestures to the intended hand gestures when the processing level was higher. With respect to the RP (readiness potential) Laplacian, the findings suggested that the ASD group had a larger number of RP components for the congruent trials than for the baseline trials. The same effect was also observed for the incongruent trials, which elicited larger P3 Laplacian than the baseline trials. That finding suggests that, in individuals with High-Functioning Autism, the cerebral work load in terms of motor preparation is equally high both when observing a compatible or an incompatible gesture to the planned hand gesture, whereas the influence of the baseline condition appears to be neutral. Unexpectedly, no significant variation was indicated among the two groups in terms of compatible and incompatible conditions. The effect of the intended action on the processing of early visual stimuli (N190), as found in previous studies, could not be replicated in this study. The results postulate the theory that automatic imitation does not exclusively depend on the disentangling of socio-cognitive cues but rather on motor preparation, contrasting the hypothesis of dysfunctions in MN in ASD.

The overall results of studies investigating the relationship between *mu* suppression in ASD and dysfunctional MN are far from conclusive. Aberrant *mu* suppression was not found to be systematically associated with dysfunctional MN, which casts doubts on the robustness of *mu* suppression as a reliable proxy for the functioning of MS or/and the appropriateness of the methodological techniques that have been employed so far in relevant research. This calls for a more in-depth examination of the function of MN and impairments of individuals with ASD at the neurobiological level, as well as interventional methods without invasive techniques. An interventional approach that has received attention in research is the Transcranial direct current stimulation (tDCS), which is a non-invasive cerebral stimulation technique that modulates cortical excitability by applying a low direct current through a set of electrodes on the scalp. In recent years, research using tDCS has gained ground as a great opportunity to causally test the role of specific neural circuits in certain motor or cognitive

functions [65]. Enhanced cortical excitability is found to be linked to anodal stimulation, whereas a weaker excitability is associated with cathodal stimulation [66]. The technique has already found application in measuring modulations in attenuated mirror neurons aiming at a potential decrease of the clinical manifestations of individuals with ASD [67], and also as a treatment method for other clinical conditions accompanied with cognitive impairments, such as schizophrenia or Alzheimer's disease [68,69].

Another important point to consider is the connection between ToM, joint attention and brain connectivity. Joint attention in particular is considered to be a predictive marker for ToM, relying on the efficient integration of information regarding mental states of oneself and of others. It therefore requires successful cooperation among the activation of perceptual neural networks. Deficits in joint attention abilities result in impairments in social engagement, constrain shared intent and imitation, and further diminish the chance of social integration and shared experience opportunities [70]. Jaime et al. [71] examined the EEG coherence during the state of perception of compatible and incompatible joint attention as well as an eye-open resting state. The researchers tested 16 high functioning adolescents with ASD (mean age: 16.2 years) and 17 neurotypically developing controls (mean age: 16.5 years). The participants were presented with 12 short video clips containing a moving red dot and a human model, evoking joint attention with two conditions: a congruent one, where the human model was following the dot with their gaze, and an incongruent one, where the human model was not. The findings of the study showed a low alpha coherence in the central-temporal area of the right hemisphere in the ASD group, which is in alignment with the findings of research studies investigating EEG coherence in adults and children [72,73]. The condition of congruence in joint attention perception did not act as an influencing factor for EEG coherence, neither in the ASD group, nor in the control group. The authors interpreted this finding as supporting that adolescents with ASD have no dysfunction in the frontal-parietal attention-oriented network. Overall, the results support the theory of underconnectivity in ASD. The theory of underconnectivity offers a different dimension, postulating that an aberrant frontal-posterior interaction exacerbates the communication and information exchange between the frontal and the posterior regions that are involved in cognitive activities such as joint attention.

Table 1 below provides an overview of the studies that were selected to be reviewed in this paper:.


**Table 1.** Overview of the selected research studies for this review.


### **4. Conclusions**

ToM is a multifaceted approach, which encompasses a variety of cognitive and neurobiological aspects and has been found to be impaired in individuals with ASD. One of the theories that attempts to account for some of these impairments is that of "broken mirror neurons", indicating dysfunctions in the proper activation of a neural circuit responsible for the efficient perception of motion activity. The aberrant firing of this neuronal circuit is suggested to have a negative impact on the ability to encode the intentions behind observed actions and further burdens the mechanism that underlies imitation, joint attention, empathy and ToM in ASD. The present review examined the most recent available studies, in particular studies conducted within the past 15 years, with respect to the connection between the function of MN in individuals with ASD and ToM-reflecting sensorimotor, social and attentional stimuli. The neuroimaging studies reviewed in this paper examined the modulation of attenuation of the *mu* rhythm in ASD using EEG screening tests as a marker for measuring MNS activity. The majority of them approached the theory of broken mirrors critically; the results, however, are contradictory, presenting divergent findings in terms of *mu* suppression and its relation to the performance of individuals with ASD in the experimental tasks of these studies. This deviation may be attributed to the large variation of phenotypical symptomatology across the spectrum of autism, as well as to the limitation of the methodological approaches of the research studies, such as limited sample numbers, a restriction to examining only specific cerebral areas, as well as an inadequate connection of the *mu* suppression emergence to other cognitive operations. Nevertheless, the review revealed discrepancies in the function of MNS in ASD, despite the fact that the activity of this neural network is differently interpreted by the researchers of each study. A clear pattern of aberrant *mu* suppression in ASD is, however, indicated in the reviewed studies, without it being exclusively attributed to dysfunctional MN. The role of MN or cerebral motor activation in general has been challenged, even in neurotypically developing infants. The study of Southgate and Begus [74] showed that nine-month-old children demonstrated motor activation in anticipation of an action, regardless of whether the action was within their own skillset of movement or not. More particularly, the study demonstrated independence in terms of coupling the perceived action with a matching motion representation, indicating that the suppression of the alpha wave is linked to action prediction but that it does not necessarily indicate the activation of MN. These findings were interpreted in alignment with the findings of the study of Kilner et al. [75], which demonstrated that MN are involved not only in the observation of an action but

also in the anticipation of a motion of another person, which facilitates the prediction of intended action goals before the execution of the action itself. The common outcome deriving from this review is that individuals with ASD exhibit deficits in ToM-related cognitive processes, such as the perception and mentalisation of actions in terms of observation execution and imitation, especially under conditions of unfamiliar social engagement. Impairments in the interpretation of social cues further burden social communication in ASD. It is worth mentioning, however, that the findings of this review suggest a relation between low performance in mentalising tasks, which is nevertheless not correlated to autistic traits. It would therefore be of particular interest to investigate *mu* suppression as a neurophysiological operation and the way in which it is linked to mechanisms such as mentalising. It is crucial to conduct further research, in order to gain a more conclusive insight regarding the mechanisms underlying the connection between ToM and neurophysiological operations.

**Author Contributions:** Conceptualization, M.A.; investigation, M.A. and V.S.; writing—original draft preparation, M.A. and V.S.; writing—review and editing, M.A. and V.S.; supervision, M.A. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was part of the SFB 1252 "Prominence in Language" in the project C03 "Reference management in bilingual narratives" (Principal Investigators: Prof. Christiane Bongartz and Prof. Jacopo Torregrossa) at the University of Cologne and was funded by the German Research Foundation (DFG).

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **The E**ff**ects of the Early Start Denver Model for Children with Autism Spectrum Disorder: A Meta-Analysis**

### **Elizabeth A. Fuller, Kelsey Oliver** †**, Sarah F. Vejnoska** † **and Sally J. Rogers \***

Department of Psychiatry and Behavioral Sciences, University of California, Davis MIND Institute, Sacramento, CA 95817, USA; efuller@ucdavis.edu (E.A.F.); kaeoliver@ucdavis.edu (K.O.); sfvejnoska@UCDAVIS.EDU (S.F.V.)

**\*** Correspondence: sjrogers@ucdavis.edu

† These authors contributed equally to this work.

Received: 16 May 2020; Accepted: 9 June 2020; Published: 12 June 2020

**Abstract:** This meta-analysis examined the effects of the Early Start Denver Model (ESDM) for young children with autism on developmental outcome measures. The 12 included studies reported results from 640 children with autism across 44 unique effect sizes. The aggregated effect size, calculated using a robust variance estimation meta-analysis, was 0.357 (*p* = 0.024), which is a moderate effect size with a statistically significant overall weighted averaged that favored participants who received the ESDM compared to children in control groups, with moderate heterogeneity across studies. This result was largely driven by improvements in cognition (*g* = 0.412) and language (*g* = 0.408). There were no significant effects observed for measures of autism symptomology, adaptive behavior, social communication, or restrictive and repetitive behaviors.

**Keywords:** autism; early intervention; Early Start Denver Model

### **1. Introduction**

The estimated prevalence of autism spectrum disorders (ASD) has continuously increased in recent decades with the most current prevalence rates estimating that 1 in 54 children under 8 years of age are diagnosed with ASD [1]. This includes an increasing prevalence of young children being diagnosed partly due to the more widespread use of early screening measures and adaptations to diagnostic tools that has led to children being diagnosed with ASD as early as 12–18 months [2]. Children this young need early intervention services that have been designed for and tested with them, given the many developmental and social-emotional differences of infants and toddlers when compared to preschoolers and older children [3]. Given the increasing prevalence estimates of ASD and the high cost of ASD treatments [4], it is critical to identify ASD intervention approaches that are appropriate and effective for supporting young children and their families.

### *1.1. Naturalistic Developmental Behavioral Interventions*

Naturalistic developmental behavioral interventions (NDBIs) are one class of ASD interventions that are particularly geared towards the needs of young children [5]. The term NDBI describes interventions that use strategies involving naturally–occurring environments and activities, child-responsive interaction styles, and teaching content and strategies derived from developmental science as well as the science of applied behavior analysis.

In a recent systematic review and meta–analysis of early interventions for children with ASD, Sandbank and colleagues [6] identified a subset of 26 group design studies that examined the effects of NDBIs and found that the NDBIs showed the strongest body of evidence compared to the other

intervention types included. However, the NDBI studies had multiple methodological and quality limitations across them, especially where 47.59% used outcome measures that were proximal to the intervention goals and 78.77% measured outcomes in contexts similar to the intervention context. Previous reviews have indicated that studies that use proximal and context-bound measures likely inflate intervention effects [7,8]. Additionally, 47.09% used outcome measures at risk of correlated measurement error (CME) due to the participation of adults in outcome measurement who have been trained in the intervention strategies. Sandbank and colleagues found that the group of 26 NDBIs resulted in significant improvements in social communication, cognition, play, and language, but, when examining results from only those studies that did not rely on a parent report (a measurement type that is susceptible to CME), only play and social communication outcomes showed significant improvements.

### *1.2. Early Start Denver Model*

The Early Start Denver Model (ESDM) is an NDBI specifically designed for the needs of very young children with ASD that has been widely studied [9]. The ESDM is one of the few comprehensive early intervention programs for ASD. Although it has a particular focus on autism-specific impairments, it teaches skills across nine developmental domains. The ESDM, which is one of the few commercially available NDBIs, has previously been identified as a promising and cost-effective intervention [10] and has been examined in two systematic reviews. The first review included 15 studies using a variety of study designs [11] and reported overall positive results. However, over half of the included studies had methodological weaknesses. A second review [12] of 10 studies found similar findings and reported that, although most of these studies had positive results, the three comparative studies had mixed findings. Problems of study quality in both meta-analyses included lack of true experimental designs, lack of blind assessment, and small sample sizes.

The purpose of this meta–analysis was to expand and improve upon the findings of these previous reviews in several ways: by including many more recently published studies, by using a meta-analytic approach that allowed for a quantitative understanding of effects, by focusing on comparative studies, and by examining effects on specific domains as well as overall effects of the intervention. This would help identify strengths and areas needing improvement for a well–known early ASD intervention.

### *1.3. Research Questions*

This systematic review and meta-analysis of the effects of the ESDM on outcomes for young children with ASD was conducted to address the following questions: (1) Does the ESDM result in significant improvements in outcomes for young children with ASD, both overall and specifically in the domains of autism symptomology, language, cognition, social communication, adaptive behavior, and repetitive behaviors? (2) Are the findings affected by quality and study design features, including proximity and boundedness of measurement?

