**Effect of Pre-Existing Sarcopenia on Oncological Outcomes for Oral Cavity Squamous Cell Carcinoma Undergoing Curative Surgery: A Propensity Score-Matched, Nationwide,**

**Yu-Hsiang Tsai 1,†, Wan-Ming Chen 2,†, Ming-Chih Chen 2, Ben-Chang Shia 2,3, Szu-Yuan Wu 2,3,4,5,6,7,8,9,10,\* and Chun-Chi Huang 1,\***


**Simple Summary:** Although sarcopenia during cancer diagnosis is an independent prognostic factor for poor overall survival in patients with various cancers, whether pre-existing sarcopenia is an independent risk factor for oral cavity squamous cell carcinoma (OCSCC) remains unclear. Therefore, we conducted a head-to-head propensity score matching (PSM) study to estimate the oncological outcomes of pre-existing sarcopenia in patients with OCSCC undergoing curative surgery. Both univariate and multivariate Cox regression analyses indicated that pre-existing sarcopenia was associated with poor survival than nonsarcopenia. Old age, male sex, advanced pT, advanced pN, differentiation grade II–III, margin-positive cancer, lymphovascular invasion, and CCI ≥ 1 were significant poor prognostic factors for survival in the patients with OCSCC undergoing curative surgery.

**Abstract:** Purpose: The effect of pre-existing sarcopenia on patients with oral cavity squamous cell carcinoma (OCSCC) remains unknown. Therefore, we designed a propensity score-matched population-based cohort study to compare the oncological outcomes of patients with OCSCC undergoing curative surgery with and without sarcopenia. Patients and Methods: We included patients with OCSCC undergoing curative surgery and categorized them into two groups according to the presence or absence of pre-existing sarcopenia. Patients in both the groups were matched at a ratio of 2:1. Results: The matching process yielded 16,294 patients (10,855 and 5439 without and with pre-existing sarcopenia, respectively). In multivariate Cox regression analyses, the adjusted hazard ratio (aHR, 95% confidence interval [CI]) of all-cause mortality for OCSCC with and without pre-existing sarcopenia was 1.15 (1.11–1.21, *p* < 0.0001). Furthermore, the aHRs (95% CIs) of locoregional recurrence and distant metastasis for OCSCC with and without pre-existing sarcopenia were 1.07 (1.03–1.18, *p* = 0.0020) and 1.07 (1.03–1.20, *p* = 0.0148), respectively. Conclusions: Pre-existing sarcopenia might be a significant poor prognostic factor for overall survival, locoregional recurrence, and distant metastasis for patients with OCSCC undergoing curative surgery. In susceptible patients

**Citation:** Tsai, Y.-H.; Chen, W.-M.; Chen, M.-C.; Shia, B.-C.; Wu, S.-Y.; Huang, C.-C. Effect of Pre-Existing Sarcopenia on Oncological Outcomes for Oral Cavity Squamous Cell Carcinoma Undergoing Curative Surgery: A Propensity Score-Matched, Nationwide, Population-Based Cohort Study. *Cancers* **2022**, *14*, 3246. https:// doi.org/10.3390/cancers14133246

Academic Editors: Carlo Lajolo, Gaetano Paludetti and Romeo Patini

Received: 1 June 2022 Accepted: 26 June 2022 Published: 1 July 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 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 (https:// creativecommons.org/licenses/by/ 4.0/).

at a risk of OCSCC, sarcopenia prevention measures should be encouraged, such as exercise and early nutrition intervention.

**Keywords:** sarcopenia; nonsarcopenia; OCSCC; survival; prognosis

### **1. Introduction**

Head and neck cancer (HNC) is the third most common cancer and the fifth leading cause of cancer deaths in men in Taiwan [1] because of betel nut chewing, cigarette smoking, and alcohol use [2–10]. The median age of patients with HNC in Taiwan is 55 years, indicating that they are an economically active population [1–10]; thus, improving their survival is essential. In Taiwan, the oral cavity squamous cell carcinoma (OCSCC) subtype accounts for more than 80% of HNC, whereas in Western countries, most HNCs are oropharyngeal cancers [2–10]. This difference is likely due to the habit of betel nut chewing in Taiwan [8–10]. Moreover, there are 377,713 new cases and 177,757 new deaths per year for oral cancer in the world based on the last updated GLOBOCAN (IARC, WHO) report in 2020 [11]. Despite advancements in therapeutics [8–10], the survival rate of HNC in Taiwan has remained dismal [1]. From the perspective of preventive medicine, if a prognostic factor for survival in patients with OCSCC can be corrected before cancer diagnosis, the factor should be screened and corrected for improving survival in OCSCC.

Sarcopenia, characterized by the loss of muscle mass, strength, and performance [12–14], can occur not only in overweight and underweight individuals but also in those with normal weight [15]. Unlike cachexia, sarcopenia does not require the presence of an underlying illness [16]. In addition, although most people with cachexia are sarcopenic, most individuals with sarcopenia are not considered cachectic [16]. Sarcopenia is associated with increased functional impairment, disability, fall, and mortality rates [17]. The causes of sarcopenia are multifactorial and include disuse, endocrine function alteration, chronic diseases, inflammation, insulin resistance, and nutritional deficiencies [14]. Therefore, sarcopenia and cancer cachexia-related sarcopenia are distinct conditions. Pre-existing sarcopenia can be prevented, whereas cancer-related sarcopenia cannot be prevented but can be treated.

Sarcopenia is associated with increased mortality for most cancers, except hormonerelated cancers (endometrial, breast, ovarian, and prostate cancers) and hematopoietic cancers [18–21], thus making it a major prognostic factor for poor overall survival and mortality in patients with cancer [18–21]. Sarcopenia-related cancer mortality might be a consequence of treatment-related toxicity [22,23]. However, whether pre-existing sarcopenia is an independent risk factor for different cancers, including OCSCC, remains unclear. A propensity score matching (PSM)-based design can resolve this issue by maintaining balance among the confounding factors of the case and control groups—all in the absence of bias [24–26]. Moreover, PSM is currently the recommended standard tool for estimating the effects of covariates in studies where any potential bias may exist [24–26]. Therefore, we conducted a head-to-head PSM study to estimate the oncological outcomes of pre-existing sarcopenia in patients with OCSCC undergoing curative surgery.

### **2. Patients and Methods**

### *2.1. Study Population*

We selected patients with OCSCC who had undergone curative surgery—tumor resection and neck dissection—between 1 January 2007 and 31 December 2017 from the Taiwan Cancer Registry Database (TCRD). The follow-up period was from the index date (i.e., date of surgery) to 31 December 2018. The types and indications of neck dissection were as follows: supraomohyoid neck dissection for clinically N0 tumors [27], modified neck dissection for ipsilateral clinically positive nodes [28], and bilateral neck dissection for contralateral metastases or tumors cross the midline [29]. Adjuvant treatments indicated for patients with OCSCC were based on the National Comprehensive Cancer Network (NCCN) guidelines and patients' tolerance [30]. The TCRD contains detailed cancer-related data of patients, including the clinical stage, cigarette smoking habit, treatment modalities, pathologic data, and grade of differentiation [5,8–10,31]. The study protocols were reviewed and approved by the Institutional Review Board of Tzu-Chi Medical Foundation (IRB109-015-B).

The diagnoses of the enrolled patients were confirmed after reviewing their pathological data, and patients who were newly diagnosed as having OCSCC were confirmed to have no other cancers or distant metastasis (DM). All patients with OCSCC underwent curative-intent surgery. The inclusion criteria were as follows: being aged ≥20 years, having a diagnosis of pathologic stage I–IVB OCSCC without metastasis according to the American Joint Committee on Cancer criteria (AJCC, 7th edition), and undergoing tumor resection and neck dissection. Patients were excluded if they had a history of other cancers before the index date, an unknown pathological stage, missing sex data, unclear differentiation of tumor grade, or a nonsquamous cell carcinoma pathologic type.

### *2.2. Interventions/Exposures*

Our definition of sarcopenia is according to the previous study from the Taiwan NHIRD [32]. In order to diminish the selection bias of the definition of sarcopenia, we only recorded the sarcopenia from the rehabilitation specialists, orthopedics, or family physicians. We have also added the sensitivity analysis of the recorded sarcopenia from the rehabilitation specialists, orthopedics, and family physician with/without other specialties (including endocrinology department) (Supplementary Table S2). In Taiwan, the coding of sarcopenia was based on a previous Taiwan study [33]; sarcopenia was defined as the skeletal muscle mass index (SMI) of 2 standard deviations (SDs) or more below the normal sex-specific means for young persons. Patients diagnosed as having sarcopenia after OCSCC diagnosis and those with sarcopenia diagnosed within 1 year before OCSCC diagnosis (excluding cancer treatment-related and cancer cachexia-related sarcopenia) were excluded. We also supplied the sensitivity analysis for the comparison of washout time intervals of one year and two years (Supplementary Table S1).

### *2.3. Comparisons*

We categorized the patients into two groups depending on whether they had sarcopenia before OCSCC diagnosis: Group 1 (nonsarcopenic OCSCC) and Group 2 (pre-existing sarcopenic OCSCC). In addition, we estimated oncological outcomes (all-cause mortality, locoregional recurrence [LRR], and DM) associated with sarcopenia. Comorbidity was assessed using the Charlson comorbidity index (CCI) [6,34]. Only comorbidities which appeared 12 months before the index date were included and they were coded and classified according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) or International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes at the first admission or after >2 appearances of a diagnostic code at outpatient visits.

### *2.4. Outcomes*

The oncologic outcomes were defined as all-cause death, LRR, and DM according to the previous oncologic studies [35–37]. All-cause mortality was the primary endpoint in both the groups. The secondary endpoints were LRR and DM.

### *2.5. Design Setting*

To reduce the effects of potential confounders when comparing all-cause mortality between patients without and with sarcopenia, we performed 2:1 PSM with a caliper of 0.2 for the following variables: age, sex, years of diagnosis, AJCC pathologic stages, pathologic tumor stages (pT), pathologic nodal stage (pN), differentiation grade, surgical margin, lymphovascular invasion (LVI), adjuvant treatments, CCI scores, cigarette smoking, alcohol use, and betel nut chewing. These variables are potential prognostic factors for allcause mortality for patients with OCSCC undergoing curative surgery. A Cox proportional hazards model was used to regress all-cause mortality in patients with OCSCC with a robust sandwich estimator used to account for clustering within matched sets [38]. Potential confounding factors for all-cause mortality for OCSCC were controlled in the PSM (Table 1). After well-matched PSM, the actual real-world data can indicate the oncological outcomes of pre-existing sarcopenia in patients with OCSCC undergoing curative surgery.

**Table 1.** Characteristics of patients with oral cavity squamous cell carcinoma with and without pre-existing sarcopenia (After propensity score matching 1:2).




RT, radiotherapy; CCRT, concurrent chemoradiotherapy; CCI, Charlson comorbidity index; SD, standard deviation; IQR, interquartile range; AJCC, American Joint Committee on Cancer; y, years old; N, numbers; Gy, Gray; pT, pathologic tumor stages; pN, pathologic nodal stages.

### *2.6. Statistical Analysis*

The aforementioned variables might be independent prognostic factors for all-cause mortality with residual imbalance after PSM [39,40]. Therefore, multivariate Cox regression analyses were performed to calculate hazard ratios (HRs) to determine whether pre-existing sarcopenia is an independent predictor of all-cause mortality.

After adjustment for confounders, all statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA). In a two-tailed Wald test, *p* < 0.05 was considered significant. OS, LRR, and DM were estimated using the Kaplan–Meier method and betweengroup differences were compared using the stratified log-rank test (stratified according to matched sets) [41].

### **3. Results**

### *3.1. Study Cohorts before and after PSM*

We identified 45,219 patients with OCSCC undergoing curative surgery (39,775 without and 5445 [12.04%] with pre-existing sarcopenia) before PSM (Supplementary Table S1). Compared with the patients without pre-existing sarcopenia, those with sarcopenia were older; were predominantly women; had higher CCI scores; more likely received the diagnosis in 2015–2017; had more advanced pT and pN stages; had more poor differentiation, margin positivity, and LVI-positive tumors; and received more adjuvant concurrent chemoradiotherapy (CCRT), higher radiotherapy (RT) doses, and higher cumulative platinum doses. PSM yielded 16,294 patients (10,855 without and 5439 with sarcopenia) who were eligible for further analysis and their characteristics are summarized in Table 1. Age, sex, years of diagnosis, cancer subtypes, AJCC pathological stages, pT, pN, differentiation, surgical margin, lymphovascular invasion, adjuvant treatments, CCI scores, cigarette smoking, alcohol use, and betel nut chewing were balanced between the cohorts (all *p* > 0.05). After PSM, the crude all-cause mortality, LRR, and DM were significantly higher in the patients with sarcopenia than in those without sarcopenia (Table 1).

### *3.2. Cox Proportional Hazard Models of All-Cause Mortality*

According to multivariate Cox regression analysis, pre-existing sarcopenia was a significant predictor of all-cause mortality (Table 2). Both univariate and multivariate Cox regression analyses indicated that sarcopenia was associated with poorer OS than nonsarcopenia. The HR for the univariate model was similar to that for the multivariate Cox regression analysis. Old age, male sex, advanced pT, advanced pN, differentiation grade II/III, margin positivity, LVI positivity, and CCI ≥ 1 were significantly poor prognostic factors for OS in the patients with OCSCC. In multivariate Cox regression analyses, the adjusted hazard ratio (aHRs, 95% confidence interval [CI]) of all-cause mortality for OCSCC with and without pre-existing sarcopenia was 1.14 (1.10–1.19, *p* < 0.0001). The aHRs (95% CIs) of mortality for male sex, age 50–59 years, age ≥ 60 years, pT2, pT3, pT4A, pT4B, pN1, pN2, pN3, differentiation grades II and III, margin positivity, LVI positivity, CCI ≥ 1, cigarette smoking, alcohol use, and betel nut chewing compared with female

sex, age < 50 years, pT1, pN0, differentiation grade I, margin negativity, LVI negativity, CCI = 0, no cigarette smoking, no alcohol use, no betel nut chewing were 1.28 (1.20–1.39), 1.14 (1.07–1.19), 1.25 (1.19–1.33), 1.05 (1.01–1.31), 1.31 (1.05–1.63), 1.66 (1.33–2.11), 1.72 (1.39– 2.17), 1.11 (1.04–1.24), 1.21 (1.05–1.41), 2.03 (1.72–2.71), 1.18 (1.12–1.23), 1.21 (1.12–1.31), 1.23 (1.18–1.33), 1.59 (1.38–1.87), 1.19 (1.13–1.26), 1.10 (1.04–1.22), 1.08 (1.03–1.23), and 1.09 (1.02–1.30), respectively.

**Table 2.** Univariable and multivariable Cox proportional regression model for all-cause mortality of the propensity score-matched groups of patients with oral cavity squamous cell carcinoma with and without pre-existing sarcopenia.



RT, radiotherapy; CCRT, concurrent chemoradiotherapy; CCI, Charlson comorbidity index; AJCC, American Joint Committee on Cancer; y, years old; pT, pathologic tumor stages; pN, pathologic nodal stages; Ref., reference group; CI, confidence interval; HR, hazard ratio. \* All the aforementioned variables in Table 2 were used in multivariate analysis.

### *3.3. Cox Proportional Hazard Models of LRR and DM*

Both univariate and multivariate Cox regression analyses indicated that pre-existing sarcopenia was associated with higher risk of LRR and DM than nonsarcopenia (Tables 3 and 4). In the multivariate Cox regression analysis, the aHRs (95% CIs) of LRR and DM for OCSCC with and without pre-existing sarcopenia were 1.07 (1.03–1.18, *p* = 0.0020) and 1.07 (1.03–1.20, *p* = 0.0148), respectively. In addition, poor prognostic factors for LRR and DM were similar with those of mortality, except old age and CCI scores. The multivariable Cox model revealed that male sex, advanced pT, advanced pN, differentiation grade II–III, margin positivity, LVI positivity, cigarette smoking use, alcohol use, and betel nut chewing use were independent poor prognostic factors for LRR and DM (Tables 3 and 4).

**Table 3.** Univariable and multivariable Cox proportional regression model for locoregional recurrence of the propensity score-matched groups of patients with oral cavity squamous cell carcinoma with and without pre-existing sarcopenia.



RT, radiotherapy; CCRT, concurrent chemoradiotherapy; CCI, Charlson comorbidity index; AJCC, American Joint Committee on Cancer; y, years old; pT, pathologic tumor stages; pN, pathologic nodal stages; Ref., reference group; CI, confidence interval; HR, hazard ratio.

**Table 4.** Univariable and multivariable Cox proportional regression model for distant metastasis of the propensity score-matched groups of patients with oral cavity squamous cell carcinoma with and without pre-existing sarcopenia.



RT, radiotherapy; CCRT, concurrent chemoradiotherapy; CCI, Charlson comorbidity index; AJCC, American Joint Committee on Cancer; y, years old; pT, pathologic tumor stages; pN, pathologic nodal stages; Ref., reference group; CI, confidence interval; HR, hazard ratio.

### *3.4. Kaplan–Meier Curves of Overall Survival, LRR, and DM*

Figure 1 and Supplementary Figures S1 and S2 present survival curves for OS, LRR, and DM plotted using the Kaplan–Meier method for the PSM sarcopenia and nonsarcopenia OCSCC groups who underwent curative surgery. The OS curve for nonsarcopenic OCSCC was higher than that for sarcopenic OCSCC (Figure 1, *p* < 0.001). The 5-year OS was 56.03% and 48.93% for the patients with OCSCC without and with pre-existing sarcopenia, respectively. Moreover, the cumulative LRR and DM rates were significantly higher for sarcopenic OCSCC than nonsarcopenic OCSCC in the log-rank test (Supplementary Figures S1 and S2, *p* values were all <0.0001 for LRR and DM, respectively).

**Figure 1.** Kaplan–Meier overall survival curves for the propensity score-matched sarcopenia and nonsarcopenia groups (controls).

### **4. Discussion**

Sarcopenia is an independent prognostic factor for poor survival in patients with HNC undergoing surgery, RT, or CCRT [20,42–47]. However, these studies included heterogeneous definitions of sarcopenia, inconsistent treatments for HNCs, different HNC subtypes, inhomogeneous HNC stages, very small sample sizes, and inconsistent cancer subtypes including oropharyngeal, hypopharyngeal, oral cavity, and laryngeal cancers [20,42–47]. None of these studies differentiated between sarcopenia as pre-existing or that related to cancer cachexia. Accordingly, their result that sarcopenia is a poor prognostic factor for survival outcomes might be due to cancer-related cachexia-induced sarcopenia or cancer treatment-related sarcopenia instead of pre-existing sarcopenia [20,42–47]. However, sarcopenia is different from cancer cachexia [14,16,17]. The causes of sarcopenia are multifactorial [14] and include muscle disuse, changes in endocrine function, chronic diseases, inflammation, insulin resistance, and nutritional deficiencies; many of these conditions can be detected early on and corrected through measures such as exercise or nutrition to prevent sarcopenia progression [48–51]. Therefore, we estimated the oncological outcomes of pre-existing sarcopenia in the patients with OCSCC undergoing curative surgery to determine the effect of pre-existing sarcopenia on OCSCC. To our knowledge, this is the first head-to-head PSM, largest, and longest follow-up study evaluating the effect of pre-existing sarcopenia on patients with OCSCC undergoing curative surgery. Our data indicated that pre-existing sarcopenia is an independent poor prognostic factor for mortality, LRR, and DM.

The definition of sarcopenia has been inconsistent in previous studies [20,42–47]. In patients with HNC receiving RT or CCRT, sarcopenia has been reported to be associated with poor OS and disease-free survival outcomes [42–45,47]. Only one report including patients with HNC receiving surgical excision demonstrated that sarcopenia appears to be a significant negative predictor of long-term OS in patients with HNC undergoing major surgery [43]. Stone et al. defined sarcopenia by using cross-sectional abdominal imaging performed within 45 days prior to surgery [43]. However, this definition precluded the differentiation of pre-existing sarcopenia from cancer cachexia-related sarcopenia [43]. This renders any results on the effect of sarcopenia unclear [43] and does not affect clinical practice in patients with HNC because cachexia is a well-known poor prognostic factor for OS in HNCs [52,53]. Our study is the first to present a clear definition of pre-existing sarcopenia (diagnosed ≥1 year before the diagnosis of OCSCC) in a homogenous group of patients with the same subtype of HNC (OCSCC) undergoing curative surgery. Therefore, our finding that pre-existing sarcopenia is the poor prognostic factor for OS, LRR, and DM might encourage the implementation of early screening for sarcopenia and intervention such as resistance exercise, protein supplementation, and vitamin D for patients at a high risk of OCSCC (betel nut chewing, cigarette smoking, or alcohol abuse) [48–51]. These valuable outcomes would provide references for the health government to establish health policies to correct, interrupt, or prevent the progression of pre-existing sarcopenia, particularly in the susceptible population.

Performing a randomized controlled trial (RCT) to evaluate oncological outcomes in patients with OCSCC undergoing curative surgery with and without pre-existing sarcopenia is difficult because sarcopenia cannot be treated using a tangible intervention [54]. Traditionally, striking a balance among the confounding factors of mortality in patients with OCSCC with and without sarcopenia (i.e., the case and control groups, respectively)—a main requirement of the RCT design—is impossible [54]. Although the main advantage of the PSM methodology is the more precise estimation of the covariate effect, PSM cannot control for factors not accounted for in the model. Moreover, PSM is predicated on an explicit selection bias of those who could be matched; in other words, individuals who could not be matched are not part of the scope of inference.

In the current study, our multivariable Cox regression analysis results indicated that age ≥ 50 years, male sex, advanced pT, advanced pN, differentiation grade II–III, margin positivity, LVI positivity, CCI ≥ 1, cigarette smoking, alcohol use, and betel nut chewing are significant poor prognostic factors for mortality—corroborating the results of previous studies (Table 2 and Figure 1) [1–10,31,55–59]. Moreover, male sex, advanced pT, advanced pN, differentiation grade II-III, margin positivity, LVI positivity, cigarette smoking, alcohol use, and betel nut chewing were the poor independent prognostic factors for LRR and DM in patients with OCSCC undergoing curative surgery (Tables 3 and 4 and Supplementary Figures S1 and S2). Age > 50 years was associated with the risk of mortality in patients with HNC undergoing curative surgery, consistent with our results [3,31]. In Taiwan, male sex and high CCI scores are known poor prognostic factors for OS in patients with HNC undergoing curative surgery [3,31,59]. Our data indicated that advanced pT/pN, margin positivity, and LVI positivity are associated with an increase in all-cause mortality, LRR, and DM, consistent with previous studies and NCCN guidelines [3,30,55–57]. In our multivariable analysis, poor prognostic factors for oncological outcomes for patients with OCSCC undergoing curative surgery were similar to those reported in previous studies [1–10,30,31,55–59]. Pre-existing sarcopenia was the only independent poor prognostic factor for OS, LRR, and DM for OCSCC that was never reported in previous studies. Although cancer cachexia is a well-known poor prognostic factor for survival in HNC [52,53], ours is the first study to establish pre-existing sarcopenia as an independent prognostic factor for OCSCC.

The mechanism through which pre-existing sarcopenia serves as a poor prognostic factor for OS, LRR, and DM might be associated with multiple factors including the metabolic processes of insulin resistance and systemic inflammation [14,16,17]. Patients with sarcopenia might have systemic inflammation that reduces liver cytochrome activities and drug clearance and metabolic processes, leading to a poor therapeutic effect [60]. In addition, inflammation by sarcopenia can cause a decrease in skeletal muscle density. A decreased muscle density is related to intramuscular lipid accumulation and favored by systemic inflammation, thus leading to a vicious cycle [60]. Therefore, early intervention to break this cycle is critical in patients with sarcopenia [48–51]. According to an epidemiological study in Taiwan, the incidence of oral cancer was 123-fold higher in patients who smoked, consumed alcohol, and chewed betel quid than in abstainers [2]. Patients with sarcopenia with risk factors for OCSCC [60] are the susceptible population for poor OS. Early screening for and treatment of sarcopenia for the susceptible population might improve survival outcomes in case they develop OCSCC.

This study has several limitations. First, the cohort derived from an Asian population in Taiwan. Although no evidence indicating a significant difference in survival of OCSCC between Asian and non-Asian populations has been reported, the current results should be cautiously extrapolated to non-Asian populations. Second, this study was performed on a big database and thus it is a real challenge to rule out an ecological bias (attributed to confounding or risk factors). PSM cannot control for factors not accounted for in the model and is predicated on an explicit selection bias of the variables that were matched. Third, patients with antecedents of other cancers were excluded. The field cancerization theory is well accepted on this anatomical area, i.e., a patient with oral cancer has a higher risk to develop future aerodigestive carcinomas (and vice versa) [4,61,62]. However, the primary endpoint in the current study is the all-cause death between sarcopenia and nonsarcopenia OCSCC, OCSCC patients combined with other cancers will have higher mortality attributed to more aggressive treatments or more advanced stages on the other cancers, whatever synchronous or metachronous cancers [4,61,62]. In order to decrease the bias of all-cause death from the other cancers in the OCSCC patients, patients with antecedents of other cancers were excluded. Fourth, the diagnoses of all comorbid conditions were based on *ICD-9-CM* or *ICD-10-CM* codes in this study. Nevertheless, the Taiwan Cancer Registry Administration reviews charts and interviews of beneficiaries in the TCRD to verify the accuracy of the diagnoses, and it audits hospitals with outlier chargers or practices and subsequently heavily penalizes them if it identifies any malpractice or discrepancies. However, to obtain precise population specificity and disease occurrence data, a large-scale RCT

carefully comparing patients with OCSCC with or without sarcopenia is warranted, but such RCTs may be difficult to execute.

