*Article* **Parental Knowledge, Attitude, and Practices on Antibiotic Use for Childhood Upper Respiratory Tract Infections during COVID-19 Pandemic in Greece**

**Maria-Eirini Oikonomou † , Despoina Gkentzi \*,†, Ageliki Karatza, Sotirios Fouzas, Aggeliki Vervenioti and Gabriel Dimitriou**

> Department of Paediatrics, Medical School, University of Patras, Rion 26504, Greece; mareiroik@yahoo.gr (M.-E.O.); akaratza@hotmail.com (A.K.); sfouzas@upatras.gr (S.F.); aggelikivervenioti@gmail.com (A.V.); gdim@upatras.gr (G.D.)

**\*** Correspondence: gkentzid@upatras.gr or gkentzid@gmail.com

† These authors contributed equally to this work.

**Citation:** Oikonomou, M.-E.; Gkentzi, D.; Karatza, A.; Fouzas, S.; Vervenioti, A.; Dimitriou, G. Parental Knowledge, Attitude, and Practices on Antibiotic Use for Childhood Upper Respiratory Tract Infections during COVID-19 Pandemic in Greece. *Antibiotics* **2021**, *10*, 802. https://doi.org/10.3390/ antibiotics10070802

Academic Editor: Albert Figueras

Received: 4 June 2021 Accepted: 29 June 2021 Published: 1 July 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/).

**Abstract:** This cross-sectional study aims to assess parents' knowledge, attitude, and practices on antibiotic use for children with URTIs symptoms in Greece in the era of the COVID-19 pandemic. We distributed a questionnaire to a random sample of parents who visited primary health care centers in Patras, Greece. Out of 412 participants, 86% believed that most infections with common cold or flu symptoms were caused by viruses, although 26.9% believed that antibiotics may prevent complications. Earache was the most common symptom for which antibiotics were needed. Most of them (69%) declare being considerably anxious about their children's health during the COVID-19 pandemic. The majority (85%) knew that COVID-19 was of viral origin, yet half of them declared uncertain whether antibiotics were needed. All demographic characteristics, except for gender, were found to have a significant effect on parents' knowledge, attitude, and practices on antibiotic use for URTIs and COVID-19. Factor analysis revealed six groups of parents with common characteristics associated with misuse of antibiotics. Our findings highlight the need to decrease misconceptions regarding antibiotic use by providing relevant education for parents targeting particular characteristics, especially during the COVID-19 pandemic. Continuous education of healthcare providers in the field is also of paramount importance.

**Keywords:** upper respiratory tract infections; antibiotics; antimicrobial resistance; parental knowledge; attitudes; practices; child; Greece; COVID-19

#### **1. Introduction**

Upper respiratory tract infections (URTIs) in children account for more than 10% of the total outpatient and emergency department visits and are considered the main cause of children's absenteeism from school as well as parents' from the workplace [1,2]. The discovery of antibiotics is a very important achievement of the 20th century, and their proper and judicious use reduces mortality and morbidity [3]. However, antibiotics are often used inappropriately, especially for the treatment of URTIs in children, even though there is sufficient evidence to support the viral origin of these conditions and the fact that this practice does not change the duration and severity of symptoms [4,5]. Many issues have been raised because of overuse or misuse of antibiotics, such as the high cost of health services, rise of side effects such as diarrhea, and increased antibacterial resistance [6–8]. There is a strong association between antibiotic overuse and the development of resistance, and countries with high rates of antibiotic consumption have reported a high incidence of resistant pathogens [9,10]. As a result, over the past decade, the World Health Organization (WHO) recognized the emergence of antibiotic resistance as one of the 10 top global public health threats facing humanity [11]. Based on numerous reports, factors leading

to antimicrobial overuse in children are complex, involving easy access to antibiotics for self-medication [12], overprescribing by physicians due to diagnostic uncertainty [13], or parents' wish, which is mainly driven by lack of knowledge and misperceptions [14–17]. A recent review demonstrated that one-third of the population of low and middle-income countries have significant knowledge gaps in the field [18]. The quality of communication between parents and pediatricians plays an important role in the decision of prescribing [19]. Associations between demographic characteristics and parental knowledge, attitude, and practices towards antibiotic use in URTIs have also been described [13,20–24].