### **2. Materials and Methods**

### *2.1. Eligibility Criteria*

Eligibility criteria are presented in Table 1. Studies were included in the meta-analysis if the study enrolled participants with ASD or at risk for ASD under age 6. The intervention type was restricted to the ESDM, but could include individual, group, or parent-implemented ESDM, or interventions that were derived from ESDM (e.g., Infant Start [13]). Study design was restricted to group comparison studies (randomized control trials or quasi experimental designs). Included studies were required to have a non–ESDM treatment comparison group, which could include: treatment as usual, waitlist control, or parent education only, or a treatment comparison that did not include ESDM interventions. Studies that did not have a comparison group (e.g., single case design or pre/post design) were excluded. Studies had to report at least one child outcome that provided adequate information to calculate a standardized mean difference effect size (e.g., means and SDs or F statistics). Studies had to be published in English to be eligible for inclusion due to the language restraints of the coders. Follow-up studies were excluded as the only data from the timepoint closest to the end of the intervention.



### *2.2. Search Procedure*

A total of nine databases were searched through Proquest: (American Psychological Association (APA) PsycArticles, APA PsycInfo, APA PsycTests, Dissertations and Theses at the University of California, Education Resources Information Center (ERIC), Linguistics and Language Behavior Abstracts, PAIS, ProQuest Dissertations and Theses A&I, Sociological Abstracts). The final search was completed in October 2019. Unpublished or "gray" literature was searched using the online databases of dissertations and theses as well as proceedings from relevant conferences (e.g., International Society for Autism Research) and reference lists. The search and study selection process were completed by the first author.

### *2.3. Data Extraction and Coded Variables*

All child outcome measures that were reported were recorded from each study. If a study reported both a total or overall score and subscale scores, only the total/overall score was used. However, the subscale scores were used for the appropriate outcome–specific meta–analysis. For example, if a study reported both the overall developmental quotient from the Mullen Scales of Early Learning (MSEL) and the subscales, the overall score was used in the overall outcome analysis, and the expressive and receptive language subscales were used in the language outcomes analysis [14]. Only outcomes from the timepoint most proximal to the end of the intervention were included.

Study-level characteristics were recorded, including the location in which the study took place, length of intervention delivery (in weeks), intensity of delivery (hours per week), mean child age (in years), percent of participants that were male, the primary person implementing the intervention (parent or professional, which included researcher, teacher, or therapist), whether the intervention included a parent training component, the format of the intervention delivery (individual, group, or mixed), and the fidelity of the intervention implementation, if reported.

Study quality indicators were recorded, including the use of random assignment and the use of assessors who were blind or naïve of the group assignment. The measurement–quality variables were coded using definitions and flowcharts described in Sandbank and colleagues [6]. Each measure was coded according to the proximity and context of the measure.

*Measurement proximity.* Proximity of the measurement was coded as distal or proximal. Distal measures were defined as those behaviors measured using developmentally–scaled tests meant to measure general development. Proximal measures were defined as those in which the measurement directly measured the goals of the intervention. For example, the MSEL would be considered a distal measure, whereas a child's ability to imitate would be considered a proximal measure since this is a behavior that is specifically targeted in the ESDM curriculum.

*Measurement context.* The context of measurement was coded as generalized or context bound. Generalized outcomes were defined as outcomes that were measured in a context differing from the intervention context of at least one dimension (setting or interaction partner). Context–bound measures (CME) were defined as those that were taken in the same context as the intervention was delivered. For example, measuring a child's language using a subscale of the MSEL would be coded as generalized because it uses different materials, interaction styles, and a likely interaction between the partner and setting, whereas measuring a child's language during an intervention session with their usual therapist would receive a context-bound code. Parent questionnaires were coded as generalized because they are intended to capture the child's generalized tendency to behave in the home context. The use of parent/teacher reports was also coded. Potential for CME was defined as any measure involving an adult trained in the intervention. This included a parent report if and only if the parent had been trained in the intervention.

### *2.4. Analytic Strategies*

The standardized mean difference effect size was calculated using Hedges' g to compare group differences (treatment vs. control) at post-test. Hedges' *g* corrects the slight bias in Cohen's *d* that occurs in studies with small sample sizes, and is, therefore, a more conservative estimate of effect in a sample of studies with high variability [15]. When studies did not report means and standard deviations, the effect size was calculated from an F-statistic, derived from a group\*time ANOVA to mitigate the concern of effect-size inflation [16].

A robust variance estimation (RVE) meta-analysis was conducted using the robumeta package on R [17]. The RVE meta–analysis accounts for the nesting of multiple effect sizes within one study [18]. This method was selected rather than traditional meta-analyses, which use only one effect size per study, to account for the fact that the ESDM targets a variety of skills and its efficacy is generally assessed using more than one outcome measure. Separate meta analyses were conducted for each subskill analysis using separate RVE meta-analyses. Meta-regression analyses were conducted to understand the contributing factors of study-level characteristics (dose and person implementing) and study quality indicators. The heterogeneity of effect sizes was examined using τ<sup>2</sup> and *I <sup>2</sup>*. Between study variance represented by τ2, which is in the metric of the effect size. *I <sup>2</sup>* represents the percent of variability that is true heterogeneity across the observed effect estimates. Higher levels of *I <sup>2</sup>* indicate greater dispersion between effect sizes that may be accounted for with moderator analyses [19]. A *p* < 0.05 alpha level was selected as the level of significance for all analyses.

A primary coder (the first author) read and extracted the data from all studies. A second person independently extracted the data from each study so that all variables on 100% of the included studies were coded by two raters. Overall reliability of independent ratings across all coded measures was 97.2%. Disagreements were resolved by first verifying the information in the manuscript and then by discussing between coders, if needed, until agreement was reached so that 100% agreement on all variables was reached. All statistical analyses were completed using the verified data set.

Although efforts were made to minimize publication bias by including gray literature searches, analyses were included to detect bias. Publication bias was examined through visual analysis of a funnel plot and the Egger's test of a small study bias [20].

### **3. Results**

### *3.1. Study Selection*

The initial search identified 411 articles to be screened for inclusion. After the initial and full-text screening of the identified articles, 12 studies, including 11 published manuscripts and one dissertation [21], were included in the final analysis. A Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) flow diagram of exclusion procedures is provided in Figure 1.

**Figure 1.** Prisma diagram of study inclusion.

### *3.2. Study Characteristics*

The 12 included studies were published between 2010 and 2019. The studies took place in five different countries: Australia, Austria, China, Italy, and the United States. The studies included 640 participants (286 intervention and 354 control). The participants ranged in age from nine months to five years old with an average overall age of 2.51 years (SD = 0.89). The studies that reported on gender reported that 80.6% of the samples were male. A total of 44 different effect sizes were reported across the 12 studies. A range of outcome measures were used. Overall study characteristics are shown in Table 2 and characteristics specific to each effect size are shown in Table 3.

In five studies, the parent was the sole agent of implementation. An additional five studies used an intervention approach that incorporated parent coaching but was primarily implemented by a professional. Four studies used a group-based approach: two studies trained parents in groups [21,22] and two studies used group-delivered ESDM [23,24]. Outcomes of studies that included parents did not show significantly higher outcomes than those that did not (B = 0.289, *p* = 0.39). Overall fidelity of implementation was high (mean = 83.2%, range = 75–92%). The studies used a wide range of intervention dosages both in intensity and in length, ranging in intensity from one hour per week to 20 hours per week, and ranging in length from six weeks to 156 weeks. This resulted in total hours of intervention ranging from 12 hours to 2080 hours. However, a meta-regression showed that child outcomes were not significantly related to the length of intervention (B = −0.01, *p* = 0.46), to the hours per week of intervention (B = −0.02, *p* = 0.73), or to the total number of hours (B = 0.004, *p* = 0.66). Additional information about what interventions the control groups received during the study period is included in the Table A1 (Appendix A).


Rogers (2012) [26]

MCDI: Vocabulary MCDI: Vocabulary Produced

MCDI: Total Gestures

VABS Composite

ADOS RRB Imitative Sequence Mean Social Orienting Mean Non-Social Orienting Mean orient to Joint Attention

Comprehension

106.5 (96.8)

 42.3 (62.0)

 28.02 (12.6)

 77.4 (9.6)

 4.0 (1.9)

 4.6 (3.5)

 0.5 (0.3)

 0.7 (0.3)

 0.3 (0.3)

 125.7 (106.4)

 38.9 (73.7)

 29.8 (13.5)

 80.3 (11.3)

 3.8 (2.0)

 3.8 (3.4)

 0.4 (0.4)

 0.6 (0.4)

 0.3 (0.3)

 0.24 (0.20)

 0.13 (0.20)

 0.42 (0.20)

 0.00 (0.20)

−0.19 (0.20)

 0.05 (0.20)

−0.13 (0.20) −0.27 (0.20)

−0.07 (0.20)

 Yes

 Yes

 Yes

 Yes

 Yes

 No

 No

 No

 No

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 Yes

 No

 No

 No

 No

 No

 No

 No

 No

 No

 No

 Yes

 Yes

 Yes

 Yes


PDDBI: Pervasive

Developmental

 Disorder Behavior Inventory, CME: Correlated Measurement

 Error.

### *3.3. Overall Outcomes*

Figure 2 shows the results of the RVE meta–analysis examining the effects of the ESDM on all included outcome measures. The effect size weight is shown for each of the 44 outcome measures arranged by the study. Larger black boxes around the effect sizes represent larger weights in the meta–analysis, and bars represent the confidence intervals. The RVE aggregated effect size resulted in an overall effect size of *g* = 0.357 (*p* = 0.024). This moderate and statistically significant effect size suggests a significant advantage for children who received the ESDM intervention compared to children enrolled in control groups. However, a moderate amount of between–study heterogeneity was observed in this analysis (*I* <sup>2</sup> = 64.84%, τ<sup>2</sup> = 0.16). The majority of studies showed confidence intervals that overlapped with zero, which indicated that the RVE aggregated effect was driven by a few studies or by specific outcome measures. This further assessed the subgroup analyses below.

**Figure 2.** Main effect of the Early Start DenverModel (ESDM) intervention on developmental and symptom outcomes. *Note.* ADOS: Autism Diagnostic Observation Schedule, RRB: Restrictive andRepetitive Behavior, MSEL: Mullen Scales of Early Learning, DQ: Developmental Quotient, ELC: Early Learning Composite, MCDI: MacArthur Bates Communicative Development Inventory, VABS: Vineland Adaptive Behavior Scales, CSBS: Communication and Symbolic Behavior Scales, GDS: Griffith Developmental Scales, CARS: Childhood Autism Rating Scale; WASI FSIQ: Wechsler Abbreviated Scale of Intelligence Full Scale Intelligence Quotient, PDDBI: Pervasive Developmental Disorder Behavior Inventory. *Note.* Black boxes indicate the weight of each effect size and bars indicate the confidence interval. The overall effect size is indicated by the open diamond and dotted line (*g* = 0.357).

### *3.4. Study Quality Indicators*

The studies were analyzed for their use of study design elements. Study level quality elements (blind assessors, random assignment) are reported in Table 2, and effect–size specific elements (if the measure was distal, generalized, relied on parent report, or showed potential risk of CME) are reported in Table 3. Thirty-eight of the forty-four (83.2%) elements included measures used developmentally–scaled, distal measures of child outcomes. Forty-three of the included measures (97.7%) used generalized contexts to measure child outcomes. Fourteen outcome measures used parent report measures (31.8%), and 10 outcome measures (22.7%) had potential risk for CME (nine of these 10 studies due to the use of parent report measures). Blind assessors were used in 72.2% of eligible studies (eight out of eleven studies with one study not being included since it only used a parent report so that no assessors were used). Six of the 12 studies (50%) used a randomized study design. A meta regression analysis showed that child outcomes were not significantly associated with distal outcomes (B = 0.28, *p* = 0.47), generalized outcomes (−0.38, *p* = 0.20), parent report (B = −0.08, *p* = 0.70), use of blind assessors (B = 0.15, *p* = 0.74), or the use of a random assignment (−0.02, *p* = 0.95). Furthermore, the inclusion of these variables did not account for the observed heterogeneity (*I* <sup>2</sup> = 76.22%, τ<sup>2</sup> = 0.26). Because of the high overlap between the use of parent measures and the potential risk of CME, only the variable for the use of parent measures was retained in the meta-regression analysis.