Despite these limitations, a major strength of our study is the use of a nationwide population-based registry with detailed baseline information. The TCRD is linked with Taiwan's National Cause of Death Database; thus, in the current study, we could perform a lifelong follow-up for most patients. Moreover, this study is the first, largest, and longest follow-up comparative cohort study to estimate the primary endpoint of OS in patients with OCSCC with and without pre-existing sarcopenia undergoing curative surgery. The covariates between the two groups were homogenous and any bias between the two groups was removed through PSM (Table 1). Considering the magnitude and statistical significance of the observed effects in the current study, the limitations are unlikely to have affected our conclusions.

### **5. Conclusions**

Our results indicate that pre-existing sarcopenia is a significantly poor prognostic factor for OS, LRR, and DM in patients with OCSCC undergoing curative surgery. Individuals with a high risk of OCSCC, such as those who have a habit of betel nut chewing, alcohol, or smoking, should be screened for sarcopenia and intervention in terms of exercise and nutrition should be promoted.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cancers14133246/s1, Figure S1: Kaplan–Meier overall cumulative locoregional recurrence curves for the propensity score–matched sarcopenia and nonsarcopenia groups (controls); Figure S2: Kaplan–Meier overall cumulative distant metastasis curves for the propensity score–matched sarcopenia and nonsarcopenia groups (controls); Table S1: Sensitivity analysis of washout time-intervals of one year and two years for definition of preexisting sarcopenia; Table S2: Sensitivity analysis of preexisting sarcopenia recorded by special specialties and all specialties.

**Author Contributions:** Conception and Design: Y.-H.T., W.-M.C., M.-C.C., B.-C.S., C.-C.H. and S.-Y.W.; Collection and Assembly of Data: Y.-H.T., C.-C.H. and S.-Y.W.; Data Analysis and Interpretation: W.-M.C., B.-C.S. and S.-Y.W.; Administrative Support: S.-Y.W.; Manuscript Writing: Y.-H.T., C.-C.H. and S.-Y.W.; Final Approval of Manuscript: All authors. All authors have read and agreed to the published version of the manuscript.

**Funding:** Lo-Hsu Medical Foundation, LotungPoh-Ai Hospital, supports Szu-Yuan Wu's work (Funding Number: 10908, 10909, 11001, 11002, 11003, 11006).

**Institutional Review Board Statement:** The study protocols were reviewed and approved by the Institutional Review Board of Tzu-Chi Medical Foundation (IRB109-015-B).

#### **Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data sets supporting the study conclusions are included in the manuscript. We used data from the National Health Insurance Research Database and Taiwan Cancer Registry database. The authors confirm that, for approved reasons, some access restrictions apply to the data underlying the findings. The data used in this study cannot be made available in the manuscript, the supplementary files, or in a public repository due to the Personal Information Protection Act executed by Taiwan's government, starting in 2012. Requests for data can be sent as a formal proposal to obtain approval from the ethics review committee of the appropriate governmental department in Taiwan. Specifically, links regarding contact info for which data requests may be sent to are as follows: http://nhird.nhri.org.tw/en/Data\_Subsets.html#S3 and http://nhis.nhri.org.tw/ point.html (accessed on 5 February 2021).

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

### **Abbreviations**

RT, radiotherapy; CCRT, concurrent chemoradiotherapy; CCI, Charlson comorbidity index; SD, standard deviation; IQR, interquartile range; AJCC, American Joint Committee on Cancer; y, years

old; N, numbers; Gy, Gray; pT, pathologic tumor stages; pN, pathologic nodal stages; CI, confidence interval; HR, hazard ratio; OCSCC, oral cavity squamous cell carcinoma; PSM, propensity score matching; LRR, locoregional recurrence; DM, distant metastasis; HNCs, head and neck cancers; TCRD, Taiwan Cancer Registry Database; NCCN, National Comprehensive Cancer Network; AJCC, American Joint Committee on Cancer criteria; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; ICD-10-CM, International Classification of Diseases, Tenth Revision, Clinical Modification; CCI, Charlson comorbidity index; LVI, lymphovascular invasion; RCT, randomized controlled trial.

### **References**


### *Article* **Oral Health Status in Patients with Head and Neck Cancer before Radiotherapy: Baseline Description of an Observational Prospective Study**

**Cosimo Rupe <sup>1</sup> , Alessia Basco 1, Anna Schiavelli 1,\*, Alessandra Cassano 2, Francesco Micciche' 3, Jacopo Galli 4, Massimo Cordaro <sup>1</sup> and Carlo Lajolo 1,†**


**Simple Summary:** Patients with head and neck cancer (HNC) are often considered as a group with compromised oral conditions, but this idea is not sufficiently supported by data in the literature. This study examined the oral condition—specifically the presence of caries and periodontal disease—of a cohort of patients with HNC waiting to start radiation therapy treatment and possible correlations between oral health, different types of HNC and various risk factors. The results confirm that the oral status of many patients with HNC is poor even before radiotherapy treatments and that smoking habit and tumor site are associated with poor oral health. These findings underline the importance of a dentist within a head and neck tumor board (TB), so that oral health can be restored as soon as possible.

**Abstract:** (1) Background: The general hypothesis that HNC patients show compromised oral health (OH) is generally accepted, but it is not evidence-based. The objective of this baseline report of a prospective observational study was to describe the oral health of a cohort of patients with HNC at the time of dental evaluation prior to radiotherapy (RT). (2) Materials and Methods: Two hundred and thirteen patients affected by HNC who had received an indication for RT were examined with the support of orthopantomography (OPT). The DMFt of all included subjects, their periodontal status and the grade of mouth opening were recorded. (3) Results: A total of 195 patients were ultimately included: 146/195 patients (74.9%) showed poor OH (defined as having a DMFt score ≥ 13 and severe periodontitis). The following clinical characteristics were correlated with poor oral health in the univariate analysis: tumor site, smoking habit and age of the patients (in decades); χ2 test, *p* < 0.05. (4) Conclusions: This study confirms that the OH of HNC patients is often compromised even before the beginning of cancer treatment and, consequently, highlights how important it is to promptly schedule a dental evaluation at the moment of diagnosis of the cancer.

**Keywords:** head and neck cancer; oral status; periodontitis; dental caries; DMFt

**Citation:** Rupe, C.; Basco, A.; Schiavelli, A.; Cassano, A.; Micciche', F.; Galli, J.; Cordaro, M.; Lajolo, C. Oral Health Status in Patients with Head and Neck Cancer before Radiotherapy: Baseline Description of an Observational Prospective Study. *Cancers* **2022**, *14*, 1411. https://doi.org/10.3390/ cancers14061411

Academic Editor: Petra Wilder-Smith

Received: 24 January 2022 Accepted: 4 March 2022 Published: 10 March 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 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 (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

The head and neck region is an anatomical heterogeneous area that can give rise to a variety of malignancies and show different risk factors, prognoses and treatments. Head and neck cancers (HNCs) represent the seventh most common malignancy worldwide [1].

The general hypothesis that HNC patients show a high prevalence of caries and periodontitis and, therefore, compromised oral health (OH) even before cancer therapy (i.e., radiotherapy, RT) is generally accepted, but it is not evidence-based. In fact, it is possible to highlight a lack of clinical data about the OH of these patients before oncological treatments.

Several studies reported that the majority of HNC patients did not attend any dental visit during the year preceding the cancer diagnosis and that many of these patients consulted a dental specialist only in cases of acute pain or other urgencies [2–4]. The overlapping of some risk factors—the most important being smoking habit—might be another possible explanation for the compromised conditions of HNC patients. Tobacco smoking is considered the main risk factor for the majority of HNCs and one of the main risk factors for the onset and progression of periodontitis and for its response to treatment [5–8]; furthermore, hyposalivation following prolonged exposure to tobacco smoking could increase the risk of caries development [9,10].

Furthermore, especially when RT is performed, preserving OH becomes crucial in the multidisciplinary management of these patients, since RT increases the risk of developing dental caries, leading to tooth loss, a well-known risk factor for major complications such as osteoradionecrosis [11–13].

Considering this, it is easy to imagine that HNC patients have a higher probability of developing dental diseases. Nevertheless, data available from the literature are scarce, often inaccurate or incomplete, and many articles do not stratify the statistical analysis according to the primary location of the cancer. The present study is the first report of a prospective protocol aiming to evaluate the OH of an HNC cohort undergoing RT.

The primary objective of this cross-sectional study was to describe the OH conditions of a cohort of HNC patients evaluated during the dental visit preceding RT. The secondary objective was to identify a correlation between the clinical characteristics of the patients and their OH status.

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

This study was conducted according to the Declaration of Helsinki, and all patients signed an informed consent form. The protocol was approved by the Ethics Committee of the Università Cattolica del Sacro Cuore (Ref. 22858/18) and was registered at ClinicalTrials.gov (ID: NCT04009161).

Patients affected by HNC attending the Oral Medicine, Head and Neck Department— Fondazione Policlinico Universitario A. Gemelli—IRCSS, between March 2017 and September 2021 were consecutively recruited in this study.

The following inclusion criteria were considered: HNC diagnosis and indication for RT.

The exclusion criteria were the impossibility of accurately evaluating OH conditions (i.e., outcomes of oncologic surgery incompatible with the dental procedures to diagnose caries and periodontitis) and patients having already received RT in the head and neck region.

All patients were visited prior to RT, with the support of an orthopantomograph (OPT). Firstly, anagraphic and anamnestic data were carefully recorded, particularly focusing on the oncologic history of the patient and on exposure to risk factors for oncologic and dental diseases.

Subsequently, the clinical evaluation of the following parameters was performed: presence of dental caries and DMFt score, periodontal health, maximal mouth opening (MMO).

The DMFt index is the key measure of caries experience in dental epidemiology [14]. It sums the number of decayed teeth, missing teeth due to caries and filled teeth in the permanent dentition. An examination for dental caries in permanent teeth is performed, examining 32 teeth. The permanent dentition status of each tooth (crown and root) is recorded as a score, where 0 corresponds to a tooth that shows no evidence of treated or untreated caries, and 1 corresponds to the case of tooth decay (treated or untreated) or a missing tooth (due to caries) [15].

The diagnosis of caries was performed through the clinical examination with the help of a dental explorer and a mouth mirror and, when in doubt, with the support of an intraoral radiograph (periapical or bitewing), performed with the help of film holders (Dentsply Sirona, Rome, IT). A bitewing radiograph was performed in every case in which visual inspection of the interproximal tooth surface was not possible. Nevertheless, when a diagnosis of an endodontic or periodontal lesion had to be performed, a periapical radiograph was taken. Caries involving the dentine were considered in the DMFt score (ICDAS™ code 3 and higher) [16–18].

Clinical evaluation of periodontitis was performed according to international standards [19]. A full-mouth periodontal examination was performed by the same operator (L.C.), with more than ten years of experience in periodontology, using an NCP15 periodontal probe and collecting the following data (six sites for each tooth): periodontal probing depth (PPD), the distance between the tip of the periodontal probe and the gingival margin; gingival recession (REC), the distance between the gingival margin and the cementoenamel junction; clinical attachment loss (CAL) for each assessed site; furcation involvement (FI), according to the Hamp classification [20]; number of tooth losses due to periodontitis; tooth mobility; full-mouth plaque score (FMPS) [21]; and full-mouth bleeding score (FMBS) [22].

After data collection, the periodontal cases were staged according to the diagnostic criteria of the 2017 classification: CAL ≥ 2 mm affecting two nonadjacent teeth, buccal or oral CAL ≥ 3 mm and PPD > 3 mm affecting two or more teeth were the diagnostic criteria to define a periodontitis case. Interdental CAL from 3 to 4 mm was the parameter which shifted the diagnosis to stage II periodontitis, while more severe CAL or at least one tooth lost due to periodontitis was the criterion which determined the shift to stage III or IV periodontitis. The differential diagnosis between stage III and IV periodontitis was driven by the following parameters: tooth loss due to periodontitis ≥ 5, masticatory dysfunction due to secondary occlusal trauma, bite collapse, drifting or flaring, which were the diagnostic criteria for stage IV periodontitis [19]. The clinical charts of the patients visited before 2017 were rescreened to stage the periodontal cases according to the abovementioned classification. OPT was used as a support to complete the diagnosis and staging of periodontitis; in case of uncertainty, an intraoral radiograph was performed, compatible with the outcomes of the major oncologic surgery.

The M parameter (teeth missed due to caries), as well as the number of teeth lost due to periodontitis, was evaluated by analysing old radiographic exams provided by the patients. In case old radiographic exams were unavailable, the patients were asked about the reason for previous teeth extractions.

The MMO was defined as the greatest distance (mm) between the incisal edge of the maxillary central incisor and the incisal edge of the mandibular central incisor and was measured by using a modified vernier caliper [23]. The MMO of the edentulous patients was measured by removing every removable prosthesis, and the edentulous ridges were used as reference points.

The following variables were recorded: sex, age, risk factors (smoking, diabetes), previous or scheduled oncological treatment (chemotherapy and surgery), site, histological type and stage of the tumor, DMFt, stage of periodontitis and MMO.

The oral health (OH) parameter was defined as a dichotomous variable, and DMFt and periodontal staging were used to define OH status, defined as "poor" in cases of DMFt ≥ 13 and/or stage III or IV periodontitis and as "good" only in cases of lower values of each of these variables. The DMFt score of 13 was chosen as a cut-off defining good OH, since it has been reported to be the mean value of DMFt in non-developing countries [24,25]. Stage III and IV periodontitis were chosen as cut-off values, since they define "severe" periodontitis, according to the 2017 classification [19].

STROBE guidelines were followed to write this paper (Table S1).

### *Statistical Analysis*

The sample size was calculated according to the simple causal sampling formula. Considering a DMFt ≥ 13 and/or stage III or IV periodontitis as predictive of poor OH, and setting the possible prevalence of poor OH at 85% and a desired precision of 5%, 195 patients were included in the final sample.

Qualitative variables were described using absolute and percent frequencies, whereas quantitative variables were summarized either as the mean and standard deviation (SD), if normally distributed, or as the median, otherwise.

The following variables were evaluated as absolute values and reclassified in ranges. DMFt was reclassified according to the established cut-off defining a poor OH condition: DMFt ≥13; periodontitis was reclassified into three categories: absence of periodontitis, stage I or II periodontitis and stage III or IV periodontitis; MMO was reclassified according to the reduced mouth opening cut-off: MMO ≤ 25 mm [26,27]. DMFt and periodontal staging were used to define OH status as either "poor" or "good", as described in the Materials and Methods section.

Correlation analysis between the OH parameters (DMFt and periodontitis) and the clinical characteristics of the patients was performed. The Kolmogorov–Smirnov test was performed to evaluate the normal distribution of the quantitative variables. The Mann– Whitney U test and Kruskal–Wallis test were performed to compare continuous variables with nonparametric distributions, whereas parametric variables were analyzed using ANOVA. Pearson's χ2 test and Fisher's exact test were used to compare discontinuous variables. A logistic regression model was built to evaluate factors affecting the probability of the main outcome variable ("poor OH").

The statistical analysis was stratified according to the following variables: tumor site; patient age (by decade); and smoking habit.

Univariate analysis was performed to determine risk factors associated with poor OH (as defined in the Materials and Methods section), and the risk factors were introduced in a stepwise logistic regression analysis to identify independent predictors of poor OH. All statistical analyses were performed using IBM SPSS Statistics software (IBM Corp. Released 2017. IBM SPSS Statistics for Apple, Version 25.0 Armonk, NY, USA: IBM Corp).

### **3. Results**

### *3.1. General Characteristics of the Population*

Two hundred and thirteen patients were consecutively assessed and enrolled, while eighteen patients were excluded, since they did not fulfil the inclusion criteria (their clinical conditions did not allow clinical evaluation). The final sample included 195 patients (67 female and 128 male subjects), with a mean age of 60.4 years (SD: 12.4; range: 22–92). The mean time between the cancer diagnosis and the dental evaluation was 37.2 days (SD: 12.02; range: 15–64).

The general characteristics of the population are presented in Table 1. It is worth mentioning that the studied population represents a sample of a HNC population, reflecting the heterogeneous characteristics and risk factors for each malignancy.

**Table 1.** General characteristics of the population and correlation with OH.



<sup>a</sup> Major oncologic surgery (i.e., fibula free flap for mandible reconstruction, glossectomy). SCC: squamous cell carcinoma.

### 3.1.1. Oral Health

The clinical and radiographic evaluation showed that 8/195 (4.1%) subjects were totally edentulous, 115/195 (59%) showed a DMFt score ≥ of 13 and 150/195 (76.9%) were affected by periodontitis. Among these 150 patients, 107 (71.3%) showed stage III or IV periodontitis. Only 3/195 patients had a DMFt score = 0 (1.53%), while the median DMFt score was 16.91 (range: 0–32; SD: 9.1). A total of 146 patients out of 195 (74.9%) showed poor OH. The results describing the oral health of the studied population are reported in Table 2.


**Table 2.** Oral health parameters of the studied population and correlation with OH. SD: standard deviation.

### 3.1.2. Tumor Localization and OH Conditions

Patients with different tumor sites showed different OH conditions (χ2 test, *p* <0.05), with the larynx being associated with poor OH (86.4% of the cases) and the rhinopharynx being associated with good OH conditions (56.5%). The prevalence of DMFt ≥ 13 was higher in salivary gland (80%) and laryngeal (75%) patients than in patients with other tumor sites (χ2 test, *p* < 0.05). The subjects affected by laryngeal tumors also had a high prevalence of stage III or IV periodontitis, although this association was not statistically significant. The results of the statistical analysis, stratified according to the localization of the tumor, are reported in Table 3.

### 3.1.3. Smoking and OH Conditions

Smoking habit was correlated with the diagnosis of periodontitis: 74.8% of severe periodontal patients (stage III or IV) were smokers or former smokers (χ2 test, *p* < 0.05). The habit of smoking was also correlated with DMFt ≥ 13 (71.3%; χ2 test, *p* < 0.05) and poor OH (70.5%; χ2 test, *p* < 0.05). Multiple logistic regression analysis confirmed that smoking

habit was a risk factor for severe periodontitis (OR = 4.78; 95% CI = 2.01–11.36; *p* < 0.05), for DMFt ≥ 13 (OR = 2.30; 95% CI = 1.19–4.44; *p* < 0.05) and, therefore, for poor OH (OR = 3.27; 95% CI = 1.46–7.33; *p* < 0.05). The results of the analysis, stratified according to smoking habit, are reported in Table 4, and Figures 1–3.

**Table 3.** General and oral health characteristics of the studied population, according to the localization of the tumor.



**Table 4.** General and oral health characteristics of the studied population, according to the habit of smoking.

**Figure 1.** Distribution of periodontitis according to the habit of smoking.

**Figure 2.** Distribution of DMFt < 13 according to the habit of smoking.

### 3.1.4. Age and OH Conditions

The cases of severe periodontitis (stages III and IV) were diagnosed only in subjects aged > 40 years, and 93.5% of periodontal patients were older than 49 years (χ2 test, *p* < 0.05). Additionally, the distribution of high scores of DMFt (13 or higher) was not homogeneous (χ2 test, *p* < 0.05): DMFt scores of ≥ 13 were only found among subjects aged > 40 years, with a peak in the 70–79 years decade (86.4% of the subjects who were allocated to this decade) and in the > 80 years category (75%). Consequently, poor OH conditions were more prevalent among the elderly population, with a peak in subjects aged > 70 years (95.6% of subjects being older than 70 years). All edentulous patients in the studied population were older than 60 years (χ2 test, *p* < 0.05). Multiple logistic regression analysis showed that

age (in decades) was a risk factor for periodontitis (stage I and II periodontitis: OR 1.73, 95% CI = 1.15–2.61; stage III and IV periodontitis: OR 3.30, 95% CI = 2.17–5.00; *p* < 0.05); for DMFt ≥ 13 (OR = 2.07; 95% CI = 1.53–2.79; *p* < 0.05); and for poor OH (OR = 2.98; 95% CI = 2.01–4.41; *p* < 0.05).The results of the analysis, stratified according to age, are reported in Table 5, and Figures 4–6.

**Figure 3.** Distribution OH status according to the habit of smoking.

**Figure 4.** Distribution of periodontitis according to age of the patients (in decades).

**Table 5.** General and oral health characteristics of the studied population, according to the age of the subjects (decades).



**Figure 6.** Distribution of OH status according to age of the patients (in decades).

### **4. Discussion**

The role of the dentist in the head and neck tumor board (TB) is becoming increasingly important, especially in the context of modern multidisciplinary management, which places greater emphasis on the quality of life of patients after, or during, cancer therapy. The results of the present work confirm the importance of a dental evaluation prior to RT to prepare a patient for these complex therapies.

Available studies regarding the oral status of subjects with HNC at the time of diagnosis are few and often inaccurate or incomplete [28]. From this lack and from the clinical impressions of many specialists derives the probably correct belief that HNC patients present poor OH. This idea is even more ingrained when it comes to subjects with oral cavity tumors.

The description of the oral status of the cohort of patients with HNC proposed by this study confirms, within the limits of a cross-sectional study, the generally accepted idea that subjects with HNC very often present poor OH, although this is not supported by the current literature.

In particular, the subjects of this cohort presented poor OH (DMFt ≥ 13 and/or periodontitis stage III or IV) in 74.9% of cases. The OH conditions were not equally distributed among the different tumor sites (χ2 test, *p* < 0.05): the subjects affected by SCC of the larynx (86.4%), of the salivary glands (86.6%) and of the oral cavity (78%) presented a higher prevalence of poor OH, when compared to the subjects affected by nasopharyngeal cancer (56.5% of nasopharyngeal patients presented DMFt < 13 and absence of severe periodontitis). Nevertheless, in the multivariate analysis, none of the tumor sites were revealed as an independent risk factor for poor OH.

The present work confirms that OH was more compromised the older the subjects were, with a peak (95.6% of cases) at 70 years of age and older (OR = 2.98; 95% CI = 2.01–4.41; *p* < 0.05). Multiple logistic regression analysis also showed that age was an independent risk factor for periodontitis (stage I and II periodontitis: OR 1.73, 95% CI = 1.15–2.61; stage III and IV periodontitis: OR 3.30, 95% CI = 2.17–5.00; *p* < 0.05) and for DMFt ≥ 13 (OR = 2.07; 95% CI = 1.53–2.79; *p* < 0.05).

The median DMFt value of the cohort analysed was 16.9. Fifty-nine percent of the included patients (115/195) had DMFt ≥ 13. Within the total population, only three subjects had DMFt = 0.

Although it is not possible to compare our results with those of previous works that studied cohorts of HNC patients, mainly due to the heterogenous methodology, some studies that reached similar conclusions can be found, such as those by Critchlow et al., Raskin et al. and Patel et al., who reported mean DMFt values of 19.6, 17.6 and 16.2 in HNC cohorts, respectively [28–30]. On the other hand, other studies (i.e., Jham et al. [31], Tezal et al. [32], Moraes et al. [33] and Kim et al. [34]) reported no significant correlation between HNC cancer and caries experience. Likely, the heterogeneity of the data stems from the different study designs, the criteria used in the different evaluations and the differences among the studied populations (i.e., geographical area, oral hygiene, access to dental care, distribution of different HNCs).

The percentage of subjects with periodontitis included in the present study was high (76.9%, 150/195) compared with epidemiological studies conducted in Europe, in which the prevalence of periodontitis did not exceed 70%, even in older age groups [35].

The classification of periodontitis proposed in 2017 [19] aims to remedy many of the critical issues present in epidemiological studies and to provide a more complete and detailed description of the populations under study. For this reason, in the present work, we chose to classify all cases of periodontitis based on this classification. In fact, if this study had limited itself to adopting the criteria proposed by previous classifications, many of the subjects with poor oral conditions, or a terminal dentition, would not have been included among the cases of severe periodontitis. The new classification, moreover, has made it possible to evaluate the periodontal status with a system based on two parameters (staging and grading) that, combined, provide information on the prognosis of the teeth and the complexity of the treatments required by the individual case. The combination of all this information constitutes a fundamental aid in deciding whether or not to perform extractions before RT.

Studies adopting the criteria proposed by the 2017 classification are very few [36–38], and our present work is the first to use them in a cohort of patients with HNC. Studies that have attempted to investigate a possible correlation between periodontitis and HNC are extremely diverse and often methodologically weak, as highlighted by a recent review [39]. In particular, the majority of studies did not adopt sound criteria to diagnose periodontitis [28,40–50], and only one [33] was based on a clinical evaluation integrated by the collection of truly suitable parameters (PPD and CAL).

Almost all authors who have analysed the OH of HNC patients before RT reported a high prevalence of periodontitis: Bonan et al. [51] reported a 93% prevalence of moderate or severe periodontitis, although cases were evaluated on the basis of a different classification; Moraes et al. [33] found that 80% of patients with oral and oropharyngeal SCC had generalized chronic periodontitis, almost exclusively severe. Although the results reported in the present study cannot be significantly compared with those of previous works because of methodological differences, they confirm that periodontitis, due to still unproven causes, is very common among patients with HNC.

Among the most plausible causes, the high incidence of smokers in these populations could play a key role. The data reported in our present study support this hypothesis; in fact, smokers represented 74.7% of the patients affected by stage III–IV periodontitis (80/107) (OR = 4.78; 95% CI = 2.01–11.36; *p* < 0.05). Statistical analysis showed that smoking also affected caries susceptibility (OR = 2.30; 95% CI = 1.19–4.44; *p* < 0.05) and, consequently, overall OH (OR = 3.27; 95% CI = 1.46–7.33; *p* < 0.05).