Coronavirus infectious disease 2019 (COVID-19), declared as a pandemic on the 11th of March of 2020, may cause URTI symptoms and, in the majority of children, warrants only symptomatic treatment [25–27]. No studies on parents' knowledge, attitude, and practices on antibiotic use have been thus far conducted during the pandemic. Despite numerous national educational activities in Greece on judicious antibiotic use, antibiotic consumption in the community has been found to be the highest amongst 31 European countries [28–30]. A nationwide study conducted in 2010 demonstrated that Greek parents understand the benign course of URTIs, rarely give antibiotics without medical advice, and contribute less than expected to antibiotic misuse [31]. The present study aims to explore parental knowledge, attitude, and practices (KAP) regarding antibiotic use for URTIs and COVID-19 infection in Greece, during the COVID-19 pandemic, with the view of designing educational strategies focusing on those parents who are prone to antibiotic misuse.

#### **2. Results**

#### *2.1. Parents' Demographic Characteristics*

Out of the 499 questionnaires distributed, 412 were returned (response rate 82.5%). Table 1 shows the demographics of our studied population. Of note, the majority of our study participants were women (77.2%), aged between 31–50 years old (88.1%), of Greek nationality (85.7%), and with good self-perceived access to the health care system (65.8%).

#### **Table 1.** Demographic characteristics of the study participants.



**Table 1.** *Cont.*

*2.2. Parental Knowledge on Antibiotic Use in Children with URTIs*

The vast majority of parents (81.6%) knew that infections with signs of cold or flu symptoms were caused by viruses. Most of them acknowledged that antibiotics were not necessary, and (69%) answered that symptomatic treatment was only needed for viral infections. Of note, 48.5% of parents thought that antibiotics were needed for bacterial infections while 50.7% for viral ones. Earache was the most common symptom for which parents thought that antibiotics were needed (25.2%), followed by a runny nose (9.5%). More than half of the parents (52.2%) declared that antibiotics were always needed for their child's tonsillitis and 47.1% for otitis. The majority of parents (73.6%) identified successfully antibiotics between other drugs. Half of them did not believe that antibiotic use may protect them from complications of the common cold or the flu. Most of the parents (80.8%) knew that imprudent use of antibiotics reduces their efficacy and may cause antibiotic resistance, and two-thirds of them were familiar with the fact that antibiotics may have some side effects. Most parents (73.3%) declared to be adequately informed about judicious use of antibiotics, and 91.7% of them declared that an educational program about the proper use of antibiotics would be useful.

#### *2.3. Parental Attitude and Practices towards Antibiotic Use in Children with URTIs*

Nearly half of the parents (48.8%) declared that they did not know whether their child needed an antibiotic before visiting the pediatrician. Most of the responders (69.4%) would not feel more satisfied if their pediatrician were to prescribe antibiotics for their child's cold symptoms. Moreover, 93% of the participants always comply with their pediatrician's recommendations. When asked if they would visit another pediatrician for a second opinion in case their own would not prescribe antibiotics, 82% disagreed. In addition, the majority (86.6%) would not change pediatrician because they would not prescribe antibiotics as frequently as they desired. Most of them never used leftover antibiotics, while 13.6% sometimes did. A total of 84% of the participants never insisted that their pediatrician prescribe antibiotics, and most of them (90.3%) answered that their pediatrician did not prescribe antibiotics because they insisted, while 7.8% declared that this happens frequently. A total of 85% did not purchase antibiotics for their child's cold or flu symptoms over the counter, while 15% of them adopted this practice sometimes, often, or always. Moreover, 61.4% of the parents answered that their pediatrician never recommended antibiotics over the phone, while 32.5% answered that this had happened sometimes. If their child was unwell, parents always (50.5%) or most of the time (27.6%) consulted their pediatrician, followed by previous experience where they referred to sometimes (27.9%) or often (19.7%). Moreover, 23.1% sometimes asked their relatives, while 14.6% sought advice from the internet.