### *3.5. Autism Symptoms*

Figure 3 displays the forest plot for the 10 autism symptomology outcomes that were reported across nine studies. The effect sizes are represented such that positive values indicate a reduction in autism symptomology. The aggregated effect size was *g* = 0.070 (*p* = 0.616), which indicated that children who received ESDM treatment did not show significant improvements in autism symptomology when compared to the control group. A moderate level of heterogeneity was observed (*I* <sup>2</sup> = 48.90%, τ<sup>2</sup> = 0.073).

**Figure 3.** Main effect of the Early Start Denver Model (ESDM) intervention on autism spectrum disorder (ASD) symptomology outcomes. *Note.* ADOS: Autism Diagnostic Observation Schedule, RRB: Restrictive and Repetitive Behavior, CARS: Childhood Autism Rating Scale; PDDBI: Pervasive Developmental Disorder Behavior Inventory.

### *3.6. Language*

Figure 4 displays the forest plot for the 19 language outcomes that were reported across 11 studies. The effect sizes represent both expressive and receptive language outcomes. The aggregated effect size was *g* = 0.408 (*p* = 0.011), which indicates that children who received the ESDM intervention made significant progress in language development compared to children in the control groups. A moderate level of heterogeneity was observed (*I* <sup>2</sup> = 52.70%, τ<sup>2</sup> = 0.088).


**Figure 4.** Main effect of ESDM intervention on language outcomes. *Note*. MSEL: Mullen Scales of Early Learning, DQ: Developmental Quotient, MCDI: MacArthur Bates Communicative Development Inventory, GDS: Griffith Developmental Scales, WASI: Wechsler Abbreviated Scale of Intelligence, PDDBI: Pervasive Developmental Disorder Behavior Inventory.

### *3.7. Cognition*

Figure 5 displays the forest plot for the 13 cognitive outcomes that were reported across nine studies. The aggregated effect size was *g* = 0.412 (*p* = 0.038), which indicated that children who received the ESDM intervention made significant progress in cognitive development compared to children in the control group. A moderate level of heterogeneity was observed (*I* <sup>2</sup> = 66.30%, τ<sup>2</sup> = 0.145).

**Figure 5.** Main effect of ESDM intervention on cognitive outcomes. *Note.* MSEL: Mullen Scales of Early Learning, DQ: Developmental Quotient, ELC: Early Learning Composite, GDS: Griffith Developmental Scales; WASI FSIQ: Wechsler Abbreviated Scale of Intelligence.

### *3.8. Social Communication*

Figure 6 displays the forest plot for the 19 social communication outcomes that were reported across eight studies. This included related sub-scores of the Vineland (Communication and Socialization) [32]. The aggregated effect size was *g* = 0.209 (*p* = 0.285), and was not statistically significant. A high amount of heterogeneity was observed across social communication measures (*I* <sup>2</sup> = 72.53%, τ<sup>2</sup> = 0.176).

**Figure 6.** Main effect of ESDM intervention on social communication outcomes. *Note*. ADOS: Autism Diagnostic Observation Schedule, MCDI: MacArthur Bates Communicative Development Inventory, VABS: Vineland Adaptive Behavior Scales, CSBS: Communication and Symbolic Behavior Scales, GDS: Griffith Developmental Scales.

### *3.9. Adaptive Functioning*

Figure 7 displays the forest plot for the six adaptive functioning outcomes that were reported across six studies. All of the included effect sizes were taken from the Vineland [32]. The aggregated effect size was *g* = 0.121 (*p* = 0.458), which was not statistically significant. A moderate amount of between-study heterogeneity was observed (*I* <sup>2</sup> = 49.03%, τ<sup>2</sup> = 0.062).

**Figure 7.** Main effect of ESDM intervention on adaptive functioning outcomes. *Note.* VABS: Vineland Adaptive Behavior Scales.

### *3.10. Repetitive Behaviors*

Figure 8 displays the forest plot for the five repetitive behavior outcomes that were reported across five studies. The effect sizes are represented such that positive values indicate a reduction in repetitive behaviors. The aggregated effect size was *g* = −0.016 (*p* = 0.876), which indicated that children who received ESDM treatment did not show significant improvements compared to the control group in repetitive behaviors. This finding should be taken with caution due to the low number of included effect sizes.

**Figure 8.** Main effect of ESDM intervention restricted and repetitive behaviors (RRB). *Note*: RBS: Repetitive Behaviors Scale; ADOS RRB: Autism Diagnostic Observation Schedule Restricted and Repetitive Behaviors.

### *3.11. Publication Bias*

An Egger's test of a small study bias (*p* < 0.01) indicated that there is a risk of a small study bias in this sample. A funnel plot is included in Figure A1 (Appendix B), which shows that two of the 44 effect sizes fall outside of the highlighted area, suggesting a small bias.

### **4. Discussion**

This meta-analysis examined the effects of the ESDM for young children with ASD delivered in a variety of formats on a variety of outcomes measures. Across 12 studies that included 44 unique effect sizes, the overall aggregated effect size was *g* = 0.357 (*p* = 0.024). This moderate [33] and statistically significant effect size indicates an overall advantage for children in the ESDM intervention groups compared to children in control groups (*p* = 0.024). (For reference, this represents a gain of 7.84 more points on the Mullen Developmental Quotient than the comparison group.) These significant differences were mostly driven by improvements in cognition (*g* = 0.412) and language (*g* = 0.408). There was a moderate amount of heterogeneity across studies and significant results were not observed for all studies or outcome measures. Nonsignificant differences were observed for the remaining domains: autism symptomology, adaptive behavior, social communication, and restricted and repetitive behaviors (RRBs). Although many of these effect sizes came from one lab, the 12 included studies represent data from five different countries and from interventions of both high and low intensity implemented using a variety of delivery methods including parents, local teachers or therapists, and group-based settings.

One particular strength of this meta-analysis was the general rigor of the measurements used in the included studies. Relatively few measures were at risk of CME, which occurs when measures involve parent interactions with children or parent reports of child skills in studies that have trained parents in the intervention. In the current sample, only 22.7% of studies had potential risk of this source of CME. This is a great deal fewer than the group of NDBI studies that Sandbank and colleagues [6] reported on, which found that 47% of outcomes were at risk for CME. In addition, most studies included in this meta–analysis used norm-referenced measures that were distal (83%) and generalized (97%). This is considerably more than the general pool of NDBI studies included in the Sandbank analysis in which 52% of outcomes used distal measures and 21% used generalized measures. The high rate of distal and generalized measures seen in this current sample of studies reduces concerns of effect size inflation due to the measurement error.

Given the relative strength in the quality of measurements used in the studies included in the current review, the current findings of significant improvements in language and cognition related to the ESDM compare favorably with previous reviews of ASD interventions. Sandbank and colleagues [6] found that, although NDBIs are generally making significant improvements across domains, the improvements in language and cognitive outcomes as a result of NDBIs were mostly smaller in magnitude (language: *g* = 0.21, *p* < 0.05, cognitive: *g* = 0.18, not significant). In comparison, the present analysis showed significant language and cognitive improvements of *g* = 0.408 and 0.412, respectively. The effect sizes for language in the present ESDM study is also larger than the effect size of *g* = 0.26 reported in a recent meta-analysis that examined language outcomes of multiple types of early ASD interventions [34].

### *Limitations and Future Directions*

The most prominent limitation was the heterogeneity observed in this sample. This meta-analysis combined a wide range of study designs, measures, and procedures. Twelve of the 44 outcome measures showed results in the negative direction, and the majority of outcomes had a confidence interval that included zero. Thus, the overall positive effect size should be taken cautiously.

Two of the potential contributors to the observed heterogeneity in this analysis involved dosage and delivery. A wide range of dosage was used across the 12 included studies in terms of length of intervention and intensity of intervention. Although neither length or intensity of dosage were significantly related to outcome magnitude, this lack of association should be considered with caution. In terms of delivery, five of the studies used a parent-implemented approach. In these studies, the dosage refers to the amount of time the parent was coached rather than the amount of time the parent used the strategies with the child. Four of the studies used a group–based approach. In this case, intensity of individual receipt of intervention is likely different from studies that used a one–on–one delivery approach. Thus, the true dosage of intervention is hard to quantify in some of these studies. Lack of relationship between dosage and outcomes has also been shown in several previous meta–analyses of early interventions [7,34]. Further study is needed to understand the role of dosage in intervention outcomes.

A second limitation was in the scientific rigor of the study designs. While all studies were controlled, half of the included studies used a non-randomized control design. Although the meta-regression indicated that there was not a significant relationship between the use of a non-random design and study outcomes, the negative beta weight indicates that randomized controlled studies had smaller effect sizes than the quasi–experimental studies on average. Many of the quasi-experimental studies were carried out outside of university and lab settings, including community implemented studies [28] in which it was not considered feasible or ethical to implement a randomized design. We included these quasi-experimental controlled studies despite the design limitations to represent findings of real-world applications of the ESDM. Other problems with rigor include use of measures based on parent report, outcome measures that were at risk of a correlated measurement error, and non–blinded assessors (in some of the studies).

A third limitation relates to the subgroup meta-analysis that showed nonsignificant changes on measures of autism symptomology, adaptive behaviors, repetitive behaviors, and social communication. This indicates that the ESDM intervention may be less effective at targeting these characteristics of early ASD. However, in the case of ASD severity and RRBs, this may also be partly due to an issue in measurement. Many of the outcome measures included for these domains came from the Autism Diagnostic Observation Schedule (ADOS) [35]. The ADOS is intended to capture relatively stable characteristics of ASD symptomology including social communication for diagnostic purposes and was not created with the intention of measuring a treatment-related change. A more recent measure, known as the Brief Observation of Social Communication Change (BOSC) [36], was created for this purpose, and may be a more useful tool for capturing change in these outcomes. Future studies should further examine these subdomains using more sensitive outcome measures and should consider additional intervention strategies to specifically target these areas.

A final limitation is the risk of small study bias observed. However, this concern was mitigated by using a correction for small study effects included in the RVE meta-analyses estimation and through extensive searching of gray literature, which included one unpublished study.

### **5. Conclusions**

Based on the moderate and significant overall effect size resulting from this meta-analysis involving 640 participants across 12 studies, the ESDM shows promise as an effective practice for young children with ASD in improving outcomes in some areas affected by early ASD, especially language and cognitive outcomes. Domains involving autism symptomology, social communication, adaptive behaviors, and repetitive behaviors did not show an ESDM advantage and may require additional treatment efforts and/or more sensitive outcome measures. This body of evidence has several strengths in scientific rigor including the use of distal and generalized outcome measures and lowered risk of correlated measurement error compared to other NDBI interventions, but also shows a weakness in the number of quasi-experimental non-randomized study designs. Lastly, the studies reported high fidelity of treatment implementation across a variety of delivery contexts, including five different countries, group and individual settings, and a range of implementors that included parents, community therapists, and teachers.

**Author Contributions:** Conceptualization, E.A.F. and S.J.R. Methodology, E.A.F. Validation, K.O. and S.F.V. Formal analysis, E.A.F. Writing—original draft preparation, E.A. Writing—review and editing, E.A.F., K.O., S.F.V., and S.J.R. Supervision, S.J.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received funding from NIH T32MH07312 during the time period 2018-2020. The funding sources were not involved in the study design or implementation.

**Conflicts of Interest:** Sally J. Rogers has received royalties from Guilford Press and honoraria for lectures related to this paper.