Despite the high percentage of patients with poor OH, only 8/195 (4.1%) were completely edentulous. The difference between the data reported by the present work and those of previous studies [4,31,51] may be influenced by the lower proportion of older individuals included in the present study (only 23.1% of patients were >70 years of age).

Interestingly, reduced MMO (<25 mm) did not correlate with the parameters of OH assessment. Reduced MMO is a very frequent clinical finding in HNC cohorts, as it can occur following both oncologic surgery and RT and makes dental care and inspection of the oral cavity particularly difficult, including during cancer follow-up appointments. However, a prospective study aiming to evaluate the correlation between MMO and OH is needed. It is very likely, in fact, that the greater difficulty in oral hygiene procedures, as

well as in routine dental therapies and inspection procedures, due to a reduced MMO leads to an increase in the incidence of caries and a worsening of periodontal conditions.

This cross-sectional study has several strengths. The description of the oral status of the cohort is based on validated diagnostic and prognostic criteria, obtained through clinical and radiographic evaluation. Additionally, the reported results open the way to further investigating possible correlations between OH and HNC.

This study does not solely report the prevalence of caries and periodontitis in the analysed population; it proposes, for the first time, a criterion that may allow evaluating the OH of examined patients in a global and objective way. Establishing a cut-off to divide subjects into two groups according to the OH found emphasizes how defining an oral condition as "good" or "poor" is necessary, not only to find the presence or absence of caries and/or periodontitis, but also to quantify severity and to evaluate the two diseases through an integrated system.

Presenting a representative sample from each subsite of HNCs is one of the strengths of this study, as it provides a more specific picture of the OH conditions of patients with different HNCs. However, this also implies a limitation: the analysed sample, including subjects with tumors differing profoundly in risk factors and clinical manifestations, might be inhomogeneous. However, statistical analysis stratified by tumor subsites effectively allows highlighting the different peculiarities of individual HNCs from an OH perspective.

This study also has several limitations, among which, like all studies having evaluated the OH of HNC patients using DMFt as a parameter, is the retrospective attribution of the M parameter. This consideration also applies to the retrospective attribution of the number of teeth lost due to periodontitis. This necessity could lead to overestimating the prevalence of one pathology over another. However, the use of the "OH" parameter allows us to curb the extent of this potential bias, since it integrates the two main variables of interest.

In addition, a possible bias for this study is the lack of a control group, homogeneous to the one studied in terms of age, gender and smoking habits. More studies, with a different design (i.e., case–control studies), are needed to confirm that HNC patients have poorer OH than the general population.

Another parameter rendering the characteristics of the population peculiar is that all included patients had received an indication to undergo RT, since a dental visit is overwhelmingly indicated to prevent unwanted effects of RT. With this study being a reallife monocentric experience, indication for RT was chosen since the treatment of RT patients is the most "demanding", both from oncological and dental points of view. Nevertheless, our study also includes patients that underwent major oncological surgeries. Their inclusion within our sample could have made the observed population more homogeneous in terms of OH variables, making our sample more representative of HNC patients than a population undergoing exclusive RT.

Nevertheless, it could be considered as a selection bias, since patients who underwent a major oncologic surgery often present poorer OH, due to the reduced ability to adequately perform oral hygiene procedures, resulting from surgical procedure-induced anatomic alterations. Nevertheless, the results of our study show that previous oncologic treatment did not have a statistically significant correlation with OH, somehow confirming that this possible bias did not have a great impact. This could be explained by the fact that the dental evaluation was carried out in a time-lapse not exceeding 60 days, an insufficient time frame to significantly influence the parameters analyzed in this study. Notwithstanding, the results of the present work demonstrate how HNC patients present poor OH even prior to RT, which makes their inclusion within a protocol of primary and secondary dental prevention indicated.

### **5. Conclusions**

This work highlights, with a high level of evidence, the number of HNC patients presenting poor OH in the months immediately following their malignancy diagnosis and their consequent need for prevention protocols and highly rigorous dental therapy, considering the increasing number of patients undergoing RT.

With the time window between the dental evaluation and the start of RT being particularly narrow, performing multiple extractions becomes necessary, resulting in further worsening of the periodontitis stage and masticatory function. This can only be avoided by referring the patient to a dental team, who will commence necessary therapies and preventive measures. Moreover, due to the increasing rate of recurrences and second primary tumors, an increasing number of patients receive an indication for RT.

It is important, therefore, that the figure of the dentist be regularly involved in multidisciplinary TBs for the management of head and neck patients to improve patient quality of life as much as possible and to reduce the risk of complications following oncologic treatment.

**Supplementary Materials:** The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/cancers14061411/s1, Table S1: STROBE Statement—Checklist of items that should be included in reports of *cohort studies*.

**Author Contributions:** Conceptualization, C.L. and C.R.; data curation, C.R. and A.S.; methodology, C.L. and C.R.; formal analysis, C.L. and C.R.; investigation, C.R., A.S. and A.B.; resources, J.G., F.M., M.C. and C.L.; writing—original draft preparation, C.R. and A.S.; writing—review and editing, C.R., C.L. and A.S.; supervision, C.L., A.C., J.G. and M.C.; project administration, C.L., M.C., J.G. and F.M. All authors have read and agreed to the published version of the manuscript.

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

**Institutional Review Board Statement:** This study was conducted in accordance with the Declaration of Helsinki, approved by the Ethics Committee of the Università Cattolica del Sacro Cuore (Ref. 22858/18) and registered at ClinicalTrials.gov (ID: NCT04009161).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy.

**Acknowledgments:** The authors would like to acknowledge the help received from Sunstar Europe S.A., Route de Pallatex 11, P.O. Box 32, 1163 Etoy, Switzerland. (prot. CA-19-0888).

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

### **References**


### *Article* **Comparative Analysis of Vascular Mimicry in Head and Neck Squamous Cell Carcinoma: In Vitro and In Vivo Approaches**

**Roosa Hujanen 1, Rabeia Almahmoudi 1, Tuula Salo 1,2,3,4,5,† and Abdelhakim Salem 1,2,3,\*,†**


**Simple Summary:** Head and neck squamous cell carcinomas (HNSCCs) are common and among the deadliest neoplasms worldwide, wherein metastasis represents the main cause of the poor survival outcomes. Tumour cells require blood vessels in order to grow and invade the surrounding tissues. Recently, a new phenomenon termed vascular mimicry (VM) was introduced, whereby tumour cells can independently form vessel-like structures to promote their growth and metastasis. VM has been characterized in many solid tumours, including HNSCC. A large body of research evidence shows that patients with positive VM exhibit poor treatment response and dismal survival rates. Thus, VM represents a promising therapeutic and prognostic target in cancer. However, there is limited knowledge regarding the identification of VM in HNSCC (in vitro and in vivo) and what factors may influence such a phenomenon. This study aims to address these limitations, which may facilitate the therapeutic exploitation of VM in HNSCC.

**Abstract:** Tissue vasculature provides the main conduit for metastasis in solid tumours including head and neck squamous cell carcinoma (HNSCC). Vascular mimicry (VM) is an endothelial cell (EC)-independent neovascularization pattern, whereby tumour cells generate a perfusable vessel-like meshwork. Yet, despite its promising clinical utility, there are limited approaches to better identify VM in HNSCC and what factors may influence such a phenomenon in vitro. Therefore, we employed different staining procedures to assess their utility in identifying VM in tumour sections, wherein mosaic vessels may also be adopted to further assess the VM-competent cell phenotype. Using 13 primary and metastatic HNSCC cell lines in addition to murine- and human-derived matrices, we elucidated the impact of the extracellular matrix, tumour cell type, and density on the formation and morphology of cell-derived tubulogenesis in HNSCC. We then delineated the optimal cell numbers needed to obtain a VM meshwork in vitro, which revealed cell-specific variations and yet consistent expression of the EC marker CD31. Finally, we proposed the zebrafish larvae as a simple and costeffective model to evaluate VM development in vivo. Taken together, our findings offer a valuable resource for designing future studies that may facilitate the therapeutic exploitation of VM in HNSCC and other tumours.

**Keywords:** head and neck squamous cell carcinoma; vascular mimicry; Matrigel; Myogel; zebrafish; metastasis

### **1. Introduction**

Head and neck squamous cell carcinoma (HNSCC) includes tumours of the oral cavity, hypopharynx, oropharynx, nasopharynx, and larynx [1]. Overall, HNSCC represents one

**Citation:** Hujanen, R.; Almahmoudi, R.; Salo, T.; Salem, A. Comparative Analysis of Vascular Mimicry in Head and Neck Squamous Cell Carcinoma: In Vitro and In Vivo Approaches. *Cancers* **2021**, *13*, 4747. https:// doi.org/10.3390/cancers13194747

Academic Editors: Carlo Lajolo, Gaetano Paludetti and Romeo Patini

Received: 17 August 2021 Accepted: 20 September 2021 Published: 23 September 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 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 (https:// creativecommons.org/licenses/by/ 4.0/).

of the most common cancers worldwide with relatively poor survival outcomes that remain stagnant at around 50% [2]. Such dismal prognosis of HNSCC patients has been largely attributed to tumour cell invasiveness and metastasis [3,4]. Thus, a better understanding of the different mechanisms and patterns underlying tumour cell dissemination could improve the management and survival outcomes of HNSCC patients.

Vascular mimicry (VM; a.k.a. vasculogenic mimicry) is a newly described pattern of tumour-related neoangiogenesis, whereby aggressive tumour cells can form tube-like vascular networks independently of endothelial cells [5,6]. These de novo VM structures were first described in patients with aggressive melanoma. Shortly thereafter, myriad studies have revealed many interesting characteristics of VM in various cancers including HNSCC [7,8]. In addition to satisfying the nutrient need of the primary tumour, VM is believed to provide tumour cells with an alternative route to intravasate and undergo metastasis [9,10]. In this regard, VM was shown to efficiently drive tumour cell metastasis in a polyclonal mouse model of breast cancer [10]. Furthermore, several studies revealed significant association between VM and lymph node metastasis (LNM) and hence worse prognosis in numerous malignancies [11–13]. We showed in a recent meta-analysis study that HNSCC patients with VM+ve tumours had shorter overall survival and worse clinicopathological features, including LNM, compared with the VM-ve group [8].

A vessel-like structure expressing CD31-ve/periodic acid–Schiff (PAS)+ve staining is often considered the "golden" standard to identify VM in histological samples [14]. However, in spite of the spirited debate ignited by this phenomenon, characterizing VM in patient samples has recently drawn criticism for its limitations. On the one hand, such CD31-ve/PAS+ve structures may represent irrelevant glycogen-rich areas rather than true mimetic vessels [14–16]. On the other hand, the mosaic vessels, concurrently expressing endothelial and tumour cell markers, have been overlooked in HNSCC-related studies, which show limited approaches to identify VM both in vitro and in vivo [7,8]. Therefore, we conducted a comprehensive comparative analysis of VM formation in HNSCC using a variety of procedures. We also proposed the zebrafish larvae as a feasible tool to model VM formation in vivo.

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

### *2.1. Patient Samples*

This study was approved by the National Supervisory Authority for Welfare and Health (VALVIRA) and the Ethics Committee of the Northern Ostrobothnia Hospital District. Our study comprised patients diagnosed with oral tongue SCC (OTSCC) who had undergone surgery in Oulu University Hospital during the period 1990–2010. Formalin-fixed paraffin-embedded (FFPE) samples (*n* = 30) were obtained from the pathology department of Oulu University Hospital. None of these patients had received other prior treatments.

### *2.2. CD31 and PAS Double Staining*

The FFPE specimens were deparaffinized and rehydrated and subjected to heatinduced antigen retrieval using Micromed T/T Mega Microwave Processing Lab Station (Hacker Instruments & Industries). Non-specific binding was blocked with Dako peroxidase blocking solution S2023 for 15 min (Dako, Glostrup, Denmark), followed by incubation in a 1:100 polyclonal rabbit anti-CD31 antibody (ab28364; Abcam, Cambridge, UK) for 1 h. Sections were then incubated with horseradish peroxidase for 30 min; treated with DAB (Pierce™ DAB Substrate Kit; Thermo Fisher Scientific; Waltham, MA, USA) for 5 min; and incubated with 0.5% freshly made periodic acid for 10 min. Sections were further stained with Schiff solution for 15 min and rinsed under running water for another 15 min. Slides were incubated with Cole's hematoxylin for 6 min and mounted in Mountex (HistoLab, Gothenburg, Sweden). All incubations were conducted at room temperature.

### *2.3. Double-Labelling Immunofluorescence (IF)*

Following deparaffinization and heat-induced antigen retrieval, sections were blocked for 1 h with 10% donkey normal serum (Sigma-Aldrich; St. Louis, MO, USA). Sections were then incubated overnight with a primary antibody solution containing 1:50 polyclonal rabbit anti-CD31 antibody (ab28364, Abcam, Cambridge, UK) and 1:100 monoclonal mouse antihuman pan-cytokeratin (CK) (M3515, Dako, Glostrup, Denmark) at 4 ◦C. The following day, the sections were incubated in (1) 1:200 donkey anti-mouse Alexa Fluor®-568 or donkey anti-rabbit Alexa Fluor®-488 conjugated secondary antibodies (Vector Laboratories; Burlingame, CA, USA) for 1 h and (2) 4 ,6-diamidino-2-phenylindole (DAPI; 1:1000; Sigma-Aldrich, St. Louis, MO, USA) for 10 min, and mounted with ProLong® Gold Antifade Mountant (Thermo Fisher Scientific; Waltham, MA, USA). To stain the cell-derived tubular networks, matrix-coated coverslips were fixed for 20 min in 4% paraformaldehyde (PFA; Santa Cruz Biotech., Santa Cruz, CA, USA) and then staining was continued as above. All steps were performed at room temperature unless otherwise indicated. For multiplexed immunohistochemistry (mIHC), the following antibodies were used: 1:50 polyclonal rabbit anti-CD31 antibody (ab28364, Abcam, Cambridge, UK); 1:100 monoclonal mouse anti-CD44 antibody (144M-95; Cell Marque, Rocklin, CA, USA); 1:100 monoclonal mouse antihuman E-cadherin antibody (M361201, Dako, Glostrup, Denmark); monoclonal mouse anti-CK c11 (ab7753, Abcam, Cambridge, UK); and 1:150 monoclonal mouse anti-CK (AE1/AE3; Dako, Glostrup, Denmark). mIHC was performed in the Digital Microscopy and Molecular Pathology Unit (FIMM Institute, University of Helsinki) as described previously [17].

### *2.4. Cell Line and Culture*

Thirteen primary and metastatic HNSCC cell lines were used, including HSC-3 (JCRB 0623; Osaka National Institute of Health Sciences, Japan), SCC-25 (ATCC, Rockville, MD, USA) and SAS (JCRB-0260). Ten cell lines (UT-SCC, hereafter SCC) were established directly from the patient biopsy material at the Department of Otorhinolaryngology, Head and Neck Surgery Unit, Turku University Hospital (Table S1). Of these, paired primary and metastatic cell lines (SCC-24A and -24B, respectively) were obtained from the same patient. The SCC-28 cell line was derived from a primary tumour that was first treated with radiotherapy prior to surgical resection. Cancer cell lines were cultured in 1:1 DMEM-F12 medium (Gibco/Invitrogen, Tokyo, Japan) supplemented with 10% heat-inactivated fetal bovine serum (Gibco), penicillin–streptomycin (15140-122, Thermo Fisher Scientific; Waltham, MA, USA), 50 μg/mL ascorbic acid (A1052, AppliChem, Chicago, IL, USA), 250 ng/mL amphotericin B (A2942, Sigma-Aldrich, St. Louis, MO, USA) and 0.4 μg/mL hydrocortisone (H0888, Sigma-Aldrich, St. Louis, MO, USA). Cell lines were maintained in a 95% humidified incubator of 5% CO2 at 37 ◦C. Human umbilical vein endothelial cells (HUVEC; Thermo Fisher Scientific; Waltham, MA, USA) were used as a positive control for the in vitro tubulogenesis. HUVECs were cultured in 200PRF medium supplemented with a low serum growth supplement (Thermo Fisher Scientific; Waltham, MA, USA).

### *2.5. Murine and Human-Derived 3D Matrices*

We used the commercial mouse Engelbreth–Holm–Swarm (EHS) sarcoma matrix, Matrigel (Corning, NYC, NY, USA). In addition, we used our in-house gelatinous soluble matrix "Myogel" that is derived from human leiomyoma tissue [18,19]. The preparation and usage of human leiomyoma tissue have been approved by the Ethics Committee of Oulu University Hospital (no. 35/2014). Liquid handling was performed using MultiFlo™ FX automated multi-mode reagent dispenser (BioTeK, Winooski, VT, USA).

### *2.6. In Vitro Tube Formation Assay*

The in vitro tube formation assay was performed according to a previously published protocol [20]. For the Matrigel-based assay, following a slow overnight thawing at 4 ◦C, 50 μL of Matrigel was dispensed into a 96-well plate and incubated for 30 min at 37 ◦C. Cancer cells and HUVECs were detached from 75 cm<sup>2</sup> flasks (Sigma-Aldrich, St. Louis, MO, USA) with trypsin–EDTA, resuspended in serum-free DMEM or 200PRF medium, and then counted using Scepter™ 2.0 Cell Counter (Merck Millipore, Burlington, MA, USA). Cells were seeded on the top of Matrigel at a starting density of 20 × <sup>10</sup><sup>3</sup> in 50 <sup>μ</sup>L serum-free medium and incubated at 37 ◦C.

For the Myogel-based assay, the optimal gel concentration (1 mg/mL) was determined using pilot experiments with HUVECs. The Myogel-fibrin matrix was prepared with serum-free medium using the following concentrations: 1 mg/mL Myogel, 1 mg/mL fibrinogen (341578, Merck, Darmstadt, Germany), 66.67 μg/mL aprotinin (A6279-10ML, Merck, Darmstadt, Germany), and 0.6 U/mL thrombin (T6884-100UN, Sigma-Aldrich, St. Louis, MO, USA). To test the potential effects of different gel constituents, Myogel was also combined with 1 mg/mL rat-tail type I collagen (354236; Corning, NYC, NY, USA) or 1–2% low-melting agarose (LMA; 50101, Lonza, Basel, Switzerland). In total, 50 μL of LMA was slowly pipetted into a 96-well plate to avoid bubbles and incubated overnight at 37 ◦C. The next day, Myogel was pipetted with the cells into the LMA-coated wells. The matrix-coated well plates were incubated for 12 and 24 h for endothelial- and tumour cell-derived tubulogenesis, respectively. The wells were then rinsed in phosphate-buffered solution (PBS), fixed for 20 min with 4% PFA, and stored in 4 ◦C.

### *2.7. Zebrafish Larvae Assays*

In vivo zebrafish experiments were performed in the zebrafish core facility at the University of Helsinki. All procedures were approved by the ethical committee of the regional state administrative agency (license ESAVI/13139/04.10.05/2017). Two-day postfertilization zebrafish larvae were dechorionated and anaesthetized using 0.04% Tricaine (*n* = 10 per matrix group). Fluorescently labelled with CellTrace™ Far Red (Thermo Fisher Scientific; Waltham, MA, USA), HSC-3 cells were washed in PBS and resuspended in 1:1 Matrigel or Myogel, and then microinjected into the perivitelline space using glass microinjection needles (about 1000 cells). Fish were maintained at 34 ◦C within an embryonic medium (Sigma-Aldrich, St. Louis, MO, USA) for 72 h and then collected, fixed with 10% PFA, and mounted using SlowFade Gold Antifade (Invitrogen, Carlsbad, CA, USA).

### *2.8. Imaging and Tube Formation Analysis*

For experiments on tube formation, samples were photographed with magnifications of 4×, 10× and 20× using the reverse Nikon Digital Sight DS-U3 microscope (Nikon, Tokyo, Japan). Each experiment was repeated at least three times independently. Stained section images were acquired with magnifications of 10×, 20×, and 40× using a Leica DM6000 microscope (Leica Microsystems, Wetzlar, Germany). Imaging of zebrafish larvae was performed at the Biomedicum Imaging Unit (University of Helsinki) using a Leica TCS SP8 confocal microscope. The ImageJ software (Wayne Rasband, National Institute of Health, Bethesda, MD, USA) was used by applying the "Angiogenesis Analyzer" plugin to measure several different parameters for evaluating the tube formation as described in the Results section.

### **3. Results**

### *3.1. Utility of the CD31-ve/PAS+ve Reaction in Identifying the VM in HNSCC Patients*

To identify the VM in the patient samples, we first employed the traditional staining method—a combination of the endothelial cell marker and PAS staining on FFPE sections from HNSCC patients (Figure 1A). PAS stains basement membrane components such as laminin, collagen, and glycogen, whereas CD31 was opted as a specific endothelial cell marker. Areas of PAS+ve laminin and collagen-rich networks (pink) with the surrounding tumour cells were recognized in the patient samples. Additionally, CD31+ve/PAS+ve endothelial vessels (brown/pink; Figure 1B,C) and CD31-ve/PAS+ve areas (pink; Figure 1D) were identified. However, it was often onerous to accurately identify the CD31-ve/PAS+ve structures due to the presence of necrotic areas or faint CD31 signals that can be easily masked by the surrounding PAS staining (Figure 1E; arrows show a faint signal of CD31).

**Figure 1.** Identification of vascular mimicry (VM) in tumour tissues. (**A**) Representative figures from tumour sections (*n* = 30) obtained from patients with oral tongue squamous cell carcinoma (OTSCC) and stained using a combination of endothelial cell (EC) marker (CD31) and periodic acid–Schiff (PAS) staining. (**B,C**) Normal blood vessels express CD31+ve/PAS+ve (brown/pink). Scale bar: 50 μm (**D,E**) Additionally, some CD31-ve/PAS+ve vessel-like structures (pink) were identified. However, identifying these structures was often demanding due to the presence of necrotic areas or a faint CD31 signal (black arrows). Scale bar: 50 μm. (**F**) The double-labelled immunofluorescence assay was employed using a combination of CD31 and tumour cell marker (pan-cytokeratin, CK) to investigate the presence of mosaic vessels in HNSCC sections. (**G**) EC-lined blood vessels were easily distinguished in the peritumoral areas (dashed yellow line). Scale bar: 50 μm (**H**) The intratumoral CD31+ve/CK+ve mosaic vessels were also observed (dashed yellow line; white arrow). Scale bar: 50 μm.

(**I**) These mosaic lumens were either containing RBCs, metastasizing tumour cells or clear. Scale bar: 25 μm. (**J**) The multiplexed immunohistochemistry (mIHC) platform was used to identify VM (merged; inset) as well as to explore the phenotype of VM-forming tumour cells. A negative or weak staining of CD44 was observed in morphologically normal tissues (left), while a strong staining was detected in the VM-forming regions (right). By contrast, E-cadherin (E-Cad) staining was evident in normal and VM-free cancerous tissues (left, red), while it was faint around the mosaic vessels (right). The mIHC images were taken at a magnification of 63×.

### *3.2. The Mosaic VM Pattern Reveals Tumour Cell Plasticity*

Recently, we showed that oral tongue squamous cell carcinoma (OTSCC) cells express considerable levels of the endothelial marker CD31 in vitro [21]. Due to the limited utility of the CD31-ve/PAS+ve reaction in identifying VM in HNSCC tissues, we sought to explore the presence of intratumoral mosaic vessels using CD31+ve/CK+ve double-labelled immunofluorescence (Figure 1F). Normal blood vessels were easily distinguished as CD31 expressing lumens mainly in the peritumoral stroma (Figure 1G). Interestingly, OTSCC patient samples revealed distinct and clear intratumoral CD31+ve/CK+ve mosaic VM lumens, which also contain red blood cells (Figure 1H, arrow). It has been well reported that VM formation is associated with phenotype switching or "cell stemness" (i.e., tumour cell plasticity), which is mediated by certain events such as upregulation of CD44 and loss of epithelial cell markers including E-cadherin [22,23]. This observation prompted us to explore whether the mosaic vessels can also be harnessed to examine the status of these phenotype mediators. Importantly, using the mIHC platform, the mosaic CD31+ve/CK+ve structures revealed an induced CD44-immunoreactivity, while E-cadherin staining was noticeably weaker around the mosaic vessels compared with tumoral VM-free regions (Figure 1I).

### *3.3. Metastatic HNSCC Cells Preferentially form VM in Matrigel*

Previous pioneering studies have shown that cancer cells can form VM capillary networks similar to the endothelial tubulogenesis when cultured on a collagen-rich matrix [24]. However, there is very limited knowledge concerning the effect of the extracellular matrix (ECM) on such a phenomenon. Therefore, after identifying the VM structures in patient samples, we explored whether matrix origin and constituents can influence VM formation in vitro. To this end, 13 primary and metastatic HNSCC cell lines plus HUVEC were seeded on murine- and human-derived matrices at a density of 20 × <sup>10</sup><sup>3</sup> cells/well, as described previously [20]. Of note, all cell lines with high metastatic potential (*n* = 3) formed capillary networks in Matrigel but not in Myogel. By contrast, the primary cell lines showed a greater tendency to form VM in Myogel (*n* = 4) compared with Matrigel (*n* = 2; Figure 2A). Nevertheless, HUVEC formed consistent tubes in both matrices, suggesting that ECM could be an important modulator of the tumour cell-derived tubulogenesis. At this cell density, the tubes were, however, poorly networked and occupied less than half of the matrix area and were then scored (+) as illustrated in Table 1. Combining Myogel with collagen I or LMA did not noticeably alter VM formation.