#### *2.4. Parental Knowledge Attitude, and Practices towards Antibiotic Use in Children with COVID-19 Infection*

Most parents (85%) declared that COVID-19 infection was of viral origin, and 72.3% answered that it may cause flu and cold symptoms in children. Most parents (68.4%) declared being anxious for their child's health during the COVID-19 pandemic. Approximately half of the parents (49.8%) declared unaware whether antibiotics were needed for COVID-19, and 47.8% did not know whether their child would be sick for a longer time if they would not receive antibiotics for COVID-19. Half of the parents (51.5%) declared more satisfied if their pediatrician would not recommend antibiotics for their child's cold or flu symptoms. The vast majority of parents (93.2%) answered that their pediatrician did not recommend antibiotics over the phone for their child's cold or flu symptoms during the pandemic.

#### *2.5. Effect of Demographic Characteristics and Knowledge, Practice, and Attitude on Antibiotic Use*

The correlations between sociodemographic characteristics and KAP on antibiotic use during the COVID-19 pandemic are shown in Table 2. Of note, KAP on antibiotic use during COVID-19 was not associated with parental age and gender, chronic disease in the child, and number of children in the family. Various other associations have been demonstrated between KAP and other sociodemographics. The relevant correlations between KAP and URTIs in general (not only for COVID-19) are available as Supplemental Tables S1 and S2.

Table 3 shows the results of factor analysis. We found six groups of parents who misused antibiotics, which in total accounts for 76.6% of the variation of the answers given on KAP in our study (meaning that 76.6% of our study participants belong to at least one of these groups).

*Antibiotics* **2021**, *10*, 802


\* NS: not statistically significant. \*\* For statistically significant associations for each statement, the subgroup of parents who answered incorrectly (I) in a higher percentage are noticed in parentheses.

**Table 2.** Association of demographic characteristics with knowledge/attitude/practice statements on COVID-19.

133


**Table 3.** Factor analysis of parental characteristics associated with antibiotic misuse \*.

\* Exploratory dichotomous factor analysis (promax oblique rotation and least-squares extraction method) of parental characteristics associated with antibiotic misuse. Only parameters with a factor loading >0.300 are shown.

#### **3. Discussion**

This is the first study on knowledge, attitude, and practices from parents in Greece regarding antibiotic use in young children with URTIs in the era of the COVID-19 pandemic. The response rate was 82.4%, which is rather satisfactory, especially taking into account the COVID-19 strict restrictions during visits in primary health care.

According to the results of this study, parents have a satisfactory level of knowledge on antibiotic use for URTIs, however some incorrect perceptions have been revealed. The majority of parents (81.6%) correctly answered that most URTIs are viral in origin, in contrast to parents from a similar study from Northern Ethiopia who scored lower percentages of correct answers (31.4%) [15]. In the latter study, 30% of the responders believed incorrectly that antibiotics were needed for viral infections, and 20.6% declared

to be uncertain. Of note, this percentage is even smaller than other studies in China and Pakistan (79% and 83%, respectively) [32,33]. Almost 70% of parents were aware that common cold and flu need only supportive care, although 26.9% falsely believed that antibiotic use may protect from complications while 26.9% declared uncertain, which are similar percentages to those previously reported [32–39]. A total of 15.6% answered that their children's disease will get complicated in the case of not using antibiotics in contrast to similar surveys in China [32], Cyprus [14], and the Republic of Macedonia [18], in which the percentages were higher, 50%, 48%, and 44%, respectively. In our study, most responders were familiar with antibiotics and able to distinguish them among other drugs similar to Palestinian parents [38]. In particular, 61% were worried about antibiotics' side effects, and the majority knew that imprudent use drives antimicrobial resistance, as previously depicted in similar studies worldwide [8,14,16,23].

The results of this study regarding parents' attitudes and practices towards antibiotic use for children's URTIs are of great interest. Parents do not seek antibiotics as 83% of them do not insist on this practice, and 69% are not more satisfied if this happens. Less than 10% believed that their pediatrician prescribes antibiotics due to their insistence, which is less than Palestinian parents (28%) but more than the percentage of parents from Cyprus [8,14]. Of note, only 8% of Australian parents asked their pediatrician for antibiotics (8%), whereas 82% of Chinese parents mentioned that they would not get displeased if their pediatrician rejected their demand for antibiotics [22,32]. In general, parents expect physical examination and adequate consultation for the nature of the disease and the need for antibiotics [37–40]. These findings emphasize the vital role of communication between parents and pediatricians.