### **Appendix A**


**Table A1.** Description of the control group.

### **Appendix B**

**Figure A1.** Funnel plot of included studies: effect size and standard error.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Epidemiology of Autism Spectrum Disorders: A Review of Worldwide Prevalence Estimates Since 2014**

### **Flavia Chiarotti \* and Aldina Venerosi \***

Reference Center for the Behavioural Sciences and Mental Health, Italian National Institute of Health, 00161 Rome, Italy

**\*** Correspondence: flavia.chiarotti@iss.it (F.C.); aldina.venerosi@iss.it (A.V.)

Received: 29 March 2020; Accepted: 21 April 2020; Published: 1 May 2020

**Abstract:** The prevalence of Autism Spectrum Disorder (ASD) has increased dramatically in recent decades, supporting the claim of an autism epidemic. Systematic monitoring of ASD allows estimating prevalence and identifying potential sources of variation over time and geographical areas. At present, ASD prevalence estimates are available worldwide, coming either from surveillance systems using existing health and educational databases or from population studies specifically performed. In the present article, we present a review of the ASD prevalence estimates published since 2014. Data confirm a high variability in prevalence across the world, likely due to methodological differences in case detection, and the consistent increase of prevalence estimates within each geographical area.

**Keywords:** prevalence estimate; autism; predictors; surveillance review

### **1. Introduction**

In the last decades, a large increase in the prevalence of Autism Spectrum Disorder (ASD) has been observed, generating claims about an "epidemic" of autism [1,2]. Correct estimates of ASD prevalence rates are important, firstlyin order to determine the economic and health services burden of this condition and to allocate sufficient funding and adequate services for children and adults with ASD and their families. A growing population of people with ASD implies the necessity of increased service availability including training of professionals, as well as identification of additional resources that can emerge by the recognition of cases in the population [3]. Furthermore, accurately determining ASD prevalence can help to understand which groups are exposed to disparities in healthcare access for developmental evaluations [4], besides being more at risk for ASD due to geographical and environmental factors [5].

Studies that estimate ASD prevalence result in wide variability of prevalence rates that call for paying attention on possible reasons for the observed changes in prevalence, and advice for caution when claiming that there is an autism epidemic [2,6].

One important source of variation in prevalence estimates are the methodological differences in case definition and case-finding procedures. In particular, some studies are carried out on existing administrative databases such as special education data, health or social records of national registers for case identification, or specific condition registers (defined as "administrative data" when relying on one database, or "multisource" when combining data from multiple databases). Other studies rely on a two-stage or multistage approach to identify cases in underlying populations; the first stage is often based on questionnaire requesting behavioural descriptions or checklist based on DSM, where informants could be in turn teachers, parents or health professionals (defined as "ad hoc studies"). Finally, some studies are surveys based on interviews to parents or teachers, who are required to state if the child presents a condition that can be related to ASD (defined as "reports"). Obviously,

sample size and catching area represent further characteristics of the studies that can affect prevalence estimates [7]. Indeed, surveyed areas vary in terms of service development as a function of the specific educational or health care systems of each country and of the year of the study [6]. Moreover, socio economic factors [8,9] and autism awareness [10] can influence assessment of the case and consequently prevalence estimate.

Case definition is the other challenge that affects prevalence estimate. Diagnostic category (AD, ASD, PDD), as well as age range considered are very important sources of prevalence estimate variation [7]. Changing definitions and labelling practices that change over time, as in the case of the introduction of diagnostic manuals' revisions, can produce change in labelling but also "diagnostic substitution" whereby similar symptoms can be classified under different disabilities during different time periods [11,12]. Lastly, cultural influence can affect the definition of case causing differences in the estimation of prevalence in different ethnic/cultural groups [13,14].

In the present paper, we present a brief narrative review of the most recent ASD prevalence estimates worldwide. We describe evidences according to two main criteria, i.e., the geographical setting and the case-finding procedure of the study. Finally, we attempt to demonstrate if these criteria act as predictive factors for underestimating or overestimating prevalence figures.

### **2. Prevalence Estimates**

Many prevalence studies have been performed worldwide since 1966. In 2012, Elsabbagh et al. [7] published a comprehensive review of the studies performed until 2012: these studies differed with respect to diagnostic category, diagnostic criteria, age at prevalence evaluation, extent of the targeted geographical area, and source of data on the diagnoses. These methodological differences, together with the large time span (almost 50 years from the first to the last study included in the review), at least partly account for the large differences observed in the estimated prevalence. Overall, estimates ranged from 0.19/1000 to 11.6/1000. The former estimate refers to the Autistic Disorder (AD), diagnosis based on Rutter's criteria (1978), age range 0–14 years, geographical area of West Berlin (Germany), and on data extracted from the registry of the university clinic of child psychiatry and/or the German Society for Autistic Children (1986). The latter refers to the Pervasive Developmental Disorder (PDD), diagnosis based on ICD-10, age range 9–10 years, geographical area of South Thames (UK), ad hoc study (2006). Taking into account the diagnostic category, the median prevalence estimates were 1.00/1000 for AD (from 0.19/1000 in Germany to 7.26/1000 in Sweden) and 6.16/1000 for PDD (from 3.00/1000 in Denmark to 11.6/1000 in UK). When considering PDD, the median prevalence was similar to the USA overall ASD prevalence estimated in 2000–2002, but much lower than the USA prevalence estimates since 2006.

In 2014, Tsai updated the review by [7], but only negligible differences emerged in the median prevalence estimates, which were confirmed to be 1.32/1000 for AD (from 0.19/1000 in Germany to 7.26/1000 in Sweden) and 6.19/1000 for PDD/ASD (from 3.00 in Denmark 2002 to 12.3/1000 in Netherlands) [15]. The review by Tsai included almost all papers evaluated by [7], specifically 59/59 = 100% for AD and 33/35 = 94.3% for PDD prevalence studies. In his review, Tsai examined 15 and 28 additional studies estimating AD and PDD/ASD prevalence, respectively. Table 1 reports a summary of the results of the reviews by [7,15].

Since the publication of the reviews by Elsabbagh et al. and Tsai, more prevalence studies have been performed worldwide. In the following, we report the prevalence studies published since 2014 according to the geographical area of reference. Some studies yield ASD prevalence estimate at different calendar year and/or in different age classes: where possible, we selected the more recent estimate of prevalence that referred to age 8. The list of studies with details and prevalence estimates are presented in Tables 2–4.



\*

to



*Brain Sci.* **2020**, *10*, 274

\* **Table 3.** Summary of prevalence studies published since 2014 in Asia, Australia &New Zealand.


\* Prevalence estimates and/or 95%CIs calculated from information reported in the original papers. Prevalence estimates in children aged less than 5 years are reported in italic.



### *2.1. Europe*

In Sweden, a prevalence study was performed in 2011 based on data from the Stockholm Youth Cohort (SYC) [16]. SYC is a record-linkage study collecting data longitudinally from 2001 to 2011 on all children from 0 to 17 years of age residing in Stockholm County in any time in the specified period; a multisource case ascertainment methodology was used to assess the presence of an ASD diagnosis. An overall large increase of ASD prevalence was observed between 2001 and 2011. Specifically, in children aged 0–17 years, the prevalence moved from 4.20/1000 in 2001 to 14.4/1000 in 2011, with an increase of almost 250%. This increase was mainly due to the huge increase of the ASD prevalence observed in children/adolescents without intellectual disability (almost +700%, from 1.40/1000 in 2001 to 11.0/1000 in 2011), while the increase of ASD prevalence in children/adolescents with intellectual disability was much lower (about +20%, from 2.80/1000 in 2001 to 3.40/1000 in 2011).

In Poland (West Pomeranian—WP—and Pomeranian—P—regions), Skonieczna-Zydecka and collaborators [17] estimated ASD prevalence in 2010–2014 on children from 0 to 16 years of age based on data obtained from both government and private institutions concerning ASD diagnoses and certificates of disability. The prevalence estimates in children of all ages were similar in the two regions (3.24 vs. 3.76/1000 in WP and P, respectively). In both regions, the highest prevalence was observed in children from 4 to 7 years of age (5.35 and 5.25/1000 in WP and P, respectively), yielding an overall estimate of 5.29/1000 in this age class.

In Germany, Bachmann et al. [18] conducted a study aimed at estimating at a national level the administrative prevalence of ASD in individuals aged up to 24 years, using inpatient and outpatient claims data of National health insurance from 2006 to 2012. The 2012 estimates were used to detect differences in prevalence among age groups. Children from 6 to 11 years of age showed the highest prevalence, estimated at 6.00/1000.

In the European Union, 14 countries have participated to the European project "Autism Spectrum Disorders in Europe (ASDEU)": Spain (programme lead), Austria, Belgium, Bulgaria, Denmark, Finland, France, Iceland, Ireland, Italy, Poland, Portugal, Romania, and United Kingdom. Among the goals of the project, there was the estimation of the prevalence of ASD in children aged 7–9 years in 2015. Four countries estimated the prevalence of ASD in 8 years old children using nationwide registry data (Denmark, Finland, and Iceland) or regional statistics (France); prevalence estimates were very different among countries, ranging from 4.76/1000 in South-Eastern France to 31.3/1000 in Iceland (for details, see Table 2) [19].

Eight countries (Austria, Bulgaria, Ireland, Italy, Poland, Portugal, Romania, and Spain) performed ad hoc studies following a shared protocol that required the participation of schools, teachers and parents. Teachers and parents were required to fill in questionnaires (Teacher's nomination and Social Communication Questionnaire, respectively) in order to screen children at risk of having ASD. The children at risk successively underwent a clinical assessment to confirm (or not) the diagnosis of ASD. Until now, only the results of the ASDEU ad hoc study performed in Italy have been published [20]. This study yielded a prevalence estimate of 7.99/1000 when using just the number of children certified with ASD or with other neurodevelopmental disorders in comorbidity with ASD. This prevalence rose to 10.4/1000 when including children identified through the screening procedure, and to 11.5/1000 based on a probabilistic calculation to adjust for non-responses. This estimate was much higher than those based on regional administrative databases storing data on services provided by Child and Adolescent Mental Health Units in Italy, namely SMAIL in Piemonte and ELEA in Emilia Romagna regions. These regional databases yielded in 2016 prevalence estimates of 4.20 and 4.30/1000 in children aged 6–10 years and 6.20/1000 and 5.50/1000 in children aged 3–5 years (Piemonte and Emilia-Romagna regions, respectively). A more recent regional estimate in 2018 based on administrative data from Abruzzo region yielded a higher prevalence estimate of 7.98/1000 in the age class 6–8 years, and quite similar prevalence estimate of 5.74/1000 in the age class 3–5 years [21]. These data suggest that prevalence estimates based on data extracted from registries built to meet administrative informative needs are on average lower than estimates coming from ad hoc studies,

mainly when a two-phase ascertainment design (screening and diagnosis confirmation) is used. UK data support this insight. The high prevalence observed in 2006 in South Thames, consistent with that estimated in Cambridgeshire in 2003–2004 by a school-based population study (15.7/1000) [43], was very different from the prevalence in children aged 8 years estimated by administrative data from the UK General Practice Research Database [13]. The database, activated in 1990 and storing medical records from the general practitioners, produced a much lower prevalence (from 3.58/1000 in 2004 to 4.09 and 3.90/1000 in 2009 and 2010, respectively), even lower than the median prevalence estimate reported by [7].

In Spain, a two-phase cross-sectional study in the framework of EPINED project was performed in Tarragona (year of study performance not specified), yielding prevalence estimates of 15.5/1000 and 10.0/1000 in the age classes 3–5 years and 10–12 years, respectively [22]. At about the same time, a study was performed using data from the Catalan Public Health Service on children aged 2 to 17 years. The estimated ASD prevalence in 6–10 years old children for 2017 was 11.8/1000, a rather high value for an estimate based on administrative data [23].

### *2.2. Middle-East*

Few studies have been performed up to now in Middle East countries, generally yielding prevalence estimates lower than Western Countries.

In Iran, the most updated estimate of ASD prevalence comes from a study that is part of a large-sample national population based epidemiological study concerning psychiatric disorders among Iranian children and adolescents aged 6–18 years [24]. The weighted ASD prevalence estimate for 6–18 years old subjects (computed from data reported in the paper) is approximately 1.60/1000, lower than less recent estimates from United Arab Emirates (2.90/1000 for 0–14 year children) [44], and Israel (4.80/1000 for 1–12 year children) [45].