**Figure 2.** Assessment of tumour cell-derived tubulogenesis in vitro. (**A**) Thirteen cell lines derived from head and neck squamous cell carcinoma (HNSCC) and normal human endothelial cells (HUVEC) were seeded at a starting cell density of <sup>20</sup> <sup>×</sup> <sup>10</sup><sup>3</sup> cells on Matrigel or Myogel. All of the highly metastatic HNSCC cell lines (namely, SCC-24B, HSC-3, and SAS) formed a VM meshwork in Matrigel only, while more primary cell lines formed such tubes in Myogel (*n* = 4). HUVEC formed tubes in both matrices. (**B**) At a higher starting cell density of 40 <sup>×</sup> <sup>10</sup>3, all metastatic and some primary HNSCC cell lines (*n* = 7) developed longer and more interlaced VM meshwork in Matrigel. In Myogel, primary tumour cell lines (*n* = 5) continued to form VM structures with the most extensive meshwork attained by cells originating from the floor of the mouth and gingiva (SCC-28 and SCC-44, respectively). Additionally, the metastatic cell line (SCC-24B) started to initiate consistent tubes in Myogel that were more extensive than its primary counterpart (SCC-24A). (**C**) When the starting cell density reached 60 <sup>×</sup> 103, only metastatic and merely two primary cell lines continued to develop thicker and longer VM networks in Matrigel compared with four primary cell lines in Myogel. (**D**) The HUVEC meshwork has become shorter and more interlaced in Myogel. The images were taken at a magnification of 4×.


**Table 1.** Vascular mimicry-like network formation for head and neck squamous cell carcinoma and human endothelial cell lines in two different matrices.

<sup>1</sup> Data from one representative experiment of at least three independent experiments are shown; starting cell density was as follows: **<sup>A</sup>** = 20 × <sup>10</sup>3; **<sup>B</sup>** = 40 × <sup>10</sup>3; **<sup>C</sup>** = 60 × <sup>10</sup><sup>3</sup> cells/well. Score description: (-) = no tube formation; (+) = poorly interconnected capillary networks that covered less than half of the matrix surface; (++) = cells formed well-defined interconnected capillary networks that covered not more than half of the matrix surface; (+++) = cells formed clear, well-defined interconnected capillary networks that covered more than half of the matrix surface. HUVEC: human umbilical vein endothelial cells.

### *3.4. Tumour Cell Density Influences VM Formation In Vitro*

Tumour cell-derived tubulogenesis is an important assay not only for assessing VM formation but also for testing potential anti-angiogenic drugs in vitro [21]. It is therefore important to discern the optimal number of tumour cells needed to establish mature capillaries for HNSCC-related studies. For this purpose, HNSCC cell lines were seeded on Matrigel or Myogel using different starting cell densities of 20, 40, and 60 × <sup>10</sup><sup>3</sup> cells. Notably, all metastatic and some primary HNSCC cell lines (*n* = 7) formed longer and well-networked VM structures at 40 × <sup>10</sup><sup>3</sup> cells, which covered approximately half of the Matrigel (score ++; Figure 2B). Furthermore, only metastatic (*n* = 3) and merely two primary cell lines developed thicker and longer capillary networks when tumour cell density reached 60 × <sup>10</sup>3, which spread to more than half of the Matrigel (score +++; Figure 2C). In Myogel, primary tumour cell lines (*n* = 5) continued to form VM structures with the most extensive networks attained by cells originating from the floor of the mouth and gingiva (SCC-28, grade 1; SCC-44, grade 3, respectively; Figure 2B,C). Of interest, at higher cell densities, the metastatic cell line (SCC-24B) started to initiate consistent VM

networks in Myogel that were more extensive than its primary counterpart (SCC-24A; Figure 2B,C). Moreover, the HUVEC meshwork has apparently become shorter and more interlaced in Myogel (Figure 2D). Two primary cell lines (SCC-73 and SCC-25) failed to form VM networks in either matrices, which remained dispersed in the matrices as round cell aggregates (Figure S1). The scores of VM formation using different cell densities and matrices are listed in Table 1.

Next, we used the Angiogenesis Analyser tool to assess the comparative capacity of various HNSCC cells in forming VM capillaries in vitro [25]. Angiogenic parameters including the number of junctions (branching capillary nodes), segments (capillaries delimited by two junctions), meshes (areas enclosed by segments), total meshes area (sum of mesh areas), total segments length (length sum of all segments), and total branching length were quantified. It is worth noting that the pre-irradiated SCC-28 cells formed unique "spiky" capillaries that spread evenly on Matrigel, regardless of their cell density (Figure 3A). Overall, starting cell densities of 40 and 60 × <sup>10</sup><sup>3</sup> cells were both adequate for initiating proper tubulogenesis in vitro; however, the latter density produced more looping meshwork in most cell lines (Figure 3A,B).

### *3.5. In Vitro VM Networks Reveal Different Morphological Patterns*

Interestingly, tumour cells from different head and neck regions formed varying morphological patterns of VM networks in their respective matrix. While metastatic OTSCC cell lines (e.g., HSC-3 and SCC-24B) had the typical "honeycomb-like" pattern, the larynx-derived primary cell line (SCC-8) attained thinner and somewhat smoother capillary extensions (Figure 4A). On the other hand, cells derived from the floor of the mouth (i.e., SCC-28) formed peculiar capillary networks with thick "spike-like" projections in the two matrices (Figure 4A).

### *3.6. In Vitro VM Networks Express Endothelial Cell Marker*

Having determined the optimal cell density to establish VM in 3D matrices, we next sought further in vitro verification that HNSCC cells, rather than endothelial cells, were responsible for the observed mosaic pattern in the clinical samples. Hence, tumour cell-derived VM networks on phenol red-free Matrigel were stained with the endothelial cell marker CD31. To unambiguously localize CD31 in relation to tumour cell junctions, VM capillaries were also labelled with Phalloidin–Alexa-594 to stain F-actin networks. Interestingly, tumour cell-derived VM capillaries clearly expressed CD31, which was mostly localized in the tubular extensions and around the capillary junctions (Figure 4B).

### *3.7. Larval Zebrafish as a Novel In Vivo Model for VM Formation*

Testing VM formation in vivo is currently conducted in patient-derived murine xenografts [24]. However, such models can present substantial challenges, including time consumption and cost and labour intensiveness. Therefore, we assessed the utility of zebrafish larvae as a simple and yet efficient approach to optically screening the formation of VM structures in vivo. Fluorescently labelled aggressive tumour cells (HSC-3) were resuspended in their respective matrix and microinjected into the perivitelline space of anaesthetized zebrafish (Figure 5A). Using confocal microscopy at 72 h post-injection, the xenografted tumour cells displayed VM-like structures in some of the fish (*n* = 3) belonging to the Matrigel-injected group, which attained singular or multi-tubular pattern (Figure 5B,C). By contrast, no similar structures were observed in the Myogel-injected group (Figure 5D).

**Figure 3.** In vitro tube formation analysis of tumour cell-derived vascular mimicry (VM). (**A**) The Angiogenesis Analyser plugin was used to discern the optimal starting cell density needed to establish VM meshwork in vitro. Different tube formation parameters were analysed, including the number of junctions, segments, meshes, total mesh area, total segment length, and total branching lengths of the tubular networks. A starting cell density of 60 <sup>×</sup> 103 produced consistent mature looping patterns in Matrigel for almost all the included cell lines. However, overall, the analysis shows that both 40 and 60 <sup>×</sup> 103 concentrations are sufficient to initiate VM structures in vitro. (**B**) The results were comparable in Myogel, with better tubes formed with a starting cell density of <sup>60</sup> <sup>×</sup> 103. However, at such a higher density, HUVEC meshwork became more extensively interlaced in Myogel, which limited the analyser's capacity to recognize smaller tubular areas as shown in the figure. Data from one representative experiment, presented as mean ± SD of three technical replicates, are shown.

**Figure 4.** (**A**) Tumour cells show distinct morphological patterns of tubulogenesis in vitro. The highly metastatic tongue cancer cell lines (HSC-3 and SCC-24B) formed the classical "honeycomb-like" looping pattern, while the larynx-derived cell line (SCC-8) attained thinner branches with smoother capillaries. Evidently, cells from the floor of the mouth (SCC-28)

formed peculiar and "spike-like" networks, on both Matrigel and Myogel, that were morphologically different from any other cell line. (**B**) In vitro tumour cell-derived tubulogenesis showed substantial resemblance to the endothelial ones. These cell networks expressed the endothelial cell marker CD31, which was primarily localized in the capillary extensions and junctions. The images were taken at magnifications of 10× and 20×.

**Figure 5.** Larval zebrafish model to evaluate vascular mimicry (VM) formation in vitro. (**A**) Fluorescently labelled highly metastatic tumour cells (HSC-3) were resuspended in Matrigel or Myogel and microinjected into the perivitelline space of 2-day-old zebrafish larvae. Fish were screened 27 h post-injection using confocal microscopy. (**B**,**C**) HSC-3 cells formed seemingly VM-like structures in the Matrigel-injected fish. (**D**) No similar tube formation was detected in Myogel-containing fish, supporting a similar outcome from the in vitro assays. Scale bars: 20 and 50 μm.

### **4. Discussion**

The VM has been well documented in a variety of cancers and is associated with a stemlike cell phenotype, aggressive disease course, and dismal survival outcomes [9,10,13,26]. However, the currently available approaches to identify VM in HNSCC are rather limited, thereby necessitating more research on this intriguing phenomenon [7,8]. In this study, we first revealed some challenges associated with identifying VM in HNSCC sections, wherein the mosaic vessels could be adopted to further assess the phenotype of VM-forming cells. Next, we reported the impact of ECM origin, tumour cell type, and density on the formation and morphology of HNSCC cell-derived tubulogenesis. We then delineated the optimal cell numbers needed to obtain such tubular meshwork in vitro, which also expressed the specific endothelial cell marker—CD31. Finally, we proposed for the first time a simple animal model, the zebrafish larvae, for assessing the development of VM in vivo.

Histologically, VM structures are often identified in cancer patients as PAS+ve, RBCcontaining, lumen-like structures combined with a negative staining of an endothelial cell marker [14]. However, PAS stains various ECM components including collagens, laminin, and proteoglycans and hence may not always represent the vascular mimetic structures. Using an X-ray microtomography 3D reconstruction, Racordon et al. showed that many PAS+ve areas do not display actual lumens in vitro and may instead represent glycoproteinrich regions [16]. It has therefore been recommended to be attentive when scoring PAS+ve areas to differentiate VM from non-specific ECM aggregates [7,14]. Furthermore, a strong

PAS staining may conceal the expression of endothelial cell markers, making it challenging to discern CD31-ve/ PAS+ve patterns. In a different approach, several reports described the existence of "mosaic" vessels expressing both tumour and endothelial cell markers in cancer tissues, emphasizing the importance of tumour cell plasticity in VM formation [6,27–29]. Initially, these vessels were thought to result from endothelial and tumour cell merging in blood vessel walls. However, it was later shown that tumour cells are able to form and maintain blood vessels by expressing neuropilin-2, EphA2, and laminin-15γ2 [28]. An interesting study revealed that 20–90% of the vascular endothelium in glioblastoma was derived from VM-forming tumour cells in mice; their selective targeting resulted in tumour reduction and degeneration [26]. Supporting these findings, Kim et al. found that the intratumoral VM channels were derived from CD31+ve/CD34+ve gastric tumour cells [30]. Furthermore, we recently showed, by fluorescence-activated cell sorting, that 90% of the HSC-3 cells were CD31+ve, compared with 96% of HUVEC [21]. In this study, we manifested this expression phenotypically by showing that tumour cell-derived tubes are CD31+ve with a striking resemblance to the endothelial ones. These findings suggest that intratumoral mosaic vessels may represent an additional staining approach to identifying patterned VM structures in cancer tissues.

Using the mIHC platform, the adhesion molecule CD44—a transmembrane glycoprotein receptor known to promote tumour cell plasticity—and VM were induced around the mosaic VM-forming cells [31]. A tumour cell plasticity is best seen in crucial metastatic processes such as epithelial mesenchymal transition, wherein tumour cells lose their adhesion, polarity, and epithelial cell markers including E-cadherin [32]. It is therefore interesting that mosaic VM-forming regions revealed a faint expression of E-cadherin compared with other epithelial regions. Our results advocate the use of mIHC for the simultaneous assessment of different markers associated with the development of VM.

Previous seminal works on VM have shown that aggressive cancer cells can form tubular networks when seeded on Matrigel. However, it is worth noting that considerable variations exist among different matrices based on their origin, composition, and consistency. Our findings suggest that tumour cell-derived tubulogenesis could be influenced, in part, by the matrix type. Although it is not yet clear why tumour cells have a matrixspecific ability to form VM, variations in ECM features may underpin this interesting observation. For instance, the protein composition of Myogel is substantially different from other EHS-based matrices. Further, crucial carcinogenesis-related properties, such as tumour cell invasion and response to HNSCC-targeted therapy, were more efficiently represented in Myogel than in Matrigel [18,33]. A fascinating observation is that a primary cell line (SCC-24A) formed merely a few tubes compared with an extensively interlaced network formed by its metastatic counterpart (SCC-24B), albeit both were established from the same patient. This confirms previous studies showing that VM is associated with metastatic and highly aggressive tumours. Additionally, we infer that VM competence can differ even within the same patient, signifying the need for more precise targeting of anti-angiogenic therapies. The spike-like pattern formed by tumour cells from the floor of the mouth is another intriguing observation. Interestingly, this particular cell line (SCC-28) was established from a tumour that was treated with radiotherapy prior to surgical resection. In this context, there is abundant evidence that ionizing radiation targeting cancer cells may enhance their metastatic process [34]. Additionally, the floor of the mouth is the most high-risk site for metastasis in oral cancer patients. Thus, it has been recommended that patients with SCC in this site should be offered an elective neck dissection even at early stages of the disease [35]. We encourage further research to investigate whether this peculiar tubular morphology plays a role in the metastatic potential of HNSCC and whether radiation therapy could impact VM formation. Consistent with previous data, endothelial cells formed shorter and much interlaced networks in Myogel [18].

Cancer cell-derived tubulogenesis is a valuable assay not only to evaluate VM formation in vitro but also for testing potential anti-angiogenic therapeutic approaches in HNSCC [21]. Therefore, we presented an easy standardized protocol to establish mature

capillary networks using a good number and a variety of HNSCC cell lines. Previous studies reported similar approaches to estimating such an optimal tumour cell number, for instance, in human ovarian cancer cell lines [16]. In this study, authors provided solid evidence that VM tubes in vitro represent in most cases functional hollow channels. Additionally, 15 and 75 × <sup>10</sup><sup>3</sup> starting numbers of the ovarian cancer cells produced clear tubular formation on day 4 of the experiment. In another protocol, Francescone et al. suggested a starting density of 10 − <sup>20</sup> × <sup>10</sup><sup>3</sup> cells using melanoma, glioblastoma, and breast cancer cell lines [20]. In the present HNSCC cell lines, comparable starting cell densities (20, 40, and 60 × <sup>10</sup><sup>3</sup> cells) were used to obtain VM channels within 24 h in culture, confirming that this phenomenon could vary based on the tumour cell type.

Zebrafish larvae have recently emerged as a popular in vivo model of HNSCC to mimic key tumorigenic events such as metastasis [36]. Indeed, zebrafish provides many advantages over other animal models considering its efficiency, feasibility, and cost- and labour-effectiveness [37]. Currently, most in vivo model systems of VM are conducted in murine xenografts. However, in addition to cost and labour challenges, screening of VM in these models can be made only post-mortem, restricting further follow-up studies [24]. Thus, we proposed the use of zebrafish larvae as a simple and cost-effective in vivo model of VM. Although VM-like structures were observed in some xenografts, their formation should be interpreted with caution as there is no evidence indicating that they represent actual lumens. In addition, it is not clear why these structures were not formed in all xenografts. Such disparity in the formation of in vivo mimetic vessels has been nonetheless observed in murine xenografts [36]. However, several technical limitations may arise when using the larval zebrafish model. Firstly, there is a possibility of tumour cell leakage out of the fish due to poor resealing of the yolk sac membrane. Secondly, their smaller body size restricts the number of microinjected cells and the resulting tumour size compared with larger animal models. Finally, larval assays are performed at 34 ◦C, which may not be suitable for some cell lines and hence fail to form proper tumour colonies [38]. Further studies would be paramount to optimizing this model and testing its feasibility for real-time imaging as well as for therapeutic and functional assays.

### **5. Conclusions**

In conclusion, our study provides a comprehensive comparative analysis of VM in HNSCC using a variety of experimental approaches. We, however, acknowledge some limitations, including the lack of perfusion assays to assess the functionality of the tubular networks, which has already been revealed in previous reports. Overall, our findings could offer a valuable resource for designing future studies that may facilitate the therapeutic exploitation of VM in HNSCC as well as in other recalcitrant tumours.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/cancers13194747/s1: Figure S1: Two primary cell lines (SCC-73 and SCC-25) were not able to initiate own tubulogenesis in either Matrigel or Myogel, where they remained dispersed as round cell aggregates (A, 40 <sup>×</sup> 103 cells; B, 60 <sup>×</sup> 103 cells); Table S1: Patient-derived cell lines and their corresponding data.

**Author Contributions:** Conceptualization, T.S. and A.S.; methodology, R.H., R.A. and A.S.; software, R.H. and R.A.; validation, R.H., R.A., T.S. and A.S.; formal analysis, R.H., R.A., T.S. and A.S.; investigation, R.H., R.A. and A.S.; resources, T.S. and A.S.; data curation, R.H. and R.A.; writing original draft preparation, R.H.; writing—review and editing, A.S.; visualization, R.H. and A.S.; supervision, T.S. and A.S.; project administration, T.S. and A.S.; funding acquisition, T.S. and A.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Jane and Aatos Erkko Foundation; the Minerva Foundation Institute for Medical Research; Cancer Society of Finland; Sigrid Jusélius Foundation; Helsinki University Central Hospital research funds; and an Oulu University Hospital MRC grant.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of the National Supervisory Authority for Welfare and Health (VALVIRA) and the Northern Ostrobothnia Hospital District. The preparation and usage of human leiomyoma tissue has been approved by the Ethics Committee of Oulu University Hospital (no. 35/2014; 28.4.2014). Zebrafish procedures were approved by the ethical committee of the regional state administrative agency (license ESAVI/13139/04.10.05/2017).

**Informed Consent Statement:** The patient sections in this study were obtained a long time ago (1990– 2010), and no consent forms were possible to obtain in this case. Therefore, the Finnish National Supervisory Authority for Welfare and Health (VALVIRA) has approved the use and publishing these patient materials. In addition, we confirm that there is NO use of any "identifying" personal information or images of the patients in our article.

**Data Availability Statement:** The datasets used in this study are available from the corresponding author upon a reasonable request.

**Acknowledgments:** The authors would like to thank Antti Isomäki (Biomedicum Imaging Unit, University of Helsinki) for his kind assistance with confocal microscopy imaging and Wafa Wahbi for helpful discussions on the zebrafish assays. Multiplexed immunohistochemistry was performed in the Digital Microscopy and Molecular Pathology Unit (FIMM Institute, University of Helsinki). Zebrafish experiments were performed in the zebrafish core facility at the University of Helsinki. Open access funding provided by University of Helsinki.

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

### **References**


### *Article* **Sleep Disorders and Psychological Profile in Oral Cancer Survivors: A Case-Control Clinical Study**

**Roberta Gasparro 1, Elena Calabria 1,\*, Noemi Coppola 1, Gaetano Marenzi 1, Gilberto Sammartino 1, Massimo Aria 2, Michele Davide Mignogna 1,† and Daniela Adamo 1,†**


**Simple Summary:** Sleep disorders have been increasingly investigated in several medical illnesses as their presence may affect patients' quality of life. However, the research examining sleep disorders in oral cancer is relatively weak. Indeed, the majority of the available studies present a crosssectional or retrospective designs. Moreover, very few of them have evaluated quality of sleep in oral cancer survivors (OC survivors). We aimed to carry out a case-control study with the purpose to investigate sleep disorders and mood impairment in 50 OC survivors. Our research has shown that quality of sleep is significantly affected in OC survivors compared to a healthy population and that OC survivors suffers from higher levels of anxiety and depression. Our results may suggest that an appropriate assessment of quality of sleep and psychological profile should be performed in OC survivors as a prompt treatment for both sleep and mood disorders is crucial for the overall improvement of patients' quality of life.

**Abstract:** Quality of sleep (QoS) and mood may impair oral cancer survivors' wellbeing, however few evidences are currently available. Therefore, we aimed to assess the prevalence of sleep disorders, anxiety and depression among five-year oral cancer survivors (OC survivors). 50 OC survivors were compared with 50 healthy subjects matched for age and sex. The Pittsburgh Sleep Quality Index (PSQI), the Epworth Sleepiness Scale (ESS), the Hamilton Rating Scales for Depression and Anxiety (HAM-D, HAM-A), the Numeric Rating Scale (NRS), the Total Pain Rating Index (T-PRI) were administered. The global score of the PSQI, ESS, HAM-A, HAM-D, NRS, T-PRI, was statistically higher in the OC survivors than the controls (*p*-value: <0.001). QoS of OC survivors was significantly impaired, especially with regard to some PSQI sub-items as the subjective sleep quality, sleep latency and daytime dysfunction (*p*-value: 0.001, 0.029, 0.004). Moreover, poor QoS was negatively correlated with years of education (*p*-value: 0.042 \*) and positively correlated with alcohol consumption (*p*-value: 0.049 \*) and with the use of systemic medications (*p*-value: 0.044 \*). Sleep disorders and mood disorders are common comorbidities in OC survivors; therefore, early assessment and management before, during and after treatment should be performed in order to improve the quality of life of OC survivors.

**Keywords:** oral cancer; sleep disturbance; depression; anxiety; insomnia; oral cancer survivors; psychiatric profile

### **1. Introduction**

Oral cancer is a life-threatening disease and a burden for health care systems worldwide. According to Global Cancer Statistics, GLOBOCAN, there were 354,864 new cases

**Citation:** Gasparro, R.; Calabria, E.; Coppola, N.; Marenzi, G.; Sammartino, G.; Aria, M.; Mignogna, M.D.; Adamo, D. Sleep Disorders and Psychological Profile in Oral Cancer Survivors: A Case-Control Clinical Study. *Cancers* **2021**, *13*, 1855. https://doi.org/ 10.3390/cancers13081855

Academic Editors: Carlo Lajolo, Gaetano Paludetti and Romeo Patini

Received: 8 March 2021 Accepted: 8 April 2021 Published: 13 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 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 (https:// creativecommons.org/licenses/by/ 4.0/).

of oral cavity cancer causing 177,384 deaths during 2018 [1]. Despite the improvement in diagnosis and treatment by health care providers with a subsequent decrease in mortality, the quality of life of oral cancer survivors (OC survivors) remains poor on account of the impact of this disease on mental and emotional well-being. Indeed, oral cancer patients often suffer from emotional distress, fatigue, sleep disturbance, anxiety and depression that can arise during treatment and persist long-term, aggravating the burden of the disease [2,3].

Recently, a growing interest has been focused on the evaluation of sleep disorders in relation to several medical illnesses as their presence may worsen the underlying disease and increase the rate of mortality [4]. Furthermore, sleep disorders are considered to be an extremely sensitive marker for psychiatric comorbidities which may also precede mood disorders, especially depression or anxiety, and its early detection and treatment is crucial to improve the prognosis and quality of life of patients.

Insomnia is the most frequent sleep disorder; generally, patients report a difficulty in falling asleep, and often experience restless sleep and excessive daytime sleepiness (hypersomnolence) [5].

The overall incidence of insomnia in cancer patients has been found to be three times higher than that reported in the general population and ranges from 30.0% to 93.1%, depending on the type of cancer [6,7].

This high incidence is probably related to a post-diagnosis experience marked by a series of stressors that can act as a trigger for insomnia and, if they persist, may contribute to a chronic development causing long-lasting sleep disturbance even after the cancer treatment ends.

In a recent systematic review, the prevalence of insomnia in oral cancer patients was 29.0% before, 45% during and 40% after the treatment while hypersomnolence was reported by 16% and 32% of patients before and after the treatment, respectively [8].

The persistence of sleep disorders such as insomnia and hypersomnolence may negatively affect the quality of life of OC survivors and has a powerful influence on the increased risk of infectious disease, and on the occurrence and progression of several major medical illnesses including cardiovascular diseases and mood disorders [9]. Sleep disorders activate biological mechanisms, such as inflammation which are increasingly thought to contribute to depression, and potentially increase the risk of cancer morbidity and related mortality [10]. Indeed, sleep duration has been closely related to a poor overall survival and cancer-specific death over a ten-year follow-up period [11].

In contrast to the substantial literature on depression, research examining sleep disorders in oral cancer is relatively weak, with the majority of studies using a cross-sectional or retrospective analysis. In addition, most of the studies have evaluated the prevalence of sleep disorders before the start or during the treatment while very few studies have included OC survivors in follow-up. Moreover, the role of predictors in sleep disorders remains unclear.

Therefore, we have designed a case-control study to better evaluate the difference in the prevalence of sleep disorders between OC survivors and healthy subjects. The purposes of this study were: (1) to investigate the prevalence of sleep disorders (insomnia and daytime sleepiness), pain, anxiety and depression among OC survivor patients, (2) and to evaluate the potential predictors of sleep disorders such as socio-demographic data, habits, body mass index (BMI), pain, anxiety, depression, medical comorbidities and drug intake and the staging and grading of the oral cancer.

### **2. Material and Methods**

### *2.1. Study Design and Participants*

A case-control study was carried out at the Oral Medicine Department of Federico II University of Naples in accordance with the ethical principles of the World Medical Association Declaration of Helsinki. The study was approved by the Research Ethics Committee (protocol number 188: 2014). The methods adopted conformed with the Strengthening the

Reporting of Observational Studies in Epidemiology (STROBE) guidelines for observational studies (Figure S1) [12].

The recruitment of OC survivors and healthy subjects was conducted between January and September 2018 and was based upon convenience sampling. All potentially eligible individuals were invited to participate in the present study and provided their written informed consent.

The case and the control groups were matched by age and gender. Specifically, first we recruited the patients and then calculated the gender distribution and the average age; secondly, we recruited the controls to obtain a matched sample.

Participants of either gender and aged 18 or older were included. The inclusion criteria for the OC survivors' group were: (i) clinical and histopathological findings of oral squamous cell carcinoma (OSCC) or tobacco-related verrucous cell carcinoma (VCC) (ii) patients with a follow-up of at least five years after the diagnosis of OSCC or VCC and being free from malignancy for at least one year, (iii) all stages based on the American Joint Committee on Cancer Staging Manual 8th edition and (iv) patients managed by surgery, radiotherapy and/or chemotherapy.

On the contrary, the exclusion criteria for the case group were: (i) patients affected by human papillomavirus (HPV)-related OSCC, (ii) patients affected by another type of tumor localized at the head and neck region, (iii) patients who had concomitant tumors in another organ, and (iv) patients who had experienced severe and irreversible side effects from OSCC treatment such as fibrosis, a mouth opening restriction of less than 30 mm, trismus, hyposalivation or osteoradionecrosis of the jaw.

The inclusion criteria for the control group were: (i) patients treated at the University Dental Clinic only for routine dental care during the study period; and (ii) the absence of any oral mucosal lesions or any previous history of OSCC/VCC.

For both groups the exclusion criteria were (i) breastfeeding or pregnant participants, (ii) patients affected by autoimmune disease or another debilitating condition or unstable disease (such as osteonecrosis of the jaw or dementia), (iii) participants with a medical history of a psychiatric disorder as defined by the DSM-5 or regularly treated with a psychotropic drug, (iv) drug-addicted or alcoholic participants and (v) individuals unable or not willing to give their consent or to understand and complete the questionnaires.

### *2.2. Procedure*

A comprehensive intra- and extra-oral examination was carried out by two oral medicine experts (RG and AD). Upon admission, demographic data such as gender, age, educational level (in years), marital status, employment status, risk factors (smoking and alcohol consumption) body mass index (BMI), comorbidities and associated drug use were recorded for both groups.

Details of clinical oral cancer related characteristics were also noted for the case-group, such as the clinical stage and grading at the time of diagnosis, the location of the tumor, any clinical nodal involvement, any metastasis, the type of treatment, and any need for further treatment during the 5-year follow-up. The performance status was assessed using the Eastern Cooperative Oncology Group (ECOG) scale in OC survivors whose scores range from 0 (fully active) to 5 (death), with higher values indicating a poorer performance status [13].

A predefined set of questionnaires was given to the participants of both groups in order to assess their quality of sleep (QoS), their psychological status (level of anxiety and depression) and the intensity and quality of any pain. The questionnaires comprised:


the questionnaires were administered in their Italian version and were reviewed for completeness before collection.

### *2.3. Outcome Measures*

### 2.3.1. Measures of the Quality of Sleep

The Pittsburgh Sleep Quality Index (PSQI) is a standardized questionnaire used for the assessment of the QoS and the incidence of sleep disturbances. This tool consists of 19 items which generate 7 'component' scores: subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleep medication and daytime dysfunction. The scores for each item range from 0 to 3, with higher scores indicating a poorer QoS. The items are combined to yield the seven components, each component having a score ranging from 0 to 3, and the sum of the scores for these seven components yields a global score ranging from 0 to 21. Global scores above five distinguish poor sleepers from good sleepers with a high sensitivity (90–99%) and specificity (84–87%) [14].

The Epworth Sleepiness Scale (ESS) is used to measure an individual's general level of daytime sleepiness. The tool consists of 8 items assessing the propensity for sleep in eight common situations. Subjects rate their likelihood of dozing in each situation on a scale of 0 (would never doze) to 3 (would have a high chance of dozing). The ESS score is the sum of the eight items, ranging from 0 to 24, with a cut-off value of >10 indicating excessive daytime sleepiness [15].

### 2.3.2. Measures of Psychological Factors

The Hamilton Rating Scale for Anxiety (HAM-A) is a measure of symptoms of anxiety and it consists of 14 items. Scores can range from 0 to 56, with scores from 7 to 17 indicating mild symptoms, between 18 and 24 indicating mild-to-moderate severity, and >25 indicating moderate-to-severe anxiety [16].

The Hamilton Rating Scale for Depression (HAM-D) is a measure of symptoms of depression that is comprised of 21 items pertaining to the affective field. Scores can range from 0 to 54. Scores between 7 and 17 indicate mild depression, between 18 and 24 moderate depression, and over 24 severe depression [17].

### 2.3.3. Measures of Pain

The Numeric Rating Scale (NRS-11) is a well-validated instrument for the evaluation of pain intensity. whose scale ranges from 0 to 10 (0 = no oral symptoms and 10 = the worst imaginable discomfort). Respondents are asked to report pain intensity in the last 24 h [18].

The Total Pain Rating Index (T-PRI) from the short form of the McGill Pain Questionnaire (SF-MPQ) is a measure of the quality of pain and it is a multidimensional pain questionnaire which measures the sensory, affective and evaluative aspects of the perceived pain. It comprises 15 items from the original MPQ, each scored from 0 (none) to 3 (severe). The T-PRI score is obtained by summing the item scores (range 0–45). There are no established critical cut-off points for the interpretation of the scores and, as for the MPQ, a higher score indicates worse pain [19].

### *2.4. Statistical Analysis*

Descriptive statistics, including means, standard deviations, medians and the interquartile range (IQR) were used to analyse all the socio-demographic and clinical characteristics of the two groups. For the qualitative variables, the significance was calculated by the Exact Chi Square Test. For the demographic numerical variables the significance difference between means was calculated by the parametric two-samples *t*-test procedure. The significance difference between the recorded medians of the PSQI, ESS, HAM-D, HAM-A, NRS and T-PRI, was measured by the Mann-Whitney Test.

The addition of the clinical characteristics predictors of a poor QoS in OC survivors, hierarchical multiple regression analyses were performed and unadjusted coefficient estimations were obtained for each predictor. A total of six models was computed. The

coefficient estimated for binary variables, such as smoking and alcohol consumption, measures the effect of the Yes response on the outcome estimation. For each model, we reported the adjusted R2 which measures the overall goodness of fit adjusted for the number of variables included into the model. The demographic model (model 1) was performed to test the contribution of the demographic variables to a poor QoS. Next, the clinical model (model 2), the psychological model (model 3), the daytime sleepiness model (model 4) and the pain model (model 5) were each performed after controlling for demographic variables to test the contribution of the clinical variables of the OSCC, anxiety and depression (HAM-A; HAM-D), daytime sleepiness (ESS), intensity and quality of pain (NRS, T-PRI) to a poor QoS. Finally, a standard regression analysis (model 6) was computed by entering all the variables simultaneously into the model in order to determine the relative contributions of all the variables to a poor QoS. In all the steps, standard errors of the model coefficients, which measure the statistical precision of the inference estimation of the model parameters, were provided. The IBM SPSS version 22.0 was used to conduct all the statistical analyses in this study, and *p*-value < 0.05 (two-tails) was considered as statistically significant.

### **3. Results**

The demographic characteristics, BMI and habits of the case and control groups are summarized in Table 1. A total of 100 participants were included in this study, 50 OC survivors and 50 healthy participants and no missing data were recorded.

**Table 1.** Socio-demographic profile, body mass index, disease onset, and risk factors in the 50 OC survivors and 50 controls.


The significance difference between means was measured by the *t*-student test. The significance difference between the percentages was measured by the Pearson Chi Square test. \* Significant 0.01 < *p* ≤ 0.05, \*\* Significant *p* ≤ 0.01. Legend: BMI = body mass index; OSCC = oral squamous cell carcinoma.

Of these participants, 54% (*n* = 26) and 46% (*n* = 24) were male and female for each group, respectively, with a mean age of 59.5 ± 10.1 years for the cases and 65.1 ± 14.4 years for the controls (*p*-value: 0.051). No statistically significant difference was found in terms of marital status, years of education, BMI or alcohol consumption (*p*-values: 0.115, 0.054, 0.068, 0.619, respectively). However, the number of healthy participants in full-time employment

and with a current smoking habit was significantly higher (*p*-value: <0.001 \*\* and 0.005 \*\*\* respectively) in comparison to the case group.

Table 2 shows the prevalence of systemic diseases and drug intake in the study sample. The OC survivors presented with a statistically higher number of systemic comorbidities in comparison to the control group (*p*-value: 0.012 \*), especially with respect to hypertension, hypercholesterolemia, prostatic hypertrophy and gastrointestinal diseases (*p*-values: <0.001 \*\*, 0.001 \*\*, 0.16 \* and <0.001 \*\*\* respectively). Consequently, the number of OC survivors taking medications, such as angiotensin II receptor antagonists, beta blockers, proton pump inhibitors and statin agents was significantly higher compared to the controls (*p*-value: <0.001 \*\*).


**Table 2.** Frequency of systemic diseases and drug consumption in the 50 OSCC patients and 50 controls.

The significance difference between percentages was measured by the Pearson Chi Square test. \* Significant 0.01 < *p* ≤ 0.05, \*\* Significant *p* ≤ 0.01.

Table 3 summarizes the clinical characteristics of the OC survivors. The majority of the patients were diagnosed with stages 0–1 (52%) while 48% were diagnosed with stages 3–4 and with differentiated OSCC (G1-2 88% of the patients). Most of the tumors were localized at the tongue (52%) and alveolar ridges (22%), while 16% and 10% at the buccal mucosa and hard/soft palate, respectively. All the patients with OSCC were managed with surgical treatments ranging from local conservative tumor excision (66.0%) to more invasive surgical treatments. such as hemiglossectomy (20%), maxillary osteotomy (8.0%), hemimandibulectomy (6%) and cervical neck dissection (42%). Only a few patients received, in addition, radiotherapy (16%) or chemotherapy (2%). Tracheostomy was not performed in respect of any OC survivors. Overall, the OSCC patients were further treated with

incisional or excisional biopsies over the five-year follow-up period (a mean of 4.8 +/− 2.9) due to local relapses, especially in respect of the 29 (58%) OC survivors with associated potentially malignant disorders such as lichenoid lesions 8 (16%), leukoplakia 7 (14%) erythroleukoplakia 14 (28%).


**Table 3.** Medical characteristics of the OC survivors.

At the time of the assessment, 66% of the OSCC patients presented with an ECOG performance status of 0 ("fully active") and 34% with an ECOG performance status of 1 ("restricted in physically strenuous activity").

Among the OC survivors, 52% were poor sleepers (PSQI > 5), whereas only 12% of the controls reported a poor QoS. Moreover, mild to severe anxiety was reported in 84% of the OC survivors (48% mild, 12% moderate and 24% severe anxiety) along with mild to severe depression in 74% of cases (40% mild, 16% moderate and 18% severe depression). On the contrary, only 20% and 18% of the healthy participants showed mild anxiety and depression symptoms, respectively, and no cases of moderate to severe anxiety or depression were recorded in the control group.

Table 4 shows the differences in all the psychological factors between the case and control group. A Cronbach alpha value of 0.76 and 0.91 was indicative of a good reliability of the PSQI scale in both groups. The OC survivors presented a mean of hours of sleep of 6.94 ± 1.024, while the controls slept a mean of 7.16 ± 0.681 h. A statistically significant difference was found between the medians of all the psychological variables assessed in terms of QoS, anxiety and depression and intensity and quality of pain. The OC survivors showed statistically significant higher scores in the global PSQI (*p*-value: 0.017 \*), especially for the items "subjective sleep quality", "sleep latency" and daytime dysfunction" (*p*-values: <0.001 \*\*, 0.029 \* and 0.004 \*\* respectively), and in the total ESS score (*p*-value: 0.001 \*\*) in comparison with the controls. Furthermore, statistically significant higher levels of anxiety and depression, as reflected by the total scores of the HAM-A and HAM-D, were also recorded among the OC survivors (*p*-value: <0.001 \*\*), together with higher levels of oral discomfort and pain according to the NRS and T-PRI total scores (*p*-value: 0.001). Taken together, these findings suggest that QoS and psychological status may be severely impaired in OC survivors.

**Table 4.** Differences in sleep quality, anxiety, depression and pain in 50 OSCC patients and 50 controls.


Legend: ESS = Epworth Sleepiness Scale; HAM-A = Hamilton Anxiety Scale; HAM-D = Hamilton Depression Scale; IQR = interquartile range. NRS = Numeric Rating Scale; McGill: PSQI = Pittsburgh Sleep Quality Index; T-PRI: Total Pain Rating Index. The significance difference between medians was measured by the Mann–Whitney test. \* Significant 0.01 ≤ *p* ≤ 0.05 \*\* Significant *p* ≤ 0.01.

Furthermore, in the case group, a statistically significant positive correlation was found between the global PSQI score and the HAM-A, HAM-D and T-PRI scores (*p*-values: <0.001 \*\*, <0.001 \*\* and 0.019 \* respectively) but not with the ESS and the NRS. Specifically, the majority of the PSQI sub-items (except for "use of sleep medication" and "sleep latency") were positively correlated with the HAM-A and HAM-D (except for "use of sleep medication"), whereas the T-PRI was correlated only with "sleep disturbances and daytime dysfunction" which also correlated, as expected, with the ESS. Overall, patients with a poorer QoS presented with higher levels of anxiety and depression and a worse quality of pain but not with increasing daytime sleepiness or pain intensity (Table 5).


**Table 5.** Correlation analysis between the PSQI items and anxiety, depression and pain in 50 OSCC patients and 50 controls.

Legend: ESS = Epworth Sleepiness Scale; HAM-A = Hamilton Anxiety Scale; HAM-D = Hamilton Depression Scale; NRS = Numeric Rating Scale; McGill: PSQI = Pittsburgh Sleep Quality Index; T-PRI: Total Pain Rating Index. Correlation between PSQI items and other variables was measured with the Spearman correlation analysis. \* Moderately significant 0.01 < *p* ≤ 0.05; \*\* strongly significant *p* ≤ 0.01.

> The hierarchical multiple regression analyses predicting QoS are shown in Table 6. The first model (the demographic model), testing the contribution of demographic variables and risk factors (alcohol and smoking) to QoS, showed that the PSQI was negatively correlated with years of education (*p*-value: 0.042 \*) and resulted in a strongly significant increase in the coefficient of determination (R2) (ΔR2 = 31.7%, *p*-value: 0.009). The addition of the clinical characteristics showed that the PSQI was positively correlated with alcohol consumption (*p*-value: 0.018 \*) and with the use of systemic medications (*p*-value: 0.045 \*). When entering all the variables simultaneously in the second model, we found an increase in the R2 value with a ΔR2 of 6.2%, possibly due to both the parameters, namely alcohol consumption and medications, although it was not statistically significant (*p*-value: 0.222). The third model (the psychological model), testing the contribution of anxiety and depression to QoS, showed that the PSQI was positively correlated with the HAM-A and HAM-D (*p*-value: 0.001 \*\*) and resulted in a strongly significant increase in the R2 (ΔR2 = 20.4%, *p*-value: <0.001 \*\*). The daytime sleepiness and pain models (models 4 and 5) did not result in a significant increase in the R2 value (ΔR2 = −2.1%, 0.0%; *p*-value: 0.749 and 0.377 respectively). The final full model (model 6, the standard multiple regression analysis) in which all of the variables were entered simultaneously (including demographic variable, risk factors, clinical characteristics, medications, anxiety, depression, daytime sleepiness, pain) resulted in a moderate increase in the R2 value (ΔR2 = 12.6 %; *p*-value: 0.043 \*) and could explain the 44.3% of variance of poor QoS. In this last model, depression has shown a strong correlation to sleep disorders (*p*-value: 0.001 \*\*) contributing significantly to a poor QoS.


**Table 6.** Multiple linear regression analysis predicting poor QoS (PSQI > 5) in 50 OC survivors.


McGill: PSQI = Pittsburgh Sleep Quality Index; T-PRI: Total Pain Rating Index.

### **4. Discussion**

The aim of this study has been to investigate the prevalence of sleep disorders (insomnia and hypersomnolence), anxiety and depression in OC survivors with a 5-year follow-up and to analyze potential predictors in the development of sleep disorders. The detection and treatment of factors which could influence the well-being of OC survivors are becoming increasingly important for healthcare systems in order to improve the follow-up care of these patients.

Among this population, insomnia, poor QoS, short sleep duration, excessive daytime sleepiness and sleep-related breathing are commonly reported and tend to become often chronic and pervasive in patients during and after treatment for OSCC [3].

In a recent systematic review, the prevalence of self-reported insomnia (defined with a PSQI cut-off of 5) in patients with head and neck cancer was 29% before treatment, 45% during treatment and 40% after treatment, while the prevalence rate of hypersomnolence (ESS cut-off > 10) was 16% before and 32% after treatment [8].

In this study, a higher prevalence of insomnia among the OC survivors within the 5-year follow-up was found, in comparison with the study of Santoso et al. [8] as 52% of the patients were poor sleepers (median PSQI score 6), while hypersomnolence was found in 24 % of OC survivors, in line with previous research [20,21].

With regard to the PSQI components a higher percentage of OC survivors reported an impaired subjective sleep quality, sleep latency, and daytime dysfunction.

Pain, fatigue, medical treatment, psychological profile (anxiety and depression) and comorbidities [22] may cause poor sleep in cancer patients. In this study, the full model of the multiple regression analysis, where all the variables were entered simultaneously, could explain only 44.3% of the variance of the PSQI in OC survivors, suggesting that the occurrence of insomnia could be independent of the cancer characteristics, staging of the malignancy, type of treatment (surgery, or radiotherapy), pain and presence of potentially malignant disorders. Instead, poor sleep was negatively correlated with years of education and positively correlated with mood disorders (anxiety and depression), the use of systemic medications and the consumption of alcohol. Therefore, a lower education level, the use of systemic drugs, the consumption of alcohol and the presence of anxiety and depression were predictors for poor sleep in OC survivors.

In a previous study, a lower education level, the presence of systemic comorbidities and the use of systemic drugs, adversely affected quality of life outcomes in survivors of cancer [23]. Moreover, there is evidence that sleep disorders may be associated with cardiovascular diseases and cardiovascular risk factors, such as hypertension and elevated resting heart rate in the general population [24], and that cardiovascular medications such as beta adrenergic blocking agents, ACE inhibitors, calcium channel antagonists may negatively affect sleep quality in individuals with other comorbidities, especially those with sleep disorders breathing [25].

Our results are in line with these studies, suggesting that the use of medications for systemic comorbidities could have a detrimental effect on the life of patients that over time could also influence QoS. However, medications with alcohol consumption contributed to sleep disorders on the account of 6.2% of the variance of poor QoS based on the second model of the regression analysis which suggests that medications may not have a pivotal role in explaining the higher prevalence of sleep disorders in this group of OC survivors, possibly for the absence of sleep disorder breathing and obstructive sleep apnea in our sample.

In addition, the low intensity of pain (NRS: 2) reported by OC survivors is considered as a predictor of poor sleep, as suggested by the regression analysis. Although xerostomia was not detected in our sample of patients probably because radiotherapy was prescribed in only 16% (8) of patients, Shuman et al [26] similarly reported that pain in the mouth and xerostomia (dry mouth) were strong predictors of poor sleep.

Regarding habits, alcohol abuse and tobacco smoking might play a role in the development of sleep disorders. Indeed, heavy alcohol users often experience insomnia even

after they stop their alcohol consumption, while smokers suffer more frequently from poor sleep, compared with non-smokers [27,28]. In this study, at the time of evaluation, only 16% (8) were current smokers, as the majority had stopped their smoking habit after their OSCC diagnosis. Conversely, 42% (21) continued to consume alcohol (<14 units per week), although no one was a heavy drinker. Therefore, the positive correlation between poor sleep and alcohol consumption could be related to a previous higher alcohol consumption.

While in a recent study insomnia and hypersomnolence were found to be associated with chemotherapy and radiotherapy, [23] in the present study we could not find this correlation, presumably because the majority of the patients were in stage 0/1 (52%, 26 individuals) and only 2% (1) and 16% (8) of patients, respectively, had received these protocols. A recent review article suggested that surgery may have a positive effect on sleep quality; indeed, patients with oral cancer treated with surgery were less prone to develop insomnia, probably because they considered the operation as a resolution of the disease. The authors found a prevalence of insomnia of 31.9% in oral cancer patients who had undergone surgery and of 44.9% in those who were not receiving surgery, especially females. An explanation of these results could be that women are more vulnerable to the stress related to a cancer diagnosis and subsequently to mood disorders on account of their hormonal status [29]. In the current study we did not find any differences between male and female OC survivors, all the patients having been treated with surgical procedures.

Previous studies have suggested that obesity (BMI > 30) is considered a significant predictor of sleep disorders [30]. In our study, only 16% (8) of OC survivors were overweight, however, based on the result of the regression analyses, BMI may not have contributed to sleep disorders, similarly to the findings from the study of Bardewell et al [31].

Regarding the psychological profile, the current literature has reported a prevalence of anxiety and depression, ranging from 19 to 50%, in cancer survivors, suggesting that the burden of cancer diagnosis and its treatment could have a strong impact on the psychological profile, persisting over time despite a successful operation and subsequently decreasing the quality of life of the affected patients. Moreover, Espie et al. reported that from 22% to 32% of OC survivors were anxious or depressed even ten years after the diagnosis and treatment [32]. Factors identified as contributing to an increased risk of psychological distress among oral cancer patients include persistent pain, age (generally, younger patients more seriously affected than older patients), gender (females more seriously affected than males), stage of cancer, type of treatment, and fear of cancer recurrence. Moreover, anxiety and oral dysfunction, including trismus, xerostomia, sticky saliva and problems with eating and social contacts, are also considered a barrier to any return to work after treatment among head and neck cancer survivors [33]. As a consequence, a lack of full-time employment can exacerbate the depressive symptoms.

In this study, a higher prevalence of mood disorders has been found in comparison with the current literature; indeed, anxiety and depression were identified in 84% (42%) and 74% (37) of OC survivors, respectively. In addition, in the final full model, depression was found to be the most contributive factor to poor QoS. The higher level of depression may be related to the stress associated with a fear of cancer recurrence, since almost 40% [3] of patients presented a local cancer recurrence and, therefore, underwent a subsequent operation during the five years of follow-up.

Mood disorders and poor sleep were closely interconnected, as shown by the correlation analysis. In addition, anxiety and depression were predictors of poor sleep, as confirmed by the regression analysis. No differences between male and female patients were detected, and neither the stage and treatment nor the number of operations for cancer recurrence affected the incidence of sleep disorders. In line with previous studies, an impaired mood and sleep affected the functional recovery of patients and their return to work because, despite their age, the majority of OC survivors (48%) had retired.

The results of this study suggest that the high prevalence of insomnia may be related not only to psychiatric symptoms or to a fear of cancer recurrence but could also be considered in some cases an independent variable (as shown by the regression analysis) which

needs to be addressed regardless of all the other factors. It is possible to consider that cancer itself can lead to the development of sleep disorders through inflammation. Inflammation has emerged as a crucial pathway which may be especially relevant with respect to cancer survivors. The sleep-wake cycle has emerged as a homeostatic regulator of inflammatory biology in which sleep loss induces an activation of nuclear factor KB (NF-kb) [34] and circulating levels of IL-6 [35], which coordinate the production of inflammatory mediators and systemic inflammation. In turn, pro-inflammatory cytokines are thought to contribute in part to the onset of depressive symptoms, which can amplify sleep disorders [36,37]. Moreover, chronic inflammation may predispose to a second primary recurrence [38].

Adequate sleep is a biological requirement for healthy physical, cognitive and psychological functioning so the management of sleep disturbance should be targeted by clinicians with appropriate interventions. In particular, the prominent role of cognitive behavior therapy has been studied [39]. \* Additionally, the administration of melatonin in relation to the management of the sleep-wake cycle and mood disturbance as well as with respect to the quality of life of cancer patients has been proposed [40].

The findings of the current study should be understood in the light of some limitations. First, the sample is small and all the patients were recruited at a single hospital, thus preventing the possibility of any geographical generalizability and slightly affecting the power of the regression analyses. Secondly, the exclusion of patients who had developed severe and permanent side effects due to the radiotherapy, may have produced a potential underestimation of the prevalence of sleep disorders in OC survivors. Moreover, the study design does not allow the drawing of any conclusive inferences about the temporality and causality of the relationships between the variables explored. Finally, only subjective sleep quality was investigated in this study, with objective sleep quality not being considered, and therefore additional measurement systems should be incorporated to verify our findings.

### **5. Conclusions**

Sleep disorders (including insomnia and hypersomnolence) continue to be prevalent both during and after treatment for OSCC. A lower level of education, the use of systemic drugs, the consumption of alcohol and the presence of anxiety and especially depression are predictors of poor sleep in OC survivors.

The treatment of oral cancer must clearly remain the major goal, but the treatment of any psychological comorbidities is also important in order to improve the quality of life in these patients. Therefore, healthcare professionals should be encouraged to include sleep disorders assessment at the time of diagnosis, during treatment and in follow-up consultations. Further clinical and prospective studies should be conducted not only to evaluate the real prevalence of sleep disorders but also to plan an adequate treatment over time with respect to all OC survivors.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/cancers13081855/s1, Figure S1: Flow chart.

**Author Contributions:** D.A. and M.D.M. equally contributed for the conceptualization of the study, the methodology, the data collection and curation and drafted the paper; R.G., E.C. and N.C. contributed to the data curation and collection and drafted the paper; G.M. and G.S. drafted and reviewed the paper; M.A. analyzed the data and contributed to writing the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** The authors have not been supported by any private or corporate financial institutions, nor have they received any grant for this study.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board (or Ethics Committee) of University of Naples Federico II (protocol number 188: 2014).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author. The data are not publicly available due to patient sensitive data.