As already reported in the review by [7], in 2010 a very low countrywide prevalence of 0.14/1000 had been estimated in children aged 0–14 years in the Sultanate of Oman [46]. This prevalence possibly reflected a low capacity of detecting children with ASD more than an actual low proportion of children affected. The lack of biological markers of ASD and the low availability of health services for the diagnosis of and the intervention on children with ASD were examined as factors that may account for the low prevalence [47]. More recently, Al-Mamri et al. performed a multicentre study aimed at updating the estimate of ASD prevalence among Omani children, using data retrieved from the three main centres for the diagnosis of ASD in the Sultanate of Oman in the period December 2011–December 2018 [25]. The new estimate was 2.04/1000 in the overall group of children (0–14 years of age); even if it is almost 15-fold higher than the previous one, it is still very low with respect to most of the estimates worldwide. Within the country, the highest prevalence was observed in Muscat (3.65/1000) with a prevalence in boys 3.4-fold higher than in girls (3.12/1000 vs. 0.91/1000, respectively).

Qatar is a country with a small population (2.7 million) characterized by a very high literacy rate, free and mandatory school attendance, and free healthcare for nationals and residents. A cross-sectional two-phase survey was conducted from 2015 to 2018 to estimate ASD prevalence in children aged 6 to 11 years [26]. The total prevalence (deriving from prevalence of already- and newly-diagnosed cases) was estimated at 11.4/1000, a value much higher than those observed in the other middle-east countries.

In Lebanon, a cross-sectional study was performed in 2014 in nurseries of Beirut and Mount Lebanon, to estimate ASD prevalence in toddlers aged 16–48 months using M-CHAT and a short structured questionnaire developed in the study [27]. Since it was not possible to conduct a follow-up interview to ascertain the M-CHAT results, the proportion of toddlers with a positive result at the M-CHAT was calculated, and corrected by an estimated positive predictive value, yielding a final ASD prevalence of 15.3/1000. This value is quite high and similar to the prevalence estimated in western countries.

### *2.3. Asia*

Qiu et al. [48] have published a systematic review and meta-analysis of studies on prevalence of ASD in South Asia (Sri Lanka, 2009; Bangladesh, 2009 and 2018; India, 2017; Nepal, 2018), East Asia (South Korea, 2011; China, 2011 and 2014), and West Asia (corresponding to the Middle East region: Iran, 2012; Israeli, 2013; Lebanon, 2016). Prevalence estimates show a very large variability across countries, ranging from very low values estimated in Iran, 2012 (0.63/1000) and Bangladesh, 2018 (0.76/1000), to low values estimated in India, 2017 (1.53/1000), China, 2011 (1.77/1000), India, 2017 (2.19/1000), China, 2014 (2.75/1000), Nepal, 2018 (3.42/1000), Israeli, 2013 (4.80/1000). On the contrary, large values were estimated in Bangladesh, 2009 (8.42/1000) and Sri Lanka, 2009 (10.7/1000), and very large values in Lebanon, 2016 (15.3/1000) and South Korea, 2011 (26.4/1000).

The Qiu's review did not include some recent studies performed in China [29–31], in Japan [32], and in India [33]. In China, Yang et al. (2015) [31] performed ASD assessment in 2014 in toddlers (3.8 to 4.8 years of age) who attended mainstream kindergarten in Shenzhen, estimating ASD prevalence at 26.2/1000. Sun et al. (2015) [29] evaluated ASD prevalence in children aged 6 to 11 years from two mainstream schools in Beijing, yielding an estimate of 11.9/1000. In 2019, Sun et al. [30] estimated ASD prevalence in three cities (Jilin, Shenzhen, and Jiamusi) at December 2013, using data from mainstream school only in all the cities and from the whole population in Jilin. Estimates based on mainstream school population were much lower than prevalence estimated in Beijing, ranging from 1.46/1000 in Jilin to 4.23/1000 in Shenzhen, the latter much lower than that estimated from toddlers. On the contrary, the estimate based on the overall population (from mainstream and special schools, private intervention centres, and community not attending school) in Jilin was 10.8/1000, nearer to the estimates from Beijing and from Western countries.

In Japan, a community sample survey was performed to estimate prevalence of neurodevelopmental disorders (NDD) and their co-occurrence in children aged 6–9 years, using questionnaires administered to parents and teachers [32]. The estimated prevalence of ASD, alone or in co-occurrence with other NDD, was 19.0/1000 based on parent's reports, and rose to 93.0/1000 based on teacher's reports. The latter was quite large, much larger than what observed in all other countries. In addition, the agreement rate between parent and teacher estimates was very low, suggesting that teacher's estimate could be largely overestimated and unreliable.

With regard to South Asia, Poovathinal et al. [33] performed a community-based survey in 2011–2012 in Kerala, South India. The study was part of a two-phase epidemiologic survey on chronic diseases performed on the entire regional population. The ASD prevalence in children from 6 to 10 years of age (i.e., the age class showing the highest prevalence) was estimated at 5.05/1000.

Finally, Hoang et al. [37] conducted a two-phase cross-sectional study (screening with M-CHAT and confirmation by clinical assessment) in toddlers from 18 to 30 months of age in Vietnam (Hanoi and Northern provinces). The estimated prevalence was 7.52/1000, obtained as proportion of children confirmed to have ASD on the number of children who underwent ASD assessment. However, the percentage of children undergoing ASD assessment after screening by M-CHAT was 100% in screen-positive children and 2% only in screen-negative ones. In addition, the percentage of ASD confirmation was 52.2% and 0.3% in screen-positive and screen-negative children, respectively. When taking into account the difference in the rate of ASD assessment following M-CHAT screening, and the difference in the rate of ASD confirmation between the two groups of children, the prevalence estimate rose to 10.8/1000, a value much higher than the values from previous studies in the same country, and much more similar to estimates in the Western countries.

### *2.4. Australia & New Zealand*

In Australia, Randall et al. published in 2016 a study performed within the Longitudinal Study of Australian Children (LSAC) framework [38]. Data on the parent-reported ASD diagnoses were collected for children belonging to two different cohorts, recruited in 2004 at birth (B-cohort, years of age 2004) and in kindergarten (K-cohort, years of birth 1999–2000). Data were obtained from two

different waves of the LSAC, referring to children aged 6–7 years in 2010–2011 (for B-cohort, wave 4) and in 2005–06 (for K-cohort, wave 2). Estimated ASD prevalence in 2005–2006 was 14.1/1000, and rose to 25.2/1000 in 2010–2011. Both prevalence estimates were much higher than the previous estimate of 3.92/1000 found by Icasiano et al. [49] in children aged 2–17 years during 2002, living in the Barwon region in Australia. It has to be noted that the estimate by Icasiano et al. refers to children and adolescents in a wider range of age, thus including diagnoses performed in different calendar years that are also affected by different capability of recognizing ASD. Secondly, researchers did not perform ad hoc case ascertainment, basing the estimate of prevalence on formal diagnoses of ASD made prior to data collection. As already reported for the use of data extracted from registries in UK, Italy, and China, prevalence estimates based on existing data are usually lower than estimates coming from ad hoc studies with active ascertainment of cases, and the gap is even larger with respect to estimates based on (often uncontrolled) parent-reported diagnoses (see USA below).

### *2.5. North America*

ASD prevalence estimates have been produced in three regions of Canada (Newfoundland and Labrador, NL; Prince Edward Island, PEI; Southeastern Ontario, SO) from 2003 to 2010, using data from the National Epidemiologic Database for the Study of Autism of Canada (NEDSAC) [39]. A general increase of prevalence across years was observed in the three regions, with large differences in prevalence among regions. In children aged 6–9 years, in NL region the prevalence increased from 5.20/1000 in 2003 to 10.8/1000 in 2008. In the PEI region, the prevalence passed from 5.88/1000 in 2003 to 6.13/1000 in 2008, and 9.99/1000 in 2010, and in SO region, from 8.34/1000 in 2003 to 12.4/1000 in 2008 and 16.2/1000 in 2010. Prevalence has been also estimated from2000 to 2015 using data from Quebec Integrated Chronic Disease Surveillance System (QICDSS) [40]. All residents in Quebec for at least one day from January 1, 1996, to March 31, 2015, and aged up to 24 years, were considered eligible for the prevalence study. Physician claims or hospital discharges from 2000 to 2015 reporting a diagnosis of ASD, Rett syndrome or childhood disintegrative disorder at ICD-9 or ICD-10, were used to classify the patient as having ASD. The lifetime prevalence for children aged 1 to 17 years was estimated at 1.50/1000 in 2000–2001, rising up to 12.2/1000 in 2014–2015. In general, a large variability in prevalence rates was observed among sub-areas, with higher prevalence in Montreal metropolitan area and lower in semi-urban and smaller regions.

In the USA, the Centers for the Disease Control and Prevention (CDC) launched in 2000 the Autism and Developmental Disabilities Monitoring (ADDM) Network, with the aim of tracking the number and characteristics of children with ASD in multiple communities in the United States. The ADDM Network is a multisite, multiple-source, record-based surveillance system, providing the most updated and comprehensive estimates of prevalence of ASD and other developmental disabilities in children aged 8 years; this age was chosen because of the peak ASD prevalence observed among elementary-school-aged children. Since 2010, the prevalence is estimated also in children aged 4 years. Prevalence estimates are given from 2000 and every two years (except for 2004); the most recent estimates refer to 2016 [12,41,50–55]. The ADDM Network program uses the systematic screening of databases/registries (related to health, service provision for developmental disabilities, special education) in order to extract information concerning behaviours possibly associated to developmental disorders, building a multi-information record for any child of the specific age class living in the reference geographical area. Information collected in the child's record is then examined to evaluate if the child can be diagnosed with ASD or other developmental disability, based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR). In 2014, 81% of the overall population underwent diagnostic evaluation by both DSM-IV-TR and DSM, Fifth Edition (DSM-5).

Tables 5 and 6 show the USA prevalence estimates in the overall examined populations of children aged 8 and 4 years, respectively, and in subgroups of children based on sex and Ethnicity.


*Brain Sci.* **2020**, *10*, 274



AL Alabama, AR Arkansas, AZ Arizona, FL Florida, MD Maryland, MO Missouri, NJ New Jersey, UT Utah, WV West Virginia.

As can be seen, in children aged 8 years the prevalence raised steadily from 6.60/1000 in 2002 to 14.7/1000 in 2010, remained constant from 2010 to 2012, and then raised again, arriving at 16.8/1000 in 2014 and 18.5/1000 in 2016, with an increase of 181% with respect to 2002. ASD prevalence also increased in children aged 4 years, passing from 13.4/1000 in 2010 to 17.0/1000 in 2014, but it decreased to 15.6/1000 in 2016.

In 2010, ASD prevalence was slightly lower in 4-years than in 8-years children (13.4 vs. 14.7/1000; this gap seemed to be bridged in 2014 (17.0 vs. 16.8/1000 in 4-years vs. 8-years children), suggesting an improvement in early diagnosis of ASD, but it appeared again in 2016 (15.6 vs. 18.5/1000 in 4-years vs. 8-years children).

The increase of prevalence from 2008 to 2016 corresponds to a variation in the distribution of ASD-diagnosed children with respect to the intellectual disability (ID; for details see Table 7. The proportion of ASD subjects with moderate ID remains constant across calendar years (24–25%), while the proportion of children with severe ID decreases (from 38 to 31–33%) and that of children without ID increases (from 38 to 42–46%). The prevalence increase with respect to 2008 in children grouped by IQ level suggests the hypothesis that a large part of the increase in the overall prevalence depends on the increase in children without ID, likely due to a greater ability to recognize children with milder forms of ASD (including high-functioning autism and Asperger's syndrome).

In 2018, Xu et al. reported an estimate of ASD prevalence of 24.7/1000 in children and adolescent aged 3–17 years in the 2014–2016 period, based on data from the National Health Interview Survey (NHIS), an annual health survey in the USA [56]. Similar prevalence was obtained by Kogan et al. [57] using data from the National Survey of Children's Health (NSCH) to estimate the national prevalence of parent-reported ASD diagnoses in US children from 3 to 17 years of age in 2016. The estimated value was 25.0/1000 in the overall group of children, and 26.1/1000 in children aged 6-11 years. Both NHIS and NSCH are nationwide surveys (the latter based on a larger sample than the former), and thus potentially representative of the whole country; however, data come from parents, who are asked to report if their targeted child was ever told to have ASD by a doctor or health professional. This introduces possible report biases with not quantifiable effects on the prevalence estimate: for this reason, data are not comparable with those coming from the ADDM Network surveys.


**Table 7.**

CDC-ADDM

 Network ASD prevalence estimates per 1000 children of 8 years of age, in the overall group and in IQ subgroups,

*Brain Sci.* **2020** , *10*, 274

 from 2008 to 2016 in USA,

In Mexico, a survey on children aged 8 years was performed in 2011-2012 in the city of Leon in Guanajuato [42]. Subjects enrolled in the study were students of regular (GSS) or special education (SEMR) schools. Children from GSS underwent a screening phase based on the Social Responsiveness Scale filled in by parents or teachers, and when the score passed the threshold, they were invited to undergo a diagnostic assessment. Based on these data, ASD prevalence was estimated to be 8.70/1000, lower than the prevalence estimated in USA in the same calendar year and quite similar to the USA estimate in 2006.

No prevalence estimates were found for States of Central or South America.

### **3. Factors Potentially A**ff**ecting Prevalence**

As reported above, Tables 2–4 summarize the studies published worldwide after the review by [15], concerning prevalence estimates in children and adolescents from 1 to 17 years of age (*n* = 42 studies). Prevalence estimates still vary across and within geographical areas, countries, year of study and source of data used in the study to estimate the prevalence. To detect the contribution of these factors on the variability observed in the prevalence estimates, we performed simple and multiple regression analyses. Specifically, studies were divided into two subgroups according to the age range of subjects (*Agerange*): age group 1 = age range including 7–8 years and/or lower limit of the age range above 5 years (*n* = 36); age group 2 = upper limit of the age range up to 5 years (*n* = 6). Prevalence was the dependent variable, while geographical area (*Area*: America, Asia, Australia, Europe, and Middle East), source of data (*Source*: administrative data coming from one or multisource databases, ad hoc study, and report), *Agerange* and year of study performance (*Year*) were the independent variables. As for *Year*, some studies report estimates referring to one specific calendar year (*n* = 28), in others estimates refer to periods of two (*n* = 7) or four or more years (*n* = 3), and others do not specify the calendar year of study performance (*n* = 4). When more than one year was indicated, we imputed the most recent year of the interval, and when no year was reported we imputed the year before that of study publication. Europe was used as reference level for *Area*, administrative data for *Source*, and age range including 7 and/or 8 years for *Agerange*. Since more studies could be performed in a single country within a geographical area, *Country* (e.g., Italy, France, within Europe area; Oman, Iran within Middle East area, etc.) was considered as clustering factor. The effect of *Year of study* on the prevalence estimate was evaluated in the simple regression. Since, as reported above, *Year* could not be determined with sufficient precision in the 17% of studies, and the effect of *Year* on prevalence estimates in the simple regression analysis was not significant, we did not include this variable in the multiple regression analysis. Variance inflation factors (VIF) were computed for any variable included in the multiple regression model; all VIF values were lower than 5, thus supporting the absence of multicollinearity. The results of regression analyses are presented in Table 8.

The regression analyses present obvious limitations, due to the low number of studies (*n* = 42), especially in relation to the large number of combinations of area, source of data, and age of children levels (5 × 3 × 2 = 30 different combinations), making it difficult to disentangle the effects of the different factors. However, we can draw some indication on the potential explanatory factors affecting prevalence estimates.

From the simple regression analyses a significant difference among *Areas* was observed, with Europe showing significantly lower prevalence with respect to Australia (*p* < 0.001). Prevalence estimated based on parents' or teachers' reports was significantly higher than prevalence estimated by administrative data (*p* = 0.044), while neither *Agerange* nor *Year* of study performance significantly affected prevalence estimates.


Multiple regression model (see Table 8) confirms that studies that use parents' and teachers' report predict higher prevalence in respect with administrative data (*p* < 0.001). On the contrary, since the estimates from Asia were mainly based on ad hoc studies, and those from Australia were both based on reports, when accounting for the source of data prevalence estimates from both areas turned to be significantly lower than those from Europe (*p* = 0.029 and *p* < 0.001 for Asia and Australia, respectively). Finally, estimates in younger children turned out to be significantly higher than estimates in older subjects (*p* = 0.007). Overall, the factors included in the multiple regression model explained about 54% of the variance in prevalence estimates (*R*<sup>2</sup> = 0.5314), notwithstanding that the number of independent variables in the multiple regression model (*n* = 7) was high with respect to the number of studies included (*n* = 42). This suggests the need to investigate other variables, likely related to exposure to different risk factors for autism, in order to explain the observed variability.

### **4. Discussion**

The analysis of the literature on ASD prevalence studies published since 2014 confirms a high variability of prevalence estimates worldwide. This variability is still accompanied by methodological differences among the performed studies that concern how cases are detected, which population is involved in, and, to a lesser measure, how cases are defined.

Interestingly, the longitudinal analysis of data across years within the same geographical area confirms the increase of prevalence estimates that has repeatedly drawn scientists' attention in the last twenty years [1]. Studies from Australia [38], Canada [39], Oman [24,25], and USA (see Tables 5 and 6) and some European countries (Sweden, [16]; Italy, [20]) show a substantial increase of ASD prevalence estimates over the years especially at the turn of the 2010. However, the consistency of the increase over countries is masked by the high variability of the prevalence estimates (see Tables 2–4) over the continents, with a range from 0.8/1000 in the North, Sirajganj district of Bangladesh to 93/1000 in Japan.

As previously reported, one of the putative methodological issues contributing to the high variability of ASD prevalence is the source of data from which ASD cases have been detected. From the present analysis, it emerges that the main sources of ASD cases are administrative data (mono or multisource), ad hoc studies, and surveys based on questionnaires. The simple and multiple regression analyses show that the source of data indeed affects the estimate of ASD prevalence. In particular, when ASD cases are detected by teachers' or parents report, prevalence estimated seem to be significantly overestimated. Otherwise, one or more phased population studies appear to produce higher prevalence estimate then studies based on administrative data, but the difference apparently is not significant.

As previously evidenced [58], it is possible that the methodological and qualitative advantages and disadvantages of the use of different sources of data, make it difficult to choose a specific surveillance policy about the count of ASD cases. Population-based designs are considered a high research standard because they are representative of all children in defined populations who meet selected ASD criteria and are evaluated in "natural" community settings, rather than of selected samples attending a particular setting (clinic or educational) or registered in specific research projects. However, the source of identification (teacher; parent; professionals), the lack of blindness of the assessor, the multi-phasing of the study (in relation to the sensitivity of the screening tools and the specificity of the confirmation diagnostic tools) as well as the sample setting (e.g., mainstream vs. special school [29]), and the case definition [59], all represent potential biases that may affect the prevalence estimate obtained by population studies. Otherwise, as above stated, estimates based on administrative classifications have other limitations, due to either difference between states in administrative policies and regulations for the access to the system of recording [60], and/or to socioeconomic disparities or different services availability over the countries [8,9]. Finally, as noted by some scholars, in the survey-based prevalence studies, the formulation of the questionnaire or interview to be administered to parents or teachers can influence the understanding of the questions asked [58] also taking into account educational and/or cultural factors. Furthermore, recall-bias is an intrinsic limitation of this kind of study.

Results of multiple regression also highlight differences in prevalence due to the geographical area where the study is performed, with Europe showing significantly lower prevalence than Australia and Asia. As argued above, this result appears to be at least partially due to the source of data used in the study, but it can indeed be due to several determinants such as socio-cultural [61,62] and socio-economic factors [63], including organisational factors [37,64].

As seen above, factors such as case definition and case-finding procedures, and geographical area, however, scholars reported that they appear to account only for about a 50% of the variability, thus suggesting that additional factors linked to the aetiology of ASD should be considered in explaining variability of ASD prevalence across areas and over time [65–67]. Current literature suggests that several environmental factors could affect brain development and differentiation over perinatal period resulting in neurodevelopmental disorders emerging at different time life. These studies overall focus on dynamic interactions between biological and non-biological risk factors [68]. As for ASD, CHARGE (Childhood Autism Risks from Genetics and the Environment) study is an excellent example of epidemiological study contributing to the understanding of which factors can increase the risk of ASD. Three groups of children have been enrolled in the CHARGE study: children with autism, children with developmental delay but without autism and children from the general population. All of them are evaluated for a broad array of exposures and susceptibilities [69]. Among evidences obtained through the analysis of data collected by the ASD group of children, folate prenatal intake, maternal fever, pesticides exposure, and air pollution, seem to be associated with an ASD risk increase [70]. However, CHARGE study adopts a retrospective case-control approach that ranks in the lower level of the pyramid evidence. Other evidence came from cohort studies that highlighted others possible risk factor such as parental age at birth [71]. Furthermore, some maternal factors (i.e., maternal age, pregnancy and delivery condition, drug intake, maternal autoimmunity, inflammation and chronic stress) are of increasing concern and suggest the need of further studies [72].

In conclusion, multiple and complementary systems are needed to better estimate ASD prevalence and to understand its observed changes. It is necessary to establish either surveillance systems in order to monitor the change of prevalence with time, or guidelines for the performance of ad hoc studies to compare the prevalence across geographical areas. Although the reliability of the prevalence estimates coming from the ADDM Network has been questioned [73], until now this is the only surveillance systems that tracks ASD prevalence over the years and across states, allowing to study factors that possibly give reasons for the observed prevalence increase. Finally, methodological differences across studies could not fully account for the large variation among the prevalence estimates. This suggests the need tostudy other factors, pertaining to the capability to recognize and diagnose ASD and/or to the exposure to genetic and environmental risk factors for ASD, in order to explain the prevalence variation.

**Funding:** This research received no external funding.

**Acknowledgments:** We thank Gemma Calamandrei and Laura Ricceri for critical reading of the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Systematic Review of Level 1 and Level 2 Screening Tools for Autism Spectrum Disorders in Toddlers**

**Serena Petrocchi 1,2,3,\*,**†**, Annalisa Levante 2,4,**† **and Flavia Lecciso 2,4**


Received: 28 February 2020; Accepted: 17 March 2020; Published: 19 March 2020

**Abstract:** The present study provides a systematic review of level 1 and level 2 screening tools for the early detection of autism under 24 months of age and an evaluation of the psychometric and measurement properties of their studies. Methods: Seven databases (e.g., Scopus, EBSCOhost Research Database) were screened and experts in the autism spectrum disorders (ASD) field were questioned; Preferred Reporting Items for Systematic review and Meta-Analysis (PRISMA) guidelines and Consensus-based Standard for the selection of health Measurement INstruments (COSMIN) checklist were applied. Results: the study included 52 papers and 16 measures; most of them were questionnaires, and the Modified-CHecklist for Autism in Toddler (M-CHAT) was the most extensively tested. The measures' strengths (analytical evaluation of methodological quality according to COSMIN) and limitations (in term of Negative Predictive Value, Positive Predictive Value, sensitivity, and specificity) were described; the quality of the studies, assessed with the application of the COSMIN checklist, highlighted the necessity of further validation studies for all the measures. According to COSMIN results, the M-CHAT, First Years Inventory (FYI), and Quantitative-CHecklist for Autism in Toddler (Q-CHAT) seem to be promising measures that may be applied systematically by health professionals in the future.

**Keywords:** autism; level 1 and level 2 screening tools; systematic review; COSMIN; PRISMA

### **1. Introduction**

Recently, U.S. data showed that the median age at earliest Autism Spectrum Disorders (ASD; [1]) diagnosis ranged from 28 to 39 months for children aged 4 [2] and is 40 months for children aged 8 [3]. According to these data, a screening procedure during the regular well-baby check-ups was recommended [2,4] with the aim to detect the warning signs of ASD (e.g., precursors of Theory of Mind; [5]). As suggested by several authors [6,7], the process should involve the early screening of warning signs and the subsequent diagnosis made through clinical judgement, in combination with the application of reliable and standardized gold-standard measures (e.g., the Autism Diagnostic Interview-Revised, [8]; the Autism Diagnostic Observative Schedule-2, [9]).