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

### **References**


### *Systematic Review* **Is Systemic Immunosuppression a Risk Factor for Oral Cancer? A Systematic Review and Meta-Analysis**

**Romeo Patini 1, Massimo Cordaro 1, Denise Marchesini 1, Francesco Scilla 1, Gioele Gioco 1,\*, Cosimo Rupe 1, Maria Antonietta D'Agostino <sup>2</sup> and Carlo Lajolo <sup>1</sup>**


**Simple Summary:** Immunosuppression is a medical condition in which a person's immune system is unable to function properly, or it does not function at all. It is a well-known fact that an illfunctioning immune system can favor the generation and development of potentially malignant lesions, autoimmune and allergic diseases, and even neoplasms. At present, the amount of risk for the development of oral cancer in immunosuppressed patients has not been quantitatively reported. Such a topic has been investigated, revealing that immunosuppression increases the risk of developing cancer from 0.2% to 1% (95% CI: 0.2% to 1.4%), giving further importance to the accurate follow-up of this category of patients.

**Abstract:** Even if the relationship between immunosuppression and increased incidence of systemic cancers is well known, there is less awareness about the risk of developing oral cancer in immunosuppressed patients. The aim of this review was to evaluate the association between immunosuppression and the development of oral cancer. Two authors independently and, in duplicate, conducted a systematic literature review of international journals and electronic databases (MEDLINE via OVID, Scopus, and Web of Science) from their inception to 28 April 2023. The assessment of risk of bias and overall quality of evidence was performed using the Newcastle–Ottawa Scale and GRADE system. A total of 2843 articles was identified, of which 44 met the inclusion criteria and were included in either the qualitative or quantitative analysis. The methodological quality of the included studies was generally high or moderate. The quantitative analysis of the studies revealed that immunosuppression should be considered a risk factor for the development of oral cancer, with a percentage of increased risk ranging from 0.2% to 1% (95% CI: 0.2% to 1.4%). In conclusion, the results suggest that a constant and accurate follow-up should be reserved for all immunosuppressed patients as a crucial strategy to intercept lesions that have an increased potential to evolve into oral cancer.

**Keywords:** immunosuppression; oral cancer; systematic review; meta-analysis

### **1. Introduction**

According to official data from the World Health Organization (WHO, Geneva, Switzerland), 377,713 new cases of oral and lip cancer were diagnosed in 2020, making it the 16th most common cancer in the world. It still has a severe prognosis today, as approximately 50% of oral and lip cancer patients will die in the 5 years following diagnosis, while the remaining 50% have aesthetic and functional relics that make their quality of life rather low. Historically, the main risk factors for this neoplasm are being male, having a diet low in vitamins, having MPDs, past/present viral infections, radiation exposure, having genetic

**Citation:** Patini, R.; Cordaro, M.; Marchesini, D.; Scilla, F.; Gioco, G.; Rupe, C.; D'Agostino, M.A.; Lajolo, C. Is Systemic Immunosuppression a Risk Factor for Oral Cancer? A Systematic Review and Meta-Analysis. *Cancers* **2023**, *15*, 3077. https://doi.org/10.3390/ cancers15123077

Academic Editor: Anita Kloss-Brandstätter

Received: 30 April 2023 Revised: 30 May 2023 Accepted: 1 June 2023 Published: 6 June 2023

**Copyright:** © 2023 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 (https:// creativecommons.org/licenses/by/ 4.0/).

predispositions and immunodeficiencies, and engaging in luxuriant habits such as smoking and alcohol and betel consumption [1].

Oral cancer treatment is challenging and requires a multidisciplinary approach with a team of specialists, which includes head and neck surgeons, radiation oncologists, medical oncologists, and oral oncologists [2].

Although surgery is the most common initial definitive treatment for the majority of oral cancers, adjunctive radiotherapy (RT), with or without chemotherapy (CT) may be performed [3].

The immune system performs numerous functions, among which its primary functions are defense against infections, self-control and immunosurveillance at the onset and during the proliferation of solid and liquid cancers, identifying and suppressing genetically modified cells that have already passed the normal checkpoints, and the intracellular control of proliferation. The possible role of the immune system in the development of cancers has been defined in the theory of "immune surveillance", which configures the active role of the immune system in preventing the onset of cancers [4].

Immune surveillance against cancer is the process in which the immune system identifies cancerous and/or precancerous cells and eliminates them.

According to the most recent findings, the immune system can play a role in preventing tumors, throughout different mechanisms. First, the virus-induced tumors can be prevented when a functioning immune system can eliminate or suppress viral infections. Second, this action against pathogens may cause a prompt resolution of inflammation, preventing the establishment of an inflammatory environment, which is a risk factor for carcinogenesis [5]. Third, the immune system can identify and eliminate tumor cells on the basis of their expression of tumor-specific antigens. Therefore, the theory of immunosurveillance is essentially based on two generally accepted claims: (I) most cancers are antigenic (an obvious requirement for immunological recognition) and (II) such antigenic differences can, "under appropriate conditions", elicit an immune response [4].

Despite immune surveillance, cancers develop even in the presence of a functioning immune system, and therefore, currently, we speak of "cancer immunoediting", a term which is used to describe the evolution of tumors, wherein tumor cells become less effectively recognized and killed by the immune system [6,7].

A first consideration concerns the definition that is used for patients with disorders of the immune system. The terms immunosuppression and immunodeficiency are often used interchangeably. This confusion is related to the subtle nuance that separates them. It could be specified that immunosuppression identifies a medical condition of a general malfunction of the immune system. Immunodeficiency, on the other hand, classifies the severity of this physical deficit according to two categories: primary and secondary.

Immunosuppression is a pathological condition characterized by the inhibition of one or more components of the immune system, whether natural or acquired, resulting in the impossibility of a person's immune system to function properly. However, currently, there is no description illustrating the relationship between immunoediting and immunosuppression. The incorrect functioning of the immune system can favor the development of autoimmune and allergic diseases or neoplasms. Immunodeficiencies are divided into primary (if they are derived from congenital defects) and secondary (if they are derived from infections or pharmacological treatments) classifications. This condition involves the onset of infections that develop and recur very often, manifesting themselves in a more serious and longer-lasting form.

Among the many alterations of the immune system, immunodeficiency can be caused by numerous and different causes, and it can involve acquired or innate immunity, both in the humoral and cellular components, as follows: innate pathologies (e.g., agammaglobulinemia linked to the X sex chromosome, one common variable immunodeficiency, severe combined immunodeficiency, DiGeorge syndrome, and congenital hypogammaglobulinemia), systemic diseases (e.g., autoimmune diseases, diabetes, chronic infections, and solid and liquid malignancies, such as leukemia, lymphoma, and multiple myeloma), and pharmacological therapies (e.g., chemotherapy, antirheumatics, immunosuppressants, and glucocorticoids), which are the main causes of immunodeficiency [8,9].

By definition, immunodeficiency is characterized by a functional deficit of the immune system (either congenital or acquired). Immunosuppression is a pathological condition characterized by the inhibition of one or more components of the immune system (natural or acquired), and it occurs following an intercurrent disease or autoimmune pathologies [10]. Immunosuppression also refers to pharmacological treatment with immunosuppressive drugs capable of inhibiting an immune system response [11]. Therefore, immunocompromised patients have a reduced ability to fight infections and other diseases.

Numerous studies have shown that in immunosuppressed subjects, there is a higher incidence of cancers than in a population with normal immunity [12]. The increased susceptibility to infections (i.e., HPV, candida, Helicobacter pylori, etc.) and the reduced immune response to infections in immunosuppressed subjects could represent a further mechanism that favors the onset of neoplasms. Furthermore, immunosuppression is, at the same time, one of the risk factors for the onset of oncological pathologies, but it is also a condition that could favor the loco-regional and distant growth and spread of cancers. In fact, the literature shows that immunosuppression is not only a risk factor for the genesis of a cancer but also a factor for the prognosis of its course [13].

Although the relationship between immunosuppression and the increased incidence of systemic cancers is now well documented, currently, it is not clear how much the risk of developing oral cancer increases in immunosuppressed subjects and what effect immunosuppression has on prognosis in terms of survival. The purpose of this systematic review was, therefore, to evaluate the association and the possible correlation between the state of depression of the immune system and the development of oral cancer through the evaluation of the incidence of oral cancer in patients with systemic immunosuppression and to compare that to data from official databases (Globocan, WHO), which lacked precise data on non-immunosuppressed subjects.

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

In the present systematic review, the adopted protocol followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. The review protocol was registered in PROSPERO database (CRD42021243898).

### *2.1. PICOS Question*

The following question was developed according to the population, intervention, comparison, outcome, and study design (PICOS).

Population: immunosuppressed patients who later developed oral cancer were included in this systematic review.

Intervention: patients with systemic immunodepression due to various factors (immunodepression, malnutrition, infections, autoimmune diseases, genetic immunosuppression, immunosuppression as a consequence of immunosuppressive therapy or radiotherapy, and oncologic immunosuppression) who subsequently developed oral cancer were considered.

Comparison: the rates of development of oral cancer in non-immunosuppressed patients and the rates of development of oral cancer in immunosuppressed patients were compared.

Outcome: the primary outcome was to evaluate the incidence of oral carcinoma in immunosuppressed patients.

Study design: cohorts, case controls, cross-sectional studies, and randomized clinical trials (RCTs) with no fewer than 10 patients were included. All case reports, case series with less than 10 patients, in vitro or in vivo studies based on animals, systematic reviews, letters to the editor, cases of oral cancer related to human papilloma virus (HPV), and articles published in languages other than Italian, English, and Spanish were excluded.

### *2.2. Focused Question*

The question on which attention was focused was formulated on the basis of the PICOS criteria: "Do immunosuppressed patients have a higher rate of development of oral cancer than healthy patients?".

### *2.3. Research*

The research was conducted on three databases (MEDLINE via OVID, Scopus, and Web of Science) from the start of their activity in May 2022, using a combination of key words and MeSH terms as follows: ((immunosuppression OR malnutrition OR infections OR autoimmune disease OR X-linked agammaglobulinemia OR common variable immunodeficiency OR selective immunoglobulin A deficiency OR hyper IgM syndrome OR DiGeorge syndrome OR severe combined immunodeficiency OR Wiskott–Aldrich syndrome OR acquired immunodeficiency syndrome OR AIDS OR immunosuppressive therapy OR radiotherapy OR "other systemic cancers" OR leukaemia OR lymphoma) AND "Oral Cancer"), ("Oral Carcinoma" AND (immunosuppression OR malnutrition OR infections OR autoimmune disease OR X-linked agammaglobulinemia OR common variable immunodeficiency OR selective immunoglobulin A deficiency OR hyper IgM syndrome OR DiGeorge syndrome OR severe combined immunodeficiency OR Wiskott–Aldrich syndrome OR acquired immunodeficiency syndrome OR AIDS OR immunosuppressive therapy OR radiotherapy OR "other systemic cancers" OR leukaemia OR lymphoma)), and ("Oral Neoplasms" AND (immunosuppression OR malnutrition OR infections OR autoimmune disease OR X-linked agammaglobulinemia OR common variable immunodeficiency OR selective immunoglobulin A deficiency OR hyper IgM syndrome OR DiGeorge syndrome OR severe combined immunodeficiency OR Wiskott–Aldrich syndrome OR acquired immunodeficiency syndrome OR AIDS OR immunosuppressive therapy OR radiotherapy OR "other systemic cancers" OR leukaemia OR lymphoma)). The date of the last search was 28 April 2023.

### *2.4. Manual Search*

A manual search of articles published between 2002 and 2022 in the following peerreviewed journals was performed: *Oral Oncology*, *Oral Diseases*, *Lancet Oncology,* and *Journal of Hematology and Oncology*.

### *2.5. Search of Unpublished Articles*

Unpublished literature was searched in the U.S. National Institutes of Health clinical trials registry and the European Multidisciplinary Database to identify incumbent studies and grey literature. In addition, bibliographic references of all included articles and reviews were similarly checked to identify additional potentially relevant studies and increase the sensitivity of the search.

### *2.6. Study Selection*

Based on the inclusion criteria, two authors independently and in duplicate (D.M. and F.S.) analyzed the titles and abstracts of the articles found. The authors retrieved the full versions of articles whose titles and abstracts appeared to meet the inclusion criteria or those, which reported insufficient data to make a clear decision. Next, the two authors independently read the full texts to determine whether the articles met these criteria. In cases where the two authors disagreed, agreement was sought through a comparison between the two, and when a solution could not be reached, a third senior author (R.P.) stepped in.

To calculate the agreement between the reviewers, Cohen's kappa coefficient was used. The level of agreement was considered excellent when k was greater than 0.75, fair to good when it was between 0.40 and 0.74, and poor when it was less than 0.4 [14].

All articles that met the inclusion criteria were subjected to data extraction and quality assessments. All irrelevant articles were excluded, and the reasons for exclusion were as described.

### *2.7. Extraction Data*

The data were collected using a purpose-built data extraction form. In cases where the publication did not provide all the necessary data, the corresponding author was contacted by e-mail to obtain the missing data. In the event that the two authors disagreed about one of the publications, a discussion was opened, which, in cases of disagreement, required the intervention of the third author.

In cases of redundant publications, the most recent article and the one with the largest follow-up were included.

### *2.8. Quality Assessment*

The risk of bias in the included studies was independently assessed in duplicate by two authors as part of the data extraction process.

An assessment of risk of bias was undertaken using the Newcastle–Ottawa Scale (NOS) [15]. The presence of each parameter was recorded with a green mark, while absence was recorded with a red mark (0). Papers with 1–3 green marks were classified as high risk of bias, those with 4–6 green marks were classified as medium risk, and those with 7–9 green marks were classified as low risk. A supplemental analysis was performed independently by the two examiners regarding the overall quality of the evidence for any performed meta-analysis using the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) system [16]. Any disagreement between the two reviewers (D.M. and F.S.) was solved by discussion with the author supervisor (R.P.).

Publication bias was assessed through a funnel plot, which was made using Excel software (Microsoft Excel®).

### *2.9. Heterogeneity Assessment*

The OpenMeta software was used for assessing the heterogeneity of the studies included in any conducted meta-analysis (OpenMeta, Inc.©, Zug, Zug, Switzerland). The authors calculated the comparability of the observed proportions across the results with chance alone using the I2 test. In cases where the *p*-value was <0.1, the heterogeneity was considered significant. Moreover, the same test was considered as a measure of heterogeneity across studies, following the subsequent scheme [17]: 0–40%, negligible; 30–60%, moderate; 50–90%, substantial; and 75–100%, considerable.

### *2.10. Data Analysis*

Descriptive characteristics of the studies are expressed as means/medians and/or frequencies, as appropriate, depending on the variables.

Meta-analyses were performed only when there were studies comparing similar groups and reporting the same outcomes. In such cases, the meta-analyses were performed with a fixed-effect model. A random-effect model was used only in the case of not-negligible heterogeneity across the included studies (>50%).

A forest plot was created to illustrate the effects on the meta-analysis of individual studies and the overall estimate. OpenMeta-analyst [18] was used to perform all analyses. The cut-off value of significance was set at *p* < 0.05.

### **3. Results**

### *3.1. Study Selection*

A flowchart of the search strategy and study selection is shown in Figure 1.

**Figure 1.** Flowchart of the selection of the studies for the review.

A total of 2843 articles was identified, with 2796 found through electronic searches and 47 found through other sources. Out of the 2709 studies that resulted after removal of the duplicates, 2470 were excluded as a result of title and abstract reading (inter-reader agreement, k = 0.78). Eventually, out of the 239 articles that remained to be evaluated in the full-text, 44 met the inclusion criteria and were included in either the qualitative or quantitative analyses (meta-analysis); in contrast, 195 were excluded. All information about full-text articles excluded, with reasons are included in the Supplementary Materials File (Table S1).

### *3.2. Study Characteristics*

The characteristics of the included studies are summarized in Table 1.

Both prospective (five studies) and retrospective (nine studies) cohort studies were included in the review. Twenty-four studies presented data from national registries, and therefore, they were analyzed separately. In addition, six studies presented results related to a single immunosuppression condition, namely, graft-versus-host disease, and for this reason, they were analyzed separately, as this condition is, itself, a potentially malignant disorder of the oral cavity. All studies were conducted in an institutional environment.


*Cancers* **2023**, *15*, 3077

**Table 1.**

Characteristics

 of the included studies.






NR: Not reported, P: Prospective, R:

Retrospective.

### *3.3. Assessment of the Risk of Bias*

The risk of bias is summarized in Figures 2 and 3. The methodological quality of the included studies was high for 12 studies [19,22,24,25,28,31–33,36,51,53,58], moderate for 26 studies [20,21,23,25,26,29,30,32,34,37–50,52,54–57,60–62], and low for six studies [23,26,27,35,48,59].

**Figure 2.** Risk of bias graph.

The results regarding publication bias are presented in Figures 4–6. Significant publication bias was found in the studies that presented results related to Graft Versus Host Disease (GVHD) and the national registries. The Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) system provided information on the certainty of the conclusions and the strength of the evidence (Table 2). Although the meta-analyses drew conclusions from cohort studies, which are considered to be among the best-available evidence, they were considered to have only moderate strength of evidence because of the presence of at least one study with a high risk of bias and very wide confidence intervals.

**Table 2.** GRADE summary of findings for meta-analysis on immunosuppression and oral cancer incidence.


a. Due to high heterogeneity across studies. b. Due to wide confidence intervals.


**Figure 3.** Risk of bias summary [19–62].

**Figure 4.** Funnel plot of studies with data from national registries [20,23,26,31–42,44–50,57,59].

**Figure 5.** Funnel plot of studies with data not from national registries [19,21,22,24,25,27–30,43,54,58,60].

**Figure 6.** Funnel plot of studies with data about GVHD [51–53,55,56,61].

**Figure 7.** Meta-analysis related to data coming from national registries [20,23,26,31–35,37–42,44–50,57,59].

**Figure 8.** Meta-analysis related to data not coming from national registries [19,21,22,24,25,27–30,43,54,58,60,62].

**Figure 9.** Meta-analysis regarding data about GVHD [51–53,56,61].

### *3.4. Results of the Meta-Analyses*

As reported earlier, three separate meta-analyses were conducted. The meta-analysis related to the national registries (Figure 7) was conducted on 23 studies with a total of 5,227,567 patients and found an "untransformed proportion" (PR) of 0.2% (95% CI: 0.002–0.003) (*p*-value of <0.001).

The meta-analysis concerning data not derived from the national registries (Figure 8) was conducted on 15 studies with a total of 6997 patients and found an "untransformed proportion" (PR) of 1% (95% CI: 0.006–0.014) (*p*-value of <0.001).

The meta-analysis regarding data about GVHD (Figure 9) was conducted on six studies with a total of 49,285 patients and found an "untransformed proportion" (PR) of 0.3% (95% CI: 0.001–0.005) (*p*-value of < 0.001).

The meta-analyses conducted on the three groups of patients revealed a general increased risk of developing an oral cancer in immunosuppressed populations. Such risk ranges from 0.2% to 1% depending on whether data from national registries are considered. In immunosuppressed patients, this evidence emphasizes the need to provide for a careful follow-up of suspicious lesions and potentially malignant disorders of the oral cavity.

### **4. Discussion**

### *4.1. Summary of the Main Findings*

The close relationship between the immune system and cancer immunoediting has been documented for many years for numerous cancers, including oral carcinoma, and this systematic review highlighted an incidence of oral carcinoma in immunosuppressed patients of 200 new cases per 100,000. If this is compared to data from registries on the incidence of oral cancer in the general population, which is approximately 4.1 per 100,000 subjects (ASR incidence = 4.1 per 100,000), immunosuppressed subjects have a risk of developing oral cancer that is 50 times higher than the general population. These raw data emphasize the need to establish clinical protocols for primary prevention and screening in all immunosuppressed subjects, likely with tailor-made protocols that depend on the cause of immunosuppression and the severity of the immunosuppression.

Some considerations of a methodological nature that emerged from this systematic review should be made in light of the literature. A first consideration concerns the definition that is used for patients with disorders of the immune system. The terms immunosuppression and immunodeficiency are often used interchangeably. This confusion is related to the subtle nuance that separates them. It could be specified that immunosuppression identifies a medical condition involving a general malfunction of the immune system, whereas immunodeficiency classifies the severity of this deficit into primary and secondary in relation to the cause. Furthermore, an aspect still unresolved concerns the identification of clinical and/or instrumental parameters that can identify the state of immunosuppression (considering both innate and acquired immunity, both cellular and humoral) and classify it in relation to the severity of the immunosuppression.

The present systematic review demonstrated that immunosuppression should be considered a risk factor for the development of oral cancer, with a percentage of increased risk ranging from 0.2% to 1% (95% CI: 0.2% to 1.4%). Considering the main causes of immunosuppression reported in the selected articles, there are some interesting considerations. In fact, in this systematic review, the authors decided to divide the results from the included papers into three main groups: the results derived from the literature analysis of the main reasons for immunosuppression (not from national registries), those from articles related to GVHD, and those from the registry analysis, which depict an increased risk of 1% (95% CI: 0.6% to 1.4%), of 0.3% (95% CI: 0.1% to 0.5%), and of 0.2% (95% CI: 0.2% to 0.3%), respectively. Articles referring to states of malnutrition were not included in this review, as they did not report adequate information regarding immune status.

### *4.2. Organ Transplantation*

Organ transplantation, in particular, kidney transplantation, represents one of the main causes of immunosuppression most frequently associated with the onset of oral cavity neoplasms. The increased life expectancy of transplant recipients exposes them to prolonged immunosuppressive therapy (mainly cyclosporine), which is necessary to avoid the phenomenon of transplant rejection. In the study conducted by López-Pintor [62], 500 kidney transplant patients were recruited, and during follow-up, six cases of oral cancer were reported out of 500 patients (incidence of 1 patient per 100 subjects).

The same trend was seen for patients undergoing heart transplantation (HTx). Due to new techniques introduced in transplant surgery, survival after heart transplantation has improved significantly in recent decades. In the study conducted by Jääma-Holmberg (2019) [25], the risk of oral cancer after organ transplantation was two to four times higher than that of the general population, becoming one of the main long-term complications in this group of patients. Furthermore, it would appear that oral cancer occurs with a higher frequency in subjects who have undergone thoracic organ transplantation

rather than those who have undergone abdominal organ transplants (i.e., liver and kidney). This different risk of oral cancer in relation to the type of organ transplanted could be partly related to the different pharmacological regimens adopted and partly linked to the underlying pathologies that lead to the need for transplants. Further studies should stratify the risk of oral cancer in relation to the type of organ transplanted.

### *4.3. Other Cancers*

Another cause of immunosuppression associated with a greater risk of developing oral cavity cancer is represented by the treatment of thyroid neoplasms. The number of newly diagnosed cases of thyroid cancer has increased in recent years due to technological advances and the spread of cytological tests for early diagnosis. Patients who underwent partial or total thyroidectomy and those who received radio-iodine treatment for the treatment of thyroid cancer reported an increased risk of developing oral cancer. The study by Hsu et al. (2014) [40] showed an increased association between thyroid cancer and subsequent head and neck cancer. This association found that its biochemical-molecular explanation was related to the intrinsic carcinogenic action of radio-iodine, which can possibly be enhanced by pre-existing molecular genetic mutations in a framework of immunological impairment linked to the partial or total removal of the thyroid.

### *4.4. Infectious Agents*

Other known causes of immunosuppressive states are infectious agents (i.e., HCV, HIV, and HPV). This literature review reported only one study, which was conducted by Su et al. [49] that highlighted an incidence of 698 cases of carcinoma out of 147,962 patients. The risk of oral cancer appears to be lower in HCV patients receiving pegylated interferon (PEG-IFN) therapy than that of untreated HCV patients. Further studies should investigate the role of HCV infection in oral cancer oncogenesis, with particular attention paid to the type of therapy administered to patients.

Studies investigating the role of HIV as a cause of immunosuppression were not included in this review. In fact, it is known that HIV infection causes a depletion of CD4+ T lymphocytes, with consequent impairment of the immune system. Acquired immunodeficiency could, therefore, lead to an increased risk of oral cancer. The study conducted by Precious K. Motlokwa et al. (2022) on an oral cancer population in sub-Saharan Africa did not show an increased risk of carcinogenicity in a group of HIV-infected patients [63]. This could be partly explained by new antiretroviral therapies, which allow clinicians to gain control of HIV infections and, therefore, reduce the impairment of patients' immune systems. Further studies are needed to evaluate whether there is a real risk in HIV-positive patients and whether there are associated risk factors (CD4 T lymphocyte count or traditional antiretroviral therapies vs HAART).

### *4.5. Hematopoietic Stem Cell Transplantation (HSC)*

Within the selected articles, it was possible to identify a group of articles conducted on patients undergoing hematopoietic stem cell transplantation (HSC), which now represents an essential therapy for the treatment of various haemato-lymphoproliferative diseases and other benign conditions (multiple myeloma, lymphomas, autoimmune disorders, etc.). In the study conducted by Santarone et al. (2020) [56], patients undergoing HSC transplantation reported the incidence of developing a malignancy at double the rate of the general population. In support of this, Dyer and colleagues [54] also found a similar incidence rate in patients undergoing HSC transplantation, underlining the importance of regular follow-ups with patients.