Earlier diagnosis of ASD could lead to earlier intervention for children, which could enhance their adaptation [10–12] or improve their social competence (e.g., emotional expression; see for details [13–15], prevent secondary developmental disturbances [16], and lead to better outcomes [17–19]. Screening measures that are suitable for use in young children (i.e., less than 24 months) are available, and can be classified as either Level 1 or Level 2 instruments [19]. Level 1 screening measures have been developed for the general population (unselected population) to identify children at risk of developmental disorders, including ASD. Level 2 screening tools have been developed to identify children at risk of ASD either because they are already under observation for developmental concerns, or because they failed Level 1 screening, or because they are siblings of children with ASD. The latter, as demonstrated for example by Lauritsen and colleagues [20], have a strong genetic risk. As Robins and Dumont-Mathieu [19] noted, several measures, developed for level 1 or level 2 screening, have been applied to other populations, determining a "hybrid" application of them. The present systematic review focuses on level 1 and level 2 screening measures of ASD, which can be administered to GPs and/or to parents or other professional groups (e.g., nurses, social workers).

In the last few years, eight reviews [21–28] have examined measures for the early detection of risk of ASD. Daniels and colleagues [21] focused on studies investigating approaches aiming at improving the early detection of ASD. This was a systematic review using five databases, although the authors chose to include studies limited to the United States. Garcia-Primo and colleagues [22] and Sappok and colleagues [24] conducted non-systematic reviews considering both measures for the early detection of risk for ASD and for diagnosis. The study by Garcia-Primo and colleagues [22] was limited to measures applied in Europe, published up to 2012, and the search was restricted to two databases (PubMED and PsycINFO). The review by Sappok and colleagues [24] was limited to one database (PubMED) and it considered measures developed for German and English speakers. Zwaigenbaum and colleagues' review [25] was limited to one database (PubMED) and the research strategy included papers published up to 2013. McPheeters and colleagues [23] made a valuable systematic review of the ASD screening tools for children who were referred for developmental disorders other than ASD and were under 36 months old.

Nevertheless, their search strategy included four databases and they considered studies published up to 2000. Marlow and colleagues [26] carried out a systematic review extracted data from four databases and included papers published up to 2017. The meta-analysis by Sánchez-García and colleagues [27] evaluated the accuracy of screening measures according only to their sensitivity, specificity, positive, and negative predictive values (PPV and NPV respectively); furthermore, their electronic search was limited to 5 databases and included paper published up to 2015. Finally, the review by Thabtah and Peebles [28] provided a no systematic review on screening tools administrable from toddlerhood to adulthood, but the authors did not report the search strategy (i.e., databased searched; range of publication years considered) applied and described the tools only in terms of sensitivity and specificity.

Summarizing, most of the above-mentioned reviews are not systematic [22,24,25,28], have limited search strategies to 1–5 databases [22–24,26,27], or focus on a specific geographic area as Europe or USA [21,22,24]. Furthermore, they did not analyze the psychometric and measurement properties of the measures with the exception of Sánchez- García and colleagues [27] meta-analysis which applied the Bayesian Hierarchical Model to evaluate some psychometric properties associated to accuracy. Overall, researchers cannot derive considerations regarding the methodological quality of the studies.

To overcome the limitations of the previous reviews, we provided a systematic search on level 1 and 2 screening tools for ASD and an evaluation of their psychometric properties according to the COSMIN checklist [29,30]. The COSMIN checklist is a 'standardized tool for assessing the methodological quality of studies on measurement properties' [31] developed based on a Delphi study which is a standardized.

The specific research questions were: (RQ1) What are the level 1 and level 2 screening measures to detect early signs of risk of ASD in children under 24 months of age? (RQ2) What are the psychometric properties of the studies of Level 1 and Level 2 measures and what is their quality evaluated applying the COSMIN checklist? (RQ3) Is there one (or more) promising instrument(s) for the early detection of risk of ASD according to COSMIN results?

To give the reader a full and comprehensive view of the characteristics of the Level 1 and Level 2 measures available, and since the COSMIN protocol evaluates the quality of the study, but not the quality of the tool, we collected data on sensitivity, specificity, PPV, and NPV for all the included measures and we provided a discussion about those properties.

### **2. Materials and Methods**

The systematic review is based on a published protocol [32], in which the authors reported a comprehensive description of the steps to follow, the methodology, and the process of the review. Furthermore, the authors provided the format of the tables to be used for the main descriptive data of the papers included in the review and the results of the examination of the psychometric properties. The methodology applied was developed based on the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines [33] for identifying the papers to be included in the review. An electronic search was conducted using PsychINFO, the Psychology and Behavioral Sciences Collection, Cumulative Index of Nursing and Allied Health Literature, Scopus, the Education Resources Information Center, Google Scholar, and Pubmed (including MEDical Literature Analysis and Retrieval System OnLINE). The keywords applied were: 'early diagnosis or diagnos \*', 'ASD screen \*', 'ASD detect \*', 'ASD or autism or autist \*', 'assessment tool', 'surveillance', 'develop \* surveillance', 'assess \*', 'instrument \*', 'measure \*', 'psychometric properties', 'standardiz \*', 'tool\*', and 'validat \*'. A secondary hand search was performed to include references and citations from the identified papers. The electronic search was carried out by an author who extracted the records and tabulated the references in an excel file. Two authors independently screened the records to exclude duplicates and to remove papers according to pre-defined inclusion/exclusion criteria. The two authors reported their decisions in two different excel files and they compared their findings record by record. In case of disagreement, a third author arbitrated. Finally, three clinicians and three research experts in ASD, working respectively for the Public Health Service and for Universities respectively, were questioned. Based on the inclusion/exclusion criteria, they did not suggest any other relevant existing measure/study different from those already included in the present review.

Predefined inclusion criteria were: (1) level 1 and level 2 screening measures of ASD for children under 24 months; (2) validation studies, standardization of measures, cross-cultural comparisons, longitudinal, or follow-up studies; (3) published papers in peer- reviewed journals; (4) papers written in English; and 6) a year of publication between 1990 and October 2019. Other reviews on the same topic were examined to extract citations of studies that were eligible for our final list. Furthermore, exclusion criteria were defined as following: (1) measures of the diagnosis of ASD; (2) retrospective studies and systematic reviews; (3) measures of risk detection/diagnosis of others developmental disorders; (4) procedures for the detection of ASD other than questionnaires, interviews and observation procedures (i.e., biological markers, fMRI, blood test); (5) epidemiological studies and guidelines for experts; (6) publications that are not in peer-reviewed journals; (7) papers without the specific aim to evaluate psychometric properties or validity properties of the measures; (8) dissertation thesis or conference papers.

The evaluation of the measures applied the COnsensus-based Standards for the selection of health Measurement INstrument (COSMIN) checklist [29–31]. The COSMIN checklist applies nine boxes identifying the main measurement properties: (A) internal consistency (i.e., the degree to which the items of a questionnaire correlate with each other and evaluate the same concept); (B) reliability (i.e., the ability to measure a construct over time or by different persons); (C) measurement error (i.e., the error of the score not attributed to true changes in the construct); (D) content validity (i.e., the degree to which the items reflect adequately the construct measured); (E) structural validity (i.e., evaluating whether the hypothesized latent factor(s) reaches a good fit of the data); (F) hypothesis testing (i.e., considering whether the construct measured by the questionnaire reaches the expected relations with other variables); (G) cross-cultural validity (i.e., giving information on the generalization properties of the measure when applied in a different cultural context); (H) criterion validity (i.e., the degree to which

the measure correlates with a 'gold-standard' measure); and (I) responsiveness (i.e., evaluating whether the measure predicts a change over time). Each box contains a different number of items (ranging from 5 to 18) evaluating 'design aspects and statistical methods' of a study [31] (p. 651), which require a mandatory assessment to obtain a full appraisal of the properties.

The COSMIN checklist provides a multi-step evaluation. The first step concerns the decision about which measurement properties have been assessed in a target paper among the nine boxes, and it is achieved by applying a binary scale (i.e., present vs. absent) considering the whole paper. For example, if the internal consistency (i.e., box A) is a property evaluated in a paper, then 'present' is attributed to box A for that paper.

The second step refines the evaluation undertaken in step 1. For each box marked as 'present' in step 1, the evaluator works through the questions, assigning to each of them an evaluation on a dichotomous scale ('yes' if the specific properties suggested by the question are present or 'no' if the specific properties suggested by the question are not present).

Finally, in the third step, the score obtained in step 2 is further refined. Every item marked as 'yes' in the previous step is now evaluated on a four-point Likert scale: excellent (+++), good (++), moderate (+), or poor (0).

A final evaluation for each box is obtained by considering the lowest score attributed to that box according to the worst score counts [31] (p. 651) procedure. Therefore, if even only one item in the box obtained a poor score, the measurement property for that box is rated as poor. Two authors independently applied the COSMIN checklist on 20 papers with an inter-rater agreement of Cohen's k = 0.94.

### **3. Results**

### *3.1. Overview of the Studies and Measures*

Figure 1 shows the PRISMA diagram.

The electronic search allowed to identify 691 records and a second-hand search added 26 more records. According to the inter-raters decision-making process, during the screening, two authors independently removed 365 duplicates and 300 papers according to the exclusion criteria. The final eligible number of papers included in the systematic review was 52 ([34–85]. The consistency between the two authors who screened these records was high (Cohen's k = 0.89). Sixteen measures were evaluated and classified into 3 categories: observational checklists (*n* = 4), questionnaires (*n* = 10), and interviews (*n* = 2). Table 1 reports the general details of each measure.

Table 2 showed the details of the studies included in the systematic review. Specifically, we reported the measure name, authors and year of the study, the type of the design, population recruited, the application level (1, 2, or "hybrid"), and the diagnostic accuracy properties (i.e., sensitivity, specificity, PPV, NPV).

We found six level 1 measures (i.e., the Checklist for Early Signs of Developmental Disorders; the Early Screening of Autistic Traits Questionnaire; the First Year Inventor; the Joint Attention OBServation; the Screening for Infants with Developmental Deficits and/or Autism; the Young Autism and other developmental disorders CHeckup Tool: 18- month-olds' version) administered to the general population retrieved in 6 longitudinal studies and 3 cross-sectional studies.

**Figure 1.** PRISMA flow diagram.

### *3.2. Overview of the Studies and Measures*

The search strategy allowed to find four level 2 measures (i.e., the Autism Detection in Early Childhood; the Autism Observation Scale for Infants; the Baby and Infants Screen for Children with aUtIsm Traits; the Parent Observation of Early Markers Scale) that were also retrieved from the systematic search evaluated in eleven studies with a cross-sectional design and in two studies with a longitudinal design. Those measures were administered to two groups of children. The first group consisted of children who were already receiving attention from the local mental health service due to developmental concerns, children suspected of developmental delay, or children qualified for a medical condition that could determine a developmental delay including ASD comorbidity (i.e., epilepsy, hydrocephaly, Down's syndrome, and cerebral palsy). Henceforth this group is identified as Developmental Concerns group (DC). The second group included twins or younger siblings of children with an ASD diagnosis, henceforth defined as Genetic Risk group (GR) because they have high probability to develop ASD [20]. The studies included in level 2 aimed either to: (a) test a screening measure on DC or GR groups; (b) compare DC and GR groups between them; (c) follow DC/GR group until the diagnosis; or, finally, (d) compare children from the general population to DC or GR groups.

Table 2 shows also the details of the six 'hybrid' measures (i.e., the CHecklist for Autism in Toddlers; the Developmental Behavior Checklist: Early Screen; the Modified Checklist for Autism in Toddlers; the Modified Checklist for Autism in Toddlers-Revised with Follow-up; the Quantitative-CHecklist for Autism in Toddlers; the Three-Item Direct Observation Screen) that were developed mainly for level 1 and/or level 2 screening, but they were also administered to clinical populations (i.e., children who had already received a diagnosis of ASD or of another developmental disorder). Those studies aimed either to: (a) apply the measure to a clinical sample, (b) compare samples with different diagnoses (ASD vs. PDD-NOS vs. ODD), or, finally, (c) compare children from the general population with children with an ASD diagnosis. Eleven studies were longitudinal and 19 had cross-sectional design.


Level 1 and 2 screening tools include in systematic review.






*Brain Sci.* **2020** , *10*, 180







**Table2.***Cont.*

Pervasive and 2 screening measure applied to other population (e.g., clinical sample); FUI: follow-up interview; Sens =

(Negative Predictive Value); N/A = not available. \* Authors reported PPV and NPV values from [36] study.

Developmental

 Disorder; PDD-NOS =

pervasive

developmental

 disorder—not

 otherwise specified; TD typically developing children; "hybrid" level of application = level 1

sensitivity; Spec =

specificity; PPV(NPV) = Positive Predictive Value

The sensitivity, specificity, PPV, and NPV of the measures are extensively reported in the validation studies of the M-CHAT, M-CHAT R/F, and ADEC. For other measures (i.e., CESDD, JA-OBS, POEMS, DBC-ES, and TIDOS) there is only one study, each containing information of the NPV and PPV. All the other measures did not report any positive or negative predictive values. Overall considered, the measures for which the PPVs and NPVs were reported, demonstrated from moderate to high predictive values, although for the M-CHAT results can be considered more stable compared to other measures that need further and deeper exploration of these properties. Quality of assessment of the studies

Table 3 shows the results of the evaluation of each psychometric properties of the studies through the application of the COSMIN checklist. For each box, we reported a summary of the assigned scores.

The quality of assessment revealed a heterogeneous picture. Specifically, 24 studies out of 52 received an evaluation of the internal consistency (Box A) and the scores were fair or poor, with the exception of the studies on FYI and the Q-CHAT, which received excellent scores. The reliability (Box B) was evaluated in 17 studies and the majority of the scores rating from fair to poor. Only studies considering the CHAT and POEMS received respectively an excellent and good evaluation. The measurement error (Box C) was assessed in 5 longitudinal studies and received poor or fair evaluations.

The Box D (i.e., content validity) was evaluated in 9 studies and it received excellent evaluations for studies considering AOSI, BISCUIT, CHAT, FYI, M-CHAT, POEMS, Q-CHAT, SEEK, and TIDOS. Structural validity (Box E) was evaluated in 7 studies, but only 3 received excellent scores regarding two measures (i.e., M-CHAT and Q- CHAT). The Hypothesis testing (Box F) was evaluated for several studies, which received fair or poor scores, whereas those on FYI and the M-CHAT-R/F received good evaluations, and that on M-CHAT was evaluated as excellent. For the studies on JA-OBS, the SEEK, and the YATCH-18 the property was not evaluated.

The cross-cultural validity (Box G) was examined in 11 studies and received fair or poor scores. The box criterion validity (H) was evaluated for all studies, with the exception of the one on Q-CHAT and one on SEEK. This property was rated as excellent or good in four studies for four measures (FYI, M-CHAT, M-CHAT-R/F, and Q-CHAT); whereas for all other studies it was evaluated as fair or poor. Finally, the responsiveness (Box I) was the least-evaluated property with only 3 studies receiving scores from fair to poor.

As Table 3 shows the reasons leading to the attribution of fair and poor scores are above all the missing data and the sample size criteria and the fact that they are evaluated across several measurement properties. These criteria were evaluated by the COSMIN with a conservative approach [86], which will be discussed in the following section.












**Table 3.***Cont.* Note: 4-point scale rating: +++ = excellent, ++ = good, + = fair, 0 = poor. Empty cell = COSMIN rating not evaluated. Ratings fair and poor were explained with the reason(s) in italicsleading the evaluation. Specifically, "administration not similar" means the two administration conditions to examine measure property were not similar; "comparator instrument"means that authors did not administered a gold standard measure for ASD to evaluate the criterion validity;"expertise translator" means that the expertise of measure translators waspoor or not described by authors; "hypothesis" means that the authors did not formulate the hypothesis a priori; "missing item" means that the authors did not report the percentageand/or the handling method for missing data; "no pilot study" means the translated measure did not pre-tested in a target population; "only one measurement" means the authors didnot administered the measure at least two times; "sample" means that the sample size was not adequate;"statistical method" means that authors did not calculated the right parameter(s)for the specifc property;"time interval" means that the time interval between two measurements was not adequate;"translation" means that the back-translation process was notadequately described; "unidimensionality" means that the internal consistency parameter was not calculated for each (sub)scale separately [29,30].

### **4. Discussion**

The systematic review identified six level 1 measures and four level 2 measures. Moreover, the present systematic review found that six screening tools were applied to clinical populations. Among the variety of methodologies of the level 1 and level 2 measures, the questionnaire was the most applied due to several inherent advantages. First, questionnaires are normally administered in a very short time, do not require specific knowledge or training, and are much less invasive than observational checklists or interviews. Second, they often do not require specific training on the coding system or the interpretation of the scores. For many questionnaires, the imputation of a final score and the attribution of a meaning to it do not involve any clinical interpretation or specific knowledge of ASD. Nevertheless, questionnaires have several limitations. First, the score depends on the subjectivity of the informants. Since questionnaires are designed for parents, they could under- or overestimate the early signs of risk based on their ability to detect them and to distinguish signs of risk from normal deviation from the developmental trajectories. However, the impact of this limitation could be minimized with longitudinal studies testing and comparing the level 1 and level 2 screening instruments with the goldstandard measures (e.g., Autism Diagnostic Observation Schedule-2, [8]) for the diagnosis of ASD. Another inherent limitation of the questionnaires is social desirability bias in the form of over-reporting desirable behaviors. Future research in this field is needed to develop one or more validity scales, as for other clinical psychological testing procedures (i.e., the MMPI-2; see [87]).

The second aim of the present review was to evaluate the psychometric characteristics of the included measures following the COSMIN checklist. Two main considerations could be drawn by our results, one pertaining to the quantity of the psychometric evaluations and the other to their quality. First, it should be noticed that in the studies included in our systematic review, there are several psychometric properties more frequently evaluated than others. A high number of studies contained data that allowed the evaluation of the internal consistency, reliability, hypothesis testing, and criterion validity; whereas the measurement error, content validity, structural validity, cross-cultural validity, and responsiveness have been evaluated in a low number of studies. The second element to be considered is the quality of the evaluations themselves. Indeed, a high frequency of evaluations of a given property not always corresponds to a high quality of evaluation of that property. For example, the content validity was the property less frequently assessed, compared to the others, but it was rated as excellent for all the studies examined. On the other side, the hypothesis testing was frequently evaluated, but received poor or fair scores. These findings should give an impetus to researchers to design validation studies with a focus on both the quantity of the properties and their quality.

Considered overall, one very common problem for all the studies is the treatment of missing data. Few authors explicitly quantified the missing data in their data set, and very few explained the method that they followed to treat missing data. For studies that aim to identify early signs of risk of ASD, the treatment of the missing data represents a crucial aspect. For this specific case, the imputation of data through statistical procedures risks altering the data structure and the distribution beyond the over-/underestimation of the risk of ASD. Thus, it is quite important that, in the future, researchers explain whether and how they have treated missing data in their sample, especially for the parent-reported measures, for which it is more likely to have items with no answers.

According to the COSMIN evaluation, our findings highlight the necessity of further validation studies for all the measures included in the present review. Longitudinal studies involving general population following a sample over time with the purpose of making a diagnostic evaluation are particularly needed. This will allow for an in-depth study the psychometric properties, to compare the results from different measures, and consequently to increase their criterion validity, and specifically the sensitivity and the specificity through the comparisons with the gold standard measures.

Special consideration had to be drawn regarding the Sensitivity, Specificity, PPV, and NPV of the measures because they are not included in the COSMIN checklist. These properties are extensively reported in the validation studies of the M-CHAT, M-CHAT R/F, and ADEC. For other measures (i.e., CESDD, JA-OBS, POEMS, DBC-ES, and TIDOS) there is only one study each containing information

of these properties (see Table 2 for the specific values). All the other measures did not report any positive or negative predictive values. Overall considered, the measures for which the Sensitivity, Specificity, PPVs and NPVs were reported, demonstrated from moderate to high predictive values (see also [27]), although for the M-CHAT results can be considered more stable compared to other measures that need further and deeper exploration of these properties.

The third and final research question aimed at the identification of one (or more) promising instrument(s) for the assessment of early signs of risk of ASD according to the COSMIN evaluations of the studies. We consider the questionnaires such as the FYI, the M-CHAT, and the Q-CHAT as promising screening measures because, according to the COSMIN evaluation, they have high number of psychometric properties evaluated and high methodological quality attributed to them. Although we found these measures promising, none of them can be currently considered as the gold standard in the early detection of risk of ASD and further development in this field is desirable. For example, future studies should improve sensitivity, specificity, NPV, and PPV properties of those measures since they are not considered at all for the FYI and they are barely considered for M-CHAT and Q-CHAT, as also suggested by [27].

On the contrary, the interviews and the observational checklists have both low number of validation studies (with the exception of the M-CHAT-R/F) and low methodological quality attributed to them. Further research should be developed on these methods of evaluation focusing on their psychometric properties, as it may be useful for health professionals to have a range of tools available for ASD risk detection that allows an in-depth analysis.

The present systematic review has several limitations. First, the COSMIN checklist is a standardized protocol for the assessment of the methodological quality of a study and not of the instrument itself. However, as suggested by others [see 86] the evaluation of the methodological quality of a study is the first step to determining whether its results are reliable and trustworthy. In other words, evaluating the methodological quality of a study allows to discover risk of bias in the results. Thus, the assessment of the quality of the study is directly related to the assessment of the measure administered in that study. Moreover, one of our inclusion criteria considered all the "validation studies, standardization of measures, cross-cultural comparisons, longitudinal, or follow-up studies", which are studies evaluating measurement and validity properties of a screening measure. Therefore, we applied the COSMIN checklist to evaluate measurement properties of studies that, in turn, evaluate the measurement properties of the screening measures. Thus, the evaluation of the properties of a study, in this case, is a proxy of the evaluation of the measure validated in that study.

Second, the worse score counts policy of the COSMIN could lead to a negatively biased view of the measure. In this vein, the COSMIN itself explains that every item of its evaluation represents an important part of the overall assessment, so a poor rating for any item should be considered as a serious flaw. Furthermore, we would like to focus on the COSMIN evaluation of the sample size. According to [31], the sample size is evaluated as excellent when it is ≥ 100, as good when it ranges 50-99, fair when it ranges 30-49 is fair, and poor when it is < 30. This categorization is a good criterion when applied to the general population, while when risk and/or clinical groups are considered, the COSMIN sample evaluation should be carefully considered according to the prevalence rate of ASD. According to this premise, recently, the researcher who developed the COSMIN protocol reformulated the evaluation of the sample size (see [86]).

Third, the Sensitivity, Specificity, Positive Predictive Value (PPV) and the Negative Predictive Value (NPV) are not evaluated in the COSMIN checklist. Within the context of screening measures for ASD, it is important that professionals are confident when using a given tool. In this field, the predictive values provide valuable information on the probability of a tool to identify that people with high scores indeed have high risk (PPV) and, vice versa, that people with low score have low risk (NPV). To avoid the omission of such important information, we extracted values of the NPVs and PPVs from the studies, we reported them in Table 2 and we discussed the evidence.

Finally, like every systematic review, the definition of inclusion criteria could have limited the electronic search, and we could have omitted several studies.

The present systematic review has two main strengths. First, the review provides an updated and complete overview of the current level 1 and level 2 screening measures for ASD. Second, our findings provide researchers and clinicians (i.e., pediatricians, GP, psychologist) the analytical knowledge on psychometric properties of the measures through the evaluation of the methodological quality of their validation studies. The outcomes of the systematic search and the results of the evaluation of the psychometric properties, through the application of the COSMIN criteria, may guide researchers and clinicians in their selection of one (or more) instrument(s), according to their specific purposes. A critical and reasoned choice of a measure combined with the good communication between clinical and patients [88] could allow for defining systematic screening procedure on general population. This is the first step for early identification of risk of ASD, which, in turn, may lead to a timely diagnosis and ultimately to better outcomes for children [10,17,18] and families [89].

**Author Contributions:** Conceptualization, F.L. and S.P.; methodology, S.P. and A.L.; software, A.L.; investigation, A.L.; data curation, S.P. and A.L.; writing—original draft preparation, S.P.; writing—review and editing, S.P. and A.L.; supervision, F.L.; project administration, F.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Review*