Furthermore, GVHD is among the adverse events associated with HSC transplantation. This clinical condition represents an adverse immunological phenomenon following HSC transplantation. GVHD oral lesions are among the so-called potentially malignant disorders, as they have a greater risk of neoplastic degeneration than healthy mucosa. Furthermore, the most frequently used therapy in the treatment of GVHD involves the use

of immunosuppressive agents (e.g., both topical and high potency systemic corticosteroids and calcineurin inhibitors), which, although they reduce the inflammatory component of GVHD lesions, could increase the risk of developing a secondary malignancy. The risk of developing malignancy in patients with chronic GVHD was significantly increased compared with the general population, with a standard incidence ratio (SIR) of 1.8 and a 95% confidence interval (95% CI) of 1.5–2.0. The risk is much higher for cancer of the oral cavity (SIR = 15.7, 95% CI, 12.1–20.1), cancer of the esophagus (SIR = 8.5, 95% CI, 6.1–11.5), colon cancer (SIR = 1.9, 95% CI, 1.2–2.7), skin cancer (SIR = 7.2, 95% CI, 3.9–12.4), and cancers of the nervous system (SIR = 4.1, 95% CI, 1.2–8.4). The risk of developing oral, esophageal, or skin cancer appears to have a maximum incidence 1 year after transplantation [61].

### *4.6. Strengths and Limitations of the Present Systematic Review*

Finally, the data obtained from this systematic review were partly extrapolated from the analysis of national registers from China, Japan, Republic of Korea, India, Taiwan, and Nordic Scandinavian countries. As these databases have a large amount of data, they can lead to significant statistical variations capable of creating very significant discrepancies in the results. In light of this, a meta-analysis dedicated solely to the analysis of the data obtained from these registries was conducted in this systematic review. It is also known that cancer of the oral cavity has a notably high incidence in the aforementioned countries (e.g., China and India) due to the different cultural and social habits. The funnel plot shown in Figure 4 revealed the presence of some studies with particularly discrepant data with respect to the confidence interval of the meta-analysis. Specifically, the study conducted by Levi et al. was discrepant to the funnel plot, and for this reason, it was removed from the statistical analysis and presented only in a qualitative form.

From a methodological point of view, all the studies included in this review had the main objective of investigating the incidence of cancer in other sites. Therefore, further prospective observational studies evaluating the occurrence of oral cancers in immunosuppressed patients as the main outcome while also taking into account the main risk factors of oral cancer that may influence this association (e.g., smoking, candida, HPV, and alcohol) are required. Moreover, it is essential to consider adequate follow-ups to avoid an underestimation of the real incidence of oral carcinomas. The time factor certainly plays an important role in the carcinogenic process, considering that a prolonged state of immunosuppression can increase the risk of the onset of neoplasms.

### **5. Conclusions**

The results obtained from the systematic review indicated that immunosuppression is to be considered a risk factor for the development of oral cancer.

Particular attention and accurate follow-ups with all immunosuppressed patients are, therefore, essential in order to intercept clinical situations at an early stage that could evolve into oral cancer.

Further studies are needed to investigate the effective role of immunosuppression in carcinogenesis and to identify any risk factors.

**Supplementary Materials:** The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers15123077/s1, Table S1. Full-text articles excluded, with reason [64–218].

**Author Contributions:** Conceptualization, C.L.; methodology, R.P.; validation, C.L., M.A.D. and R.P.; formal analysis, G.G.; data curation, F.S. and D.M.; writing—original draft preparation, R.P., F.S. and D.M.; writing—review and editing, C.L. and R.P.; visualization, C.R. and M.A.D.; supervision, M.C. All authors have read and agreed to the published version of the manuscript.

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

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data are available upon request to the corresponding author.

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

### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

### *Systematic Review* **Osteoradionecrosis of the Jaws Due to Teeth Extractions during and after Radiotherapy: A Systematic Review**

**Carlo Lajolo 1,2, Cosimo Rupe 1,2, Gioele Gioco 1,2,\*, Giuseppe Troiano 3, Romeo Patini 1,2,\*, Massimo Petruzzi 4, Francesco Micciche' 5,6 and Michele Giuliani <sup>3</sup>**


**Simple Summary:** Teeth extractions before or after radiotherapy (RT) could be procedures at high risk for osteoradionecrosis (ORN) onset. This systematic review was performed to investigate the ORN incidence following teeth extractions during and after RT for head and neck (H&N) cancer and to evaluate any other possible risk factor. The results highlight how post-RT teeth extractions are a major risk factor for ORN onset (ORN incidence of 5.8%), especially in the mandible, with a diminishing trend in the last years.

**Abstract:** Teeth extractions before or after radiotherapy (RT) could be procedures at high risk for osteoradionecrosis (ORN) onset. This systematic review was performed to investigate the ORN incidence following teeth extractions during and after RT for head and neck (H&N) cancer and to evaluate any other possible risk factor. Methods: This systematic review was conducted according to PRISMA protocol, and the PROSPERO registration number was CRD42018079986. An electronic search was performed on the following search engines: PubMed, Scopus, and Web of Science. A cumulative meta-analysis was performed. Results: Two thousand two hundred and eighty-one records were screened, and nine were finally included. This systematic review revealed an ORN incidence of 5.8% (41 patients out of 462, 95% CI = 2.3–9.4); 3 ORN developed in the maxilla. No other clinical risk factors were detected. Conclusion: Post-RT teeth extractions represent a major risk factor for ORN development, especially in the mandible, with a diminishing trend in the last years. Further research on other possible risk factors might improve this evidence.

**Keywords:** osteoradionecrosis; jaw; head and neck cancer; radiotherapy; tooth extraction

### **1. Introduction**

Among the most common malignancies worldwide, head and neck (H&N) cancers represent the seventh one [1], and almost 75% of patients are treated with radiotherapy (RT), which is either curative or adjuvant or palliative [2]. Unfortunately, RT may cause several side effects, [3] among which osteoradionecrosis (ORN) of the jaws is the most serious.

Signs and symptoms of ORN can vary from pain, sequestration of necrotic bone, and fistulas, to more severe cases with the fracture of the mandible, which can result in sepsis, which is potentially life-threatening, or require major surgical procedures and provoke oral feeding difficulties [4].

ORN can be defined as exposed irradiated bone that fails to heal over a period of three months without evidence of persisting or recurrent tumor; nevertheless, the ORN

**Citation:** Lajolo, C.; Rupe, C.; Gioco, G.; Troiano, G.; Patini, R.; Petruzzi, M.; Micciche', F.; Giuliani, M. Osteoradionecrosis of the Jaws Due to Teeth Extractions during and after Radiotherapy: A Systematic Review. *Cancers* **2021**, *13*, 5798. https:// doi.org/10.3390/cancers13225798

Academic Editor: David Wong

Received: 10 October 2021 Accepted: 16 November 2021 Published: 18 November 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 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 (https:// creativecommons.org/licenses/by/ 4.0/).

definition remains a debated topic, due to the following issues: the possibility of ORN onset without bone exposure and the duration of bone exposure necessary to achieve a definite diagnosis, which varies from 1 to 6 months, according to the literature [5,6]. Furthermore, definitions retrieved in literature do not mention the possibility that patients could present jaw bones necrosis due to antiresorptive therapy (medication-related osteonecrosis of the jaws—MRONJ) [7], which may be administered for other tumors and must be excluded in the differential diagnosis or, at least, taken into debt consideration.

Hypovascularity and hypocellularity subsequent to bone irradiation [6] and the following fibro-atrophic process [8] seem to be crucial in the ORN pathogenesis, forming fragile tissues susceptible to necrosis, especially in cases of tissue damage, such as teeth extractions.

Teeth extractions after radiotherapy are recognized as the most important risk factor for the ORN onset [9–13], with a reported incidence ranging between 2% and 22% of patients [14,15], according to the different studied populations and the different diagnostic parameters.

Nabil et coll. (2011) [9] conducted a systematic review that revealed an overall ORN incidence of 7% in patients who underwent tooth extractions after RT; nevertheless, the high number of factors contributing to the ORN pathogenesis (i.e., tumour site, TNM, oncologic therapeutic protocol, oral general status, site of tooth extraction, flap elevation, antibiotics, and hyperbaric oxygen therapy) make the information necessary to prevent ORN onset after tooth extraction insufficient and inadequate, due to the complexity of the topic.

This systematic review was performed to assess (i) the ORN rate following postradiotherapy tooth extractions; (ii) what is the time-lapse between RT and teeth extraction associated with a lower incidence of ORN; (iii) which other risk factors are associated with the ORN onset; (iv) whether any protocol could prevent or reduce the ORN rate; and (v) whether the ORN rate following the pre-RT tooth extraction is lower than the ORN rate following post-RT tooth extraction.

### **2. Methods**

This systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement criteria [16]. PROSPERO Registration was performed, and the following ID was assigned: CRD42018079986.

### *2.1. Inclusion and Exclusion Criteria*

Inclusion and exclusion criteria are resumed in Table 1.

### *2.2. Search Strategy and Selection of Studies*

An electronic search was performed on the following search engines: PubMed, Scopus, and Web of Science, without specifical filters, from January 1978 to November 2021.

The electronic search strategy was conducted by using a combination of the following MeSH terms and free text words: "Osteoradionecrosis" AND "Dentistry", "Osteoradionecrosis" AND "Prevention", "Osteoradionecrosis" AND "Tooth Extraction", and "Osteoradionecrosis" AND "Tooth Removal".

Two reviewers (G.T. and G.G.) assessed the studies' eligibility in a standardized independent manner. If there was any disagreement, it was evaluated by a third reviewer (C.L.) for the final decision. The screening process was conducted according to the PRISMA flow-diagram (Figure 1). A manual search was also conducted on the following journals: Oral Oncology, Clinical Oral Investigations, Oral Diseases, and European Journal of Oral Sciences. In addition, reference lists of the included articles were manually searched, in order to retrieve any possible full-length papers which could be included.


**Table 1.** Inclusion and exclusion criteria adopted for this systematic review.

Studies not specifying whether ORN developed at the extraction site.

Studies on therapies of patients with ORN were included only if the ORN was effectively due to dental extractions and if the total number of patients receiving tooth extractions was clearly stated. Because many definitions of ORN have been proposed, confusion exists regarding its diagnosis, mainly concerning the time of bone exposure. The assessment of the period of bone exposure is crucial to achieving an ORN diagnosis, because it is not possible to clinically distinguish between a delayed alveolar bone healing and a true ORN. In this revision, studies without a clear definition of ORN were excluded to avoid biases. Abbreviations: ORN, osteoradionecrosis; RCTs, randomized clinical trials; H&N, head and neck.

**Figure 1.** PRISMA flow-diagram of the selection process. Nine articles were finally included in the systematic review and meta-analysis. Adapted from Moher, D et al. (2010) [16]. For more information, visit www.prisma-statement.org (Accessed on 15 November 2021).

### *2.3. Data Collection*

General information on the included papers (i.e., study design, year of publication, country, number of patients, ORN definition, and diagnostic process) and data related to patients (i.e., age and gender, tooth extraction protocol, extraction-related ORN, and other possible risk factors) were collected into a customized table.

### *2.4. Risk of Bias Assessment*

The risk of bias assessment was performed throughout the modified Newcastle Ottawa scale [17] and the Jadad scale [18] (File S1, Supplementary Materials) by 2 reviewers (C.R. and C.L.). In case of disagreement, the final assessment was performed by a third reviewer (G.T.).

### *2.5. Statistical Analysis*

A cumulative meta-analysis was performed with a random effects model in accordance to DerSimonian–Laird method. The pooled proportion (PP) of the rate of ORN occurrence was calculated. The results of the meta-analysis were presented throughout a forest plot graph. The software Open Meta-Analyst version 10 was used to perform the statistical analysis.

### **3. Results**

### *3.1. Results of Search and Study Selection*

The electronic search provided 2281 records (PubMed: 1395 papers, Scopus: 621 papers, Web of Science: 265 papers), and 84 papers were selected for full-paper evaluation. The manual search retrieved six additional articles which underwent a full-text evaluation; providing a total of 90 reviewed papers. Nine articles fulfilled the inclusion criteria and, thus, were included in qualitative and quantitative synthesis [11,12,14,15,19–23]. Table S1 (Supplementary Materials) reports the reasons for the exclusion of the other 81 full-length papers.The selection process is reported as a flow-diagram, following the PRISMA guidelines, in Figure 1.

### *3.2. Study Characteristics and Summary of Results*

This systematic review includes seven retrospective cohort studies, one prospective study, and one clinical trial.

General information on the included papers (i.e., study design, year of publication, country, number of patients, ORN definition, and diagnostic process) is reported in Table 2.

Specific information regarding patients who underwent teeth extractions is presented in Table 3.

Teeth extractions were performed during and after RT on 462 patients out of a total of 800 subjects suffering from H&N cancer. Overall, among these patients, 41 received an ORN diagnosis at the extraction site in a mean follow-up of 40.6 months. The meta-analysis revealed a 5.8% ORN incidence (95% CI = 2.3–9.4, *p* < 0.001). The analysis showed the presence of a high rate of heterogeneity between the studies (I<sup>2</sup> = 8466%). The pooled proportion (PP) and the box plot of the included articles are reported in Figure 2.

Three patients out of 41 developed ORN in the maxilla, while all of the others affected the mandible. Table S2 (Supplementary Materials) shows the details of reported ORN, although only few data could be retrieved.


**Table 2.** Population data of the selected articles: a total of 462 subjects underwent teeth extractions after RT.

\* Although it was not possible to identify a mean value, the study was included because every patient received a follow-up of at least six months. § The prescribed dose to the tissues affected by the neoplasm. <sup>a</sup> Bone exposure longer than 3 months. <sup>b</sup> Bone exposure longer than 6 months. <sup>c</sup> Bone exposure is present in 2 consecutive follow-ups (6–8 weeks for the first two years, 3–4 months after the first 2 years). Abbreviations: Tot, Total; M, Male; F, Female; n., number; RCT, Randomized Clinical Trial; P, prospective; R, Retrospective; RT, radiotherapy; ORN, osteoradionecrosis; EBR, External Beam Radiotherapy; IMRT, Intensity Modulated Radiation Therapy; BT, Brachytherapy; Gy, Gray.

**Table 3.** Characteristics of patients who underwent teeth extraction: among 462 patients who received tooth extractions after RT, 41 ORN were diagnosed.


Abbreviations: n., number; Tot, Total; ORN, osteoradionecrosis; RT, radiotherapy.

**Figure 2.** One-way forest plot of the selected articles shows the PP of the incidence of ORN in irradiated patients receiving teeth extractions during and after RT. Abbreviations: CI, Confidence Interval; Ev, Events; Trt, Total of Patients receiving teeth extractions; I2, Higgins' Hindex.

### *3.3. Risk of Bias Assessment*

The risk of bias assessment for the included papers is reported in Table 4. The methodological quality of the included studies was dis-homogeneous. Four articles out of nine reached a high score, such as Al-Bazie et al. (2016) [11,14], whereas others had an elevated risk of bias. Furthermore, the selection risk of bias was low, since all the inclusion criteria were strict, including only studies performed on a population of irradiated H&N cancer patients who received teeth extractions during and after RT. The shortcomings mostly concerned the comparability and the outcomes domains: in fact, no studies reported other confounders (i.e., antiresorptive drugs), and only a few studies reached one year of follow-up after teeth extractions and outlined the drop-out rate.


#### **Table 4.** Modified Newcastle-Ottawa Score and Jadad scale.

### *3.4. Results of Individual Studies*

Results of individual studies among patients who underwent teeth extraction after radiotherapy are reported in Tables 2 and 3.

### *3.5. Excluded Studies*

The reasons for the exclusion of the other 81 full-length papers are summarized in the Table S1 (Supplementary Materials), available electronically.

In particular, 22 studies did not reach an adequate sample size to be included; 17 studies provided an inadequate definition or diagnosis of ORN; 11 studies had a design not fulfilling the inclusion criteria (reviews, letters to editor); 15 studies analyzed a cohort not representative of the whole population of patients undergoing tooth extractions during or after RT; seven studies did not reach an adequate follow-up (six months after tooth extraction); nine studies diagnosed ORN cases, but it was not clear whether the ORN developed at post-extraction sites.

The study conducted by Schweiger et al. (1987) [24] was remarkable; nevertheless, it did not fulfill the inclusion criteria: the authors made an ORN diagnosis after one month of bone exposure. Notably, a medical examination conducted one month after tooth extraction may overestimate the ORN rate. In fact, the authors reported a higher risk of ORN incidence (8%) following post-RT dental extractions.

The study conducted by Saito et al. (2021) [25] was well conducted; nevertheless, as the authors declared in their discussion section, it was not possible to distinguish if ORN was present at the moment of the extraction or if it was a consequence of the post-RT dental extraction. This could have led to an overrating of ORN incidence (28.1%, as reported by the authors).

Another recent study, performed by Kubota et al., 2021 [26], showed good methodology. Nevertheless, the authors did not specify whether the ORN developed at the post-RT extraction site.

### **4. Discussion**

The role of dentists in the H&N cancer supportive therapy is becoming fundamental. The main objectives of dental treatment in these patients, before radiotherapy, are the removal of oral foci and, after radiotherapy, the prevention and therapy of dental diseases and the side-effects of radio-chemotherapy involving the oral cavity. Development of more accurate radiotherapy techniques (e.g., IMRT) has decreased the number of side-effects in the oro-maxillofacial district [27]; nevertheless, ORN remains the most important event, and together with severe mucositis, which sometimes undermines a patient's life, it can occur in 2% to 22% of irradiated subjects [14,15]. Since teeth extractions performed after the RT represent the main risk factor for ORN onset, dentists should prevent dental diseases to minimize the number of extractions after the RT, and in the case where extraction is necessary, dentists should apply specific protocols to decrease the risk of the onset of ORN.

However, the possible progression of dental diseases, precipitated by the consequences of RT on oral and maxillofacial tissues (e.g., radio-induced caries), and the increase in life expectancy determine the possibility to perform dental extractions in patients who received radiotherapy for H&N cancer [1,28,29]. This systematic review showed an ORN rate of 5.8% in patients undergoing tooth extractions after RT, in accordance with the systematic review conducted by Nabil et coll. (2011) [9]. Comparing the final data obtained from this systematic review (5.8% of ORN in post-RT) with those of extractions performed before radiotherapy (2.2%), reported in a systematic review already conducted by our research group [30], it seems reasonable to consider post-RT extractions as a high-risk procedure and suggest performing them before starting RT. These results are in contrast with the findings emerging from another systematic review, which did not retrieve statistically significant differences in the ORN risk between patients undergoing tooth extractions before RT and patients undergoing tooth extractions after RT [31]. Although it is not easy to find an explanation for these differences in the results, it could be related to less restrictive inclusion and exclusion criteria adopted by Beaumont S et al. (2021). Nevertheless, both the reviews show how a thorough analysis of the risk factors needs to be performed, by means of new clinical trials, in order to reach a better understanding of the pathogenesis of ORN, as further discussed in the discussion section.

However, if we analyze the incidence of ORN in the papers included in this review, it is very uneven: notably, as presented in Figure 2, incidence varies from 50% to 5.6% in the articles prior to 1990 and is up to 0% in articles published from 1990 to today. Therefore, it seems that post-RT extractions no longer involve this risk, unlike pre-RT extractions, which despite a decreasing trend, still show a certain percentage of ORN, and this has been observed in recent studies too (e.g., 7.6% in Schuurhuis, 2011 and 13% in Batstone, 2012) [32,33]. Nevertheless, a recent study conducted by Kubota H et al. (2021) reported an ORN rate of 7.5% in patients who underwent radiotherapy during the last decade [26]. Further studies are needed in order to better clarify the real incidence of ORN. This different frequency of ORN for post-RT extractions, between studies conducted before and after 1990, appears notable but is difficult to fully understand.

Possible explanations are the introduction of the more advanced technique (IMRT) that could have contributed to the progressive reduction of this incidence. IMRT selectively irradiates the tumor, giving a significantly lower dose to healthy tissues. In the 1980s, the transition from traditional 2D to conformed 3D (3DCRT) treatment represented a critical advance in RT. In 3DCRT, simulation and treatment planning are based on computerized tomography (CT), reaching a precise definition of the area affected by neoplastic disease and a more accurate dose calculation. Afterwards, the introduction of IMRT, a highly specialized typology of conformative therapy, through the modulation of the beam flow, allowed the irradiation of the target site with a non-uniform intensity, increasing the dose

only to cancer tissues. Furthermore, it allowed the use of multiple irradiation planes, including oblique and not coplanar planes, which together with the use of multilamellar collimators, ensure adequate irradiation of tumor tissues and the saving of healthy tissues, including alveolar bone. However, in many of the studies analyzed, the radiotherapy technique used was unknown.

Furthermore, the increased involvement of dentists in the management of H&N cancer patients could have improved oral conditions of patients post-RT: careful dental treatment before the beginning of RT (e.g., extraction of all teeth with uncertain prognosis), a thorough dental follow-up after the RT (e.g., interception of any possible dental diseases at early stages), and supportive therapies (i.e., oral hygiene recalls and professional fluoride therapy) may contribute to a better oral health after RT. The result of such careful management could mean (1) a lower number of extractions per patient, (2) less inflamed/infected foci, (3) a more accurate extraction planning, and (4) better general oral health conditions.

Our first consideration focuses on the critical issues of the definition and diagnosis of ORN. In accordance with the literature published in the last 15 years, we included only those studies that provided a clear definition of ORN and in which the ORN was diagnosed in the case of irradiated bone, exposed in the oral cavity, for a minimum of three months, with no local recurrences [5,6,34]. Most of the excluded studies, analyzed in full-text, provided no clear definition of the disease. We considered it essential that a clear definition of ORN was present in the study; in the literature, there are several definitions which differ from each other in the length of time of bone exposure and about the bone exposure as a main sign of ORN diagnosis. Although the bone exposure has to linger in post-extractive alveoli for a period of time such as to exclude delayed healing of the alveolus (i.e., dry socket), there is no agreement in defining the post-extraction time interval after which an ORN may be diagnosed. A short time interval could notably overestimate the real ORN rate; by contrast, a long time interval could underestimate the real rate of ORN because some ORN can heal spontaneously, going through bone sequestration, and therefore not be correctly diagnosed. Furthermore, some authors described the possibility that ORN occurs even without bone exposure [35]. Therefore, considering exposed bone as the only sign of ORN, the ORN rate could be underestimated due to a misdiagnosis or to a diagnostic delay. Further research should provide a clear definition of ORN so that it would be possible to compare the results and provide data with a stronger level of evidence.

Another relevant methodological bias that we found from the analysis of the literature concerns the outcome: most of the studies provided information on the number of patients with ORN without providing any information on the number of sites affected by ORN. Considering that ORN may occur in more than one site in the same patient, further research might provide a precise indication of the sites affected by ORN in relation to the post-extraction site. Moreover, to provide a specific risk of ORN onset at post-extractive alveoli, the studies should provide more precise information on the affected sites subjected to extraction (in the irradiated patient population). Contrariwise, most of the studies provided no information on either the sites undergoing post-RT extractions or on the number of post-extractive sites affected by ORN, except Marx et al. (1995) [20].

Some noteworthy clinical considerations concern the anatomical site of tooth extraction. Mandibular jaw appears to be a risk factor of ORN onset following teeth extractions. This systematic review reported only three cases of ORN in the maxilla, while all the other cases developed in the mandible. Unfortunately, it was not possible to clarify the ORN risk related to anatomical site, since the included articles did not report data regarding the anatomical site of extracted teeth in the overall population undergoing RT. Another clinical consideration concerns the surgical technique adopted for the extraction of teeth in patients that received irradiation. Non-surgical extractions are less invasive; however, the lifting of a flap allows the closure of the post-extraction site by first intention and the possibility to modify the bone morphology when necessary. Nowadays, little is known regarding whether any innovative surgical technique can decrease the ORN risk. Marx et coll. (1985) and Maxymiw et coll. (1991) performed all teeth extractions without lifting

a flap [12,20]. The ORN rate found by these authors was somewhat discordant: Marx diagnosed 35 ORN out of 291 extracted teeth, and Maxymiw diagnosed no ORN out of 449 extracted teeth. However, the other included articles did not report sufficient data regarding teeth extraction techniques. Further studies are necessary to confirm whether the extraction technique influences the risk of ORN.

Another little-known aspect concerns the reasons to perform dental extractions in this specific cohort of patients: none of the included articles provided information on this matter. Notably, an assessment should be performed as to whether the motivation for a tooth to be extracted could favor the onset of ORN, bearing in mind that the non-extraction of teeth affected by inflammatory-infectious processes could represent a trigger for the onset of ORN, similar to what occurs for MRONJ [36]. By contrast, it seems reasonable that extractions of teeth affected by an inflammatory-infectious process may represent a higher ORN risk procedure. However, post-irradiated alveolar bone could be affected by spontaneous ORN, miming in the early stages an inflammatory-infectious process, overestimating the risk of ORN consequent to post-RT extraction. The articles included in this review do not provide information regarding this topic.

A necessary consideration is relative to the dose received by the post-extraction sites, which could be considered a risk factor for ORN onset. The patients affected by ORN received an average dose of 68 Gy. Unfortunately, it was not possible to define a threshold, since the included articles did not provide information for the specific post-extraction alveoli. Nevertheless, a reasonable opinion is that high-dose radiation therapy increases the risk of ORN.

A highly debated topic in the literature concerns the identification of a time interval after the end of the RT, beyond which the surgical procedures may be safer or associated with a lower risk. Although a reasonable judgement seems to be that postponing the extraction can reduce the risk of ORN, alterations of bone metabolism could persist or worsen several years after the end of radiation therapy. This systematic review showed that patients who developed ORN had a mean time interval from RT to dental extractions longer than the whole population (33 months vs. 24.7 months); these data, contrariwise to general opinion, seem to suggest that a longer time-lapse between RT and ORN could not prevent the ORN onset. However, this information was reported in only two of the four studies that diagnosed ORNs and refers to average values. Specifically, Beumer et al. (1983) [15] conducted dental extractions at different time intervals (7–60 months), and the time-interval from RT to dental extraction was not associated with a higher ORN risk. Although the most recent evidence seems to confirm this result [31] and some authors suggest performing tooth extractions in the immediate post-RT period [26], it is important to consider the possible existence of a "bimodal pattern" of RT damage, showing two different peaks of risk: 12 months after the end of RT and 24–60 months after RT [10]. At present, no controlled studies allow a conclusion regarding the existence of a time interval that reduces the risk of ORN; therefore, this topic warrants further investigation.

Among the risk factors to be evaluated in the estimate of the onset of ORN, the influence of any previous or ongoing medical therapy that may enhance the risk must also be considered. The increased number of patients undergoing medical treatments with antiresorptive, antiangiogenetic, and biological drugs (e.g., denosumab, bisphosphonates) for oncological or metabolic reasons makes it necessary to conduct an accurate interview of the medical history of each patient [37]. Studies included in this review provided no information on this regard. A critical clinical consideration pertains to the different perioperative medical support protocols reported in the literature to reduce the ORN risk. Among those protocols, antibiotics associated with antiseptic rinses are the most used. There is strong evidence that the deeper zones of necrotic bone are colonized by bacteria of the oral district, so much so that the pathogenetic idea of aseptic necrosis has been repeatedly challenged over time. In the study conducted by Al-Bazie et al. (2016) [11] and Maxymiw et al. (1991) [12], the antibiotic prophylaxis with amoxicillin and penicillin V was included in the protocol and was effective in the prevention of ORN, reporting an ORN

rate of 0% (0 ORN out of 161 patients). Additionally, Marx et al. (1985) [20] and Epstein et al. (1987) [21] performed antibiotic prophylaxis; however, their studies showed a higher ORN rate of 35.4% and 5.56%, respectively (altogether, 16 ORN cases out of 91 patients, indicating an ORN rate of 17.58%). Further studies with a larger sample size are therefore needed to clarify the real usefulness of antibiotics in preventing ORN.

Hyperbaric oxygen therapy (HBO) is another peri-operative support provided. The rationale for using HBO is based on the impact of an increased amount of oxygen on hypoxic tissues. Locally, HBO increases the amount of growth factors, including those playing an active role in angiogenesis. Oxygen can also promote an antibacterial effect on the trauma site. Based on the available evidence, the effectiveness of HBO in preventing ORN is debated [38,39]. The articles included in this systematic review did not provide sufficient data regarding the effectiveness of HBO. Further trials are needed to resolve the controversy [37].

Another consideration should be done among new drugs proposed for ORN medical therapy (i.e., pentoxifylline, tocopherol) that could also represent a new approach to the prevention of ORNs [40]. Thus far, none of the studies has analyzed this aspect: future clinical studies might evaluate the preventive role of these drugs for ORN onset.

### **5. Conclusions**

This systematic review highlights that dental extractions after RT are procedures at high risk of ORN, especially in the mandible. It was impossible to draw definitive conclusions about other clinical risk factors, including the time-lapse to respect between RT and tooth extractions. Data gathered from the analyzed literature presented a higher rate of ORN (5.8%) when compared with extractions performed before RT (2.2%) [30]; even if the general trend of ORN is decreasing for both pre- and post-RT extractions, studies performed on extraction after RT presented a peculiar bimodal trend: studies before 1990 show a much higher ORN rate compared with those performed after 1990, which are proximate to 0%. Reasons for this bimodal behaviour are not completely understood; possible explanations are that the introduction of the more advanced radiotherapy techniques and the greatest role of the dental clinician for H&N cancer supportive therapy could have improved oral conditions of patients after RT. Further research among other possible risk factors should be conducted to investigate their role in ORN development.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/cancers13225798/s1, File S1: Modified Newcastle-Ottawa quality assessment tool forms for observational studies, Table S1: Articles excluded from systematic review and reasons for their exclusions, Table S2: Details of reported ORN Patients. Gender, Age, and Tumor Site of ORN Patients are not reported because only a few data were retrieved.

**Author Contributions:** All authors contributed to the study conception and design. Material preparation, data collection, and analysis and the first draft of the manuscript were performed by C.L., G.G. and C.R. All authors commented on previous versions of the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** No funding to declare.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data are available upon request to the authors.

**Conflicts of Interest:** None of the authors have any conflicts of interest to declare.

### **References**


## **Development and Validation of Prognostic Models for Oral Squamous Cell Carcinoma: A Systematic Review and Appraisal of the Literature**

**Diana Russo 1, Pierluigi Mariani 1, Vito Carlo Alberto Caponio 2, Lucio Lo Russo 2, Luca Fiorillo 3, Khrystyna Zhurakivska 2, Lorenzo Lo Muzio 2,4, Luigi Laino <sup>1</sup> and Giuseppe Troiano 2,\***


**Simple Summary:** Prognostic models to choose the right treatment schedule are needed in order to translate into practice a personalized approach. None of these models have been still entered into the clinical practice for what concern oral squamous cell carcinoma (OSCC). In this manuscript we performed a systematic review and subsequent quality assessment of already development prognostic model for OSCC with the aim to take stock of the situation on their possible clinical use.

**Abstract:** (1) Background: An accurate prediction of cancer survival is very important for counseling, treatment planning, follow-up, and postoperative risk assessment in patients with Oral Squamous Cell Carcinoma (OSCC). There has been an increased interest in the development of clinical prognostic models and nomograms which are their graphic representation. The study aimed to revise the prognostic performance of clinical-pathological prognostic models with internal validation for OSCC. (2) Methods: This systematic review was performed according to the *Cochrane Handbook for Diagnostic Test Accuracy Reviews* chapter on searching, the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) guidelines, and the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS). (3) Results: Six studies evaluating overall survival in patients with OSCC were identified. All studies performed internal validation, while only four models were externally validated. (4) Conclusions: Based on the results of this systematic review, it is possible to state that it is necessary to carry out internal validation and shrinkage to correct overfitting and provide an adequate performance for optimism. Moreover, calibration, discrimination and nonlinearity of continuous predictors should always be examined. To reduce the risk of bias the study design used should be prospective and imputation techniques should always be applied to handle missing data. In addition, the complete equation of the prognostic model must be reported to allow updating, external validation in a new context and the subsequent evaluation of the impact on health outcomes and on the cost-effectiveness of care.

**Keywords:** oral squamous cell carcinoma; nomograms; prognostic models; overall survival; prognosis; systematic review

### **1. Background**

Head and Neck Cancer (HNC) is the sixth most common type of cancer across the world with nearly 550,000 new cases per year. Most of HNCs are diagnosed as Oral

**Citation:** Russo, D.; Mariani, P.; Caponio, V.C.A.; Lo Russo, L.; Fiorillo, L.; Zhurakivska, K.; Lo Muzio, L.; Laino, L.; Troiano, G. Development and Validation of Prognostic Models for Oral Squamous Cell Carcinoma: A Systematic Review and Appraisal of the Literature. *Cancers* **2021**, *13*, 5755. https://doi.org/10.3390/ cancers13225755

Academic Editor: Petra Wilder-Smith

Received: 6 September 2021 Accepted: 13 November 2021 Published: 17 November 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 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 (https:// creativecommons.org/licenses/by/ 4.0/).

Squamous Cell Carcinomas (OSCC) and oral cancer ranks eighth among the most common causes of cancer-related deaths worldwide [1,2]. Both pharmacological and surgical protocols for OSCCs diagnosed in early stages are less aggressive and characterized by better outcomes, whilst in advanced stages, very high patients' morbidity and poor clinical outcomes are expected [3]. Despite the increased knowledge and the encouraging scientific findings of the past 20 years on such diseases, the overall 5-year survival rate for OSCC is still below 50% [4].

Nowadays, the Tumor-Node-Metastasis (TNM) staging system is employed worldwide to predict tumor prognosis and to guide physicians towards the correct treatment choice, however, survival outcomes in patients classified within the same TNM stage class could be dramatically different, with discrepancies in therapy response and tumor management [5].

One of the main limitations of OSCC-related TNM system is its main focus on the anatomical extension of the disease. However, within each staging group, the prognosis can be modified by tumor-related factors, such as genetics, patient age, sex, race or comorbidities. For this reason, the need for a more "personalized" approach to the oncologic patient was underlined in the recent eighth edition of the American Joint Committee On Cancer (AJCC) staging system [6]. It is, therefore, necessary to investigate further prognostic factors to construct prognostic models to carry out a personalized prognosis evaluation [7,8].

Recently, there has been an increased interest in the development of clinical prognostic models and, in particular, in nomograms which are their graphic representation [9]. These are a set of mathematical algorithms that can be used to predict patient outcomes by incorporating multiple variables. Clinic-pathological and genetic variables are mainly incorporated in OSCC prognostic models, showing interesting evidence of their role in patients' prognosis [10,11]. Purpose of these models is to estimate the probability or individual risk that a given condition, such as recurrence or death, will occur in a specific time by combining information from multiple prognostic factors of an individual [12].

Due to the recent interest in these new prognostic tools, and their potential important role in clinical practice, some guidelines have been defined for explanation and elaboration of clinically useful and correctly elaborated prognostic model. These Guidelines are reported in the Prognosis Research Strategy (PROGRESS) 3 and the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) [7,13]. In 2016 the AJCC developed the acceptance criteria for inclusion of risk models for individualized prognosis in the practice of precision medicine in the systematic reviews [14]. In the same year, Debray et al. developed a guide for systematic reviews and meta-analyzes of the performance of prognostic models [15]. Additionally, the Prediction Model Risk of Bias Assessment Tool (PROBAST) was also developed to assess the risk of bias and the applicability of diagnostic and prognostic prediction model studies [16].

In this scenario, this study presents a systematic review of clinical-pathological prognostic models with internal validation for OSCC, using the AJCC inclusion criteria and according to current published guidelines.

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

### *2.1. Protocol*

This systematic review was performed according to the *Cochrane Handbook for Diagnostic Test Accuracy Reviews* chapter on searching [17], the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) guidelines [18], and the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) [19]. The reviews aim was to evaluate the prognostic performance of nomograms in patients with OSCC. This protocol was designed a priori and registered on the online database PROSPERO (CRD42020219937).

### *2.2. Search Strategy*

Studies were identified by using different search engines: Medline/PubMed, ISI Web of Science and SCOPUS. In addition, partial research of the gray literature was carried out through Google Scholar. Furthermore, bibliographies of included studies were handedrevised to find further studies to include in this review. Search operations ended in October 2020. For the search strategy, MeSH terms and free text words were combined through Boolean operators as follow: (prognostic model OR prognostic index OR prediction model OR signature OR risk assessment OR prognostic assessment OR nomogram OR risk score OR model stratification) AND ((OSCC OR "oral cancer" OR tongue) NOT (gastric OR laryngeal OR pharynx OR endocrine OR colorectal OR breast OR prostate OR lung OR salivary OR review OR meta-analysis)).

### *2.3. Eligibility Criteria*

To be included, studies had to fulfill the following criteria: (i) characteristics of the prognostic model had to be reported, together with their representative alternative presentation (e.g., scoring system, nomogram, etc.) for patient diagnosed with OSCC undergoing surgery with or without adjuvant therapy; (ii) at least one between with Overall Survival (OS) and Disease-Free Survival (DFS) had to be reported as outcome; (iii) studies had to follow TRIPOD and CHARMS checklist [13,19]; (iv) the prognostic model had to be internally validated; (v) and based on clinicopathological prognostic factors; (vi) that met all the thirteen inclusion criteria described by AJCC [9]; (vii) cohort studies, retrospective studies and studies that performed external validation of a pre-existing model were included; (viii) published in English; (ix) with available full text. We excluded: (i) case reports; case series; reviews and meta-analysis; (ii) studies that intend to modify existing prediction models and not to create new ones; (iii) studies including prognostic models that are not based on measurable markers in resected tumor tissue (saliva, blood, etc.); (iv) studies that met the three AJCC exclusion criteria [9].

### *2.4. Article Selection, Data Collection Process, and Data Items*

Articles were independently selected by two of the authors (D.R., P.M.) in multiple steps. First, results of different databases were crossed, and duplicates were electronically removed by EndNote v.X9 software. Subsequently, a manual check was performed to furtherly remove previous undetected duplicates. The first screening for inclusion was performed by reading title and abstract. Full assessment for eligibility was furtherly carried out by full-text reading, judging each study as included, excluded or uncertain, according to the previously listed criteria. A third reviewer (G.T.) acted as an arbiter and calculated a value of k-statistic to ascertain the level of reviewers' agreement. In cases of disagreement, the same author (G.T.) took a final decision. From each of the selected articles, relevant information were extracted into a data extraction sheet using the TRIPOD and CHAMRS checklist, such as: author, year of publication, country where the study was carried out, the title of the paper, sample size, internal validation sample size, tumor localization sub-site, predictors (candidate and final) used to develop the models, outcome of the model (OS, DFS), method for the internal validation was carried out, modelling method, handling of missing data, model discrimination, model calibration, model presentation, handling of continuous predictors, presence of external validation, type of study.

### *2.5. Risk of Bias Assessment*

Risk Of Bias (ROB) within individual studies was assessed by using Prediction model Risk Of Bias Assessment Tool (PROBAST) [16]. PROBAST can be used to assess any type of prognostic prediction model aimed at individualized predictions regardless of the predictors used. The tool comprises four domains—population, predictor, outcome, analysis, questions are answered as "yes", "probably yes", "probably no", "no", or "no information". Risk of bias is summarized as "low", "high", or "unclear". The degree of applicability is rated as "low", "high", or "unclear" concern. The "unclear" category

should be used only when reported information is insufficient. In both cases, for both ROB and applicability, an overall judgment is provided. ROB was assessed separately for development (comprising internal validation) and external validation settings. For articles reporting both model development and external validation, the risk of bias was assessed independently.

### **3. Results**

A total of 5972 records were identified in the initial search and were screened by title and abstract by two reviewers. Among these, 66 match our eligibility criteria and were furtherly assessed by full-text reading. At the end of selection process, 6 articles were considered suitable for inclusion in this systematic review [20–25]. Details on the selection process and reasons for exclusion are shown on Figure 1.The value of k-statistic resulted 0.87, indicating an excellent level of agreement between reviewers.

**Figure 1.** Flow-chart: 5972 records were identified in the initial search and, among them, 66 were further evaluated by reading the full text. At the end of the selection process, 6 articles were considered suitable for inclusion in this systematic review.

### *3.1. Study Characteristics and Model Development*

All studies were published between 2014 and 2019. Prognostic models were mainly developed in China (50%, *n* = 3) [22,24,26], the remaining in India (33.3%; *n*= 2) [21,25] and in USA (16.6%; *n* = 1) [23] (Table 1). Patient data were collected retrospectively and hospitalbased in four studies [21,23,25,26], while in two studies these were collected from the SEER database [22,24]. Data of patients' samples and tumor characteristics are summarized on Table 1.



**1.**Featuresofthemodelsincluded.

The main investigated prognostic factor was age (100%; *n* = 6) [21–26], in four articles T stage [22,23,25,26], N status [22,24–26] and sex [21–23,26] are inspected, while three studies looked into histological grade [22,24,26] and subsite of the tumor onset [23,24,26]. Main final factors that were found to be independently associated with OS were age and race. Candidates and final prognostic factors included in prognostic models are reported on Table 2. None of the studies evaluated DFS, while OS resulted to be the main outcome (Table 1). Multivariable Cox proportional hazards was used as developer model in 50% of studies [22,24,26], alternatively to a combined modelling method using multivariable Cox proportional hazard regression models and stepdown reduction methods [21,23,25]. Only Montero et al. reported how missing data were handled, by implementation of an imputation technique [22].


**Table 2.** Predictors included in the prognostic models.


In most of the prognostic models (66%, *n* = 4) [22,23,25,26], continuous predictors were dichotomized or categorized, hence the nonlinearity of continuous predictors was assessed. For two prognostic models, cubic splines were used to test for the presence of, a non-linear association between continuous predictors and the predicted outcome [23,26].

All the studies used a nomogram as final presentation [21–26]. Methodological characteristics of prognostic models developed are summarized on Table 3.

### *3.2. Validation of the Models*

Internal validation was performed in all studies by 1000-time bootstrapping [21–23,25,26], except Sun et al. who employed a combined 500-time bootstrapping and 5-fold crossvalidation methodology [23].

As a method of discrimination, C-statistics has been used in five studies [21–25]; only one study performed AUC [26].

Four studies reported assessed calibration of the model by means of calibration plots [22–24,26], while two did not describe their calibration method [21,25].

In all studies, predictive accuracy was quantified by calculation of the Concordance index (C-index) for each outcome, all the included studies had a C-index higher than 0.6 [21–26]. External validation was performed in four studies and C-index was found to be higher than 0.6 in all the articles included [22,24–26]. Methodological features of the development and validation of prognostic models are listed on Table 3.


**Table 3.**

Methodological

characteristics

 of prognostic models developed.

### *3.3. Risk of Bias*

PROBAST was used to assess the risk of bias of included studies. Four models presented a low overall bias level [21,22,24,26], while two reported a high overall bias level [23,25]. The overall applicability level resulted to be low in all studies [21–24,26], except one [25]. Four out of six studies performed external validation of the models [22,24–26]. The overall risk of bias was low in three out of four models [22,24,26]. In the external validations, applicability was found to be low in all studies [22,24–26]. The risk of bias for each domain of the developed models and the external validations is shown, respectively, on Figures 2 and 3. The applicability for each domain, both for the developed models and for the external validations, is reported in Tables 4 and 5.

**Figure 2.** Risk of bias of the developed prognostic models: For each of the six prognostic models included in this systematic review, four domains of bias (population, predictors, outcomes, analysis) were evaluated as "high" or "low". In this way, the overall risk of bias of each article was assessed.

**Figure 3.** Risk of bias of models' external validations: For each of the four externally validated prognostic models included in this systematic review, four domains of bias (population, predictors, outcomes, analysis) were assessed as "high" or "low". In this way, the overall risk of bias of each article was assessed.


**Table 4.** Applicability of the developed prognostic models.

For each of the six prognostic models included in this systematic review, four domains (population, predictors, outcomes, analysis) were evaluated as "high" or "low". In this way, the overall applicability of each article was assessed.

**Table 5.** Applicability of models' external validations.


For each of the four externally validated prognostic models included in this systematic review, four domains (population, predictors, outcomes, analysis) were assessed as "high" or "low". In this way, the overall applicability of each article was assessed.

### **4. Discussions**

An accurate prediction of cancer survival is very important for counseling, treatment planning, follow-up and postoperative risk assessment in patients with OSCC [27]. Although the use of prognosis models is still relatively new for OSCC, these models are already widely used for other human diseases [28–31]. It is now well known that cancerrelated outcomes are influenced by several factors that are not included in the TNM system. The vast majority of these factors has not been incorporated into the staging system because they may not predict outcome "independently" in multivariate prognosis models, however many of them may work in tandem and have varying degrees of influence on each other [32,33].

This systematic review has yielded a detailed picture of prognostic models for predicting OS in patients with OSCC. Six studies included in this review correctly developed models according to the TRIPOD, all the included studies carried out internal validation of the model and four models were also externally validated [21–26]. The majority of models assessed OS in patients with squamous cell carcinoma of the tongue [22,24,26], two assessed all possible sites of tumor onset [21,23], and one model only assessed the buccal mucosa cancer [25]. All models rated OS at five years, except for Bobdey et al [25]. who only rated it at three years; furthermore, Li et al. and Sun et al., also evaluated OS at eight and three years respectively [21,23]. Among the clinical factors, those most included in the models are age, race, martial state, comorbidities and smoking; while among the histopathological ones the most investigated were T stage, N stage and M stage.

This systematic review showed methodological differences in model development. It is well known that the performance of a prognostic model is overestimated when it is just assessed in the patient sample that was used to build the model [34]. Internal validation provides a better estimate of model performance in new patients when done by adjusting overfitting, that is the difference between the accuracy of the apparent prediction and the accuracy of the prediction measured on an independent test set. Resampling techniques are a set of methods to provide an assessment of accuracy for the developed prognostic prediction models [35]. As an exception, Sun et al. [23] used a combined bootstrapping and cross-validation method, although all other studies used 1000-time bootstrapping as a resampling technique. Nevertheless, an evaluation of a model's performance by using bootstrapping or cross-validation is not enough to overcome overfitting, such type of studies should also apply shrinkage, which is a method used adjust the regression coefficients [36,37]. However, none of the studies used this technique, probably because its usefulness for models with a low number of predictors is unclear [13].

Another important finding from our review is that one-third of the studies did not report on model calibration [38]. Calibration reflects the agreement between the model's predictions and the observed outcomes. It is preferably reported graphically, usually with a calibration plot [39]. Another key aspect of the characterization of a prognostic model is discrimination, that is, the ability of a forecasting model to differentiate between those who experience the outcome event or not [13]. The most used measure for discrimination is the Concordance Index (C-index), which reflects the probability that for any pair of individuals randomly, one with and one without the outcome, the model assigns a higher probability to the individual with the outcome [40]. For survival models, many c-indices have been proposed, so it is important to underline that, from our results, the most commonly used is the discrimination model proposed by Harrell [41]. In any case, discrimination can vary in a range from 0 to 1 and is considered good when higher than 0.5, considering that all the studies included in this systematic review presented a C-index at least higher than 0.6, all of them showed a good prognostic accuracy [42]. In addition, improvements in study design and analysis are crucial to allow evidence of more reliable prognostic factors that can be incorporated into new prognostic models, or to update existing models, to improve discrimination [43]. Another important finding was the almost total lack of handling of the missing data, except for Montero et al. [22] who carried out the multivariate imputations by chained equations (MICE) [44] before conducting multivariable regression statistical analysis [23]. The absence of a mention of the missing data leads to a so-called "full case analysis". Including only participants with complete data, as well as being inefficient as it reduces the sample, can also lead to biased results due to a subsample [12]. Additionally, in only two prognostic models, continuous predictors were dichotomized or categorized, and the non-linearity of continuous predictors was examined using restricted cubic splines [23,26].

In the end, only four prognostic models performed external validation, in none of these the population in which the validation was performed was specifically reported and this data also negatively influenced the risk of bias. External validation is preferable to internal validation for testing the transportability of a model since it is impossible for the population, or distribution of predictors, in an independent population to be the same as in the model development population [45]. Secondly, to improve the generalizability of a model, it should ideally be validated in different contexts with different population [46]. Furthermore, in the literature, there are currently no external validation by independent researchers of prognostic models for OS in patients with OSCC. A reliable model should be tested by independent researchers in different contexts to ensure the generalizability of prognostic models [15].

Most of the prognostic models in the literature describe the development of the model, a small number report external validation studies and currently, there are no studies considering clinical impact or utility [7]. Identifying accurate prognostic models and performing impact studies to investigate their influence on decision making, patient outcomes and costs is a fundamental component of stratified medicine because it contributes evidence at multiple stages in translation [47].

Multivariable Cox proportional hazards regression models were used to developing the models, as indicated for survival data [48]. All included prognostic models used nomogram as model presentation, yet none of the prognostic models reported the original mathematical regression formula. This turns out to be highly limiting, firstly because this presentation format is not a simplification of a developed model, but rather a graphical presentation of the original mathematical regression formula, and secondly, because recalibration, and updating of the original formula is necessary to perform validation [49]. Furthermore, it would be advisable to provide readers with the appropriate tools for the interpretation and application of the nomogram [30].

All the studies included in this systematic review had a retrospective design, and therefore showed issues related to missing data and a lack of consistency in predictor and outcome measurement [16]. In addition, both the single-institutional studies and the SEER database lacks critical information. The former, being the cohort of similar patients, may not be relevant in predicting the risk of other patient populations. The second lacks information that could be relevant to prognosis such as comorbidities, chemotherapy and tobacco smoking [50]. Prospective cohort studies should be performed for predictive modeling since they enable not only clear and consistent definitions but also prospective measurement of predictors and outcomes [13,50].

The recognition of the methodological limitations found in the developed models and their external validation were evaluated as a high risk of bias, as indicated in the PROBAST. Domain four (analysis domain) is the one that most influenced the overall risk of bias [16,51].

### **5. Limitations**

The main limitations related to this systematic review are due to the very strict inclusion criteria to ensure the high accuracy of the contents. Certainly, having selected only internally validated models and articles written in English has strongly restricted the number of studies included. However, as this is the first systematic review of the literature on prognostic models for OSCC patients, this was done to provide clinicians and researchers with a clear picture of the correct model development method. Future systematic reviews should include a greater number of outcomes (cancer-specific survival, recurrence-free survival, etc.) and include biomolecular prognostic factors in addition to clinicopathological one.

### **6. Conclusions**

Based on the findings of this systematic review, the following recommendations could be reported: (i) model development studies should weight for overfitting by carrying out internal validation (by resampling techniques such as bootstrapping) and using shrinkage techniques, (ii) model calibration and discrimination should always be examined, (iii) imputation techniques for missing data handling should always be applied, (iv) nonlinearity of continuous predictors should be examined, (v) the complete equation of the prognostic model should always be reported to allow external validation and updating by independent research groups; (vi) prospective studies should be performed to reduce the risk of bias (vii) external validation in a new context and impact assessment on health outcomes and cost effectiveness of care should be carried out.

**Author Contributions:** D.R. and P.M. contributed to data acquisition and drafted the manuscript; V.C.A.C., K.Z., L.L.R., L.F. and L.L.M. contributed to data analysis and interpretation; K.Z., L.L. and G.T. contributed to conception, design and critically revised the manuscript. All authors read and approved the final manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are freely available in the article.

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

### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland www.mdpi.com

*Cancers* Editorial Office E-mail: cancers@mdpi.com www.mdpi.com/journal/cancers

Disclaimer/Publisher's Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Academic Open Access Publishing

mdpi.com ISBN 978-3-0365-9245-9