The findings of the survey also demonstrate that Greek parents do not treat their children without medical advice. Only 12.6% of the participants reported that they have sometimes purchased antibiotics without medical prescription by the drug store. This finding is lower than in other countries such as Peru (23.5%) and Lebanon (22.5%) [34–38]. Of note, 16.3% have sometimes used leftover antibiotics in contrast with Saudi Arabia and the Republic of Macedonia, where higher percentages were recorded, 66%, 60.8%, respectively [16,18]. Moreover, it appears that Greek parents trust their pediatricians as the vast majority would neither visit another pediatrician for a second opinion (81.3%) if their pediatrician would not prescribe antibiotics for their child's cold or flu, nor change their pediatrician in case they did not prescribe antibiotics as often as they believed that they should (86.4%). In addition, 93% always comply with their pediatrician recommendations in regards to their child's cold or flu symptoms. These findings are similar to a similar study in Palestine but differ from a study in Jordan where 39.3% of the parents answered that they would indeed change pediatrician if they prescribed antibiotics less frequently than they desired [23,39]. The pediatrician remains the main source of information for the ill child, followed by previous experience, pharmacist, relatives, and the internet. Finally, more than one-third of the parents obtained antibiotics for their child following a telephone consultation, and the same percentages have been reported in Jordan and Cyprus [14,23].

It is of interest to note that we did not find many differences in comparison with a similar survey conducted in our country a decade ago [31]. Earache was the most common symptom for which parents anticipated receiving antibiotics. Whereas in both studies, most parents declared to be aware of antimicrobial resistance due to antibiotic misuse (88% in the previous and 81% in our study), a similar percentage of parents (24.7% in the previous and 24% in our study) would administer antibiotics in case their child got flu or cold symptoms in order to recover sooner. Moreover, similar percentages of parents (42% in the previous and 38% in the current study) got a medical recommendation for antibiotics over the phone. As a result, the development and implementation of parental education measures to improve antibiotic use are crucial and still warranted at a national level.

This is the first study on parental knowledge, attitude, and practices towards antibiotic use on COVID- 19 infection in children in our country. The majority of parents (68.4%) declared to be extremely anxious as far as their children's health is concerned due to the

pandemic. Most parents knew that COVID-19 causes flu or cold symptoms and is of viral origin. However, half of them declared to be uncertain as to whether antibiotics were needed for its treatment, and a third of them were not aware if their child's illness would last for longer if no antibiotics are used. These results were predictable as COVID-19 was a new infection, and parents were not adequately informed, especially in the beginning of the pandemic. Therefore, pediatricians should invest more time in order to inform parents about COVID-19. As far as over the telephone diagnosis and recommendation of antibiotics are concerned, the vast majority of parents (93%) replied that this never happened during the pandemic, whereas 63% replied that this was never practiced by their pediatrician before. Parents expect physical examination and recommendation of the appropriate treatment. Given the circumstances during the pandemic, this might not always be feasible. Healthcare providers should therefore have a low threshold of antibiotic prescribing and continue to adhere to the antimicrobial stewardship principles during the pandemic to combat further development of resistant pathogens.

Factor analysis revealed six groups of parents with common characteristics that were associated with antibiotic misuse. Group 1 that accounts for 24% of the variation includes parents that do not have an adequate background on judicious antibiotic use. Even worryingly, group 2 (10.2% of the variation) includes parents that are confident in antibiotic use yet misuse them. For these two groups, more time needs to be invested by prescribers in educating parents in the field. On the other hand, groups 3, 4, 5, and 6 (31.5% of the variation) are parents that do not use antibiotics prudently but do listen to their pediatricians. The latter highlights the need for interventions on antibiotic prescribers, including continuing education. Educational activities for prescribers should start early on in their career, as research thus far has highlighted gaps in medical training in the field [41,42]. Novel and effective training methods on all aspects of antimicrobial stewardship and antimicrobial resistance should be incorporated in the Medical Curricula worldwide. In addition to education, better prescribing requires expert personnel, practice guideline drafting, and implementation aids, as well as setting clear goals and quantitative targets for quality in prescribing [43]. All the above have been used with apparent success in some Northern European countries and should guide relevant interventions in Greece [43].

It is of interest to highlight that our study population included a small percentage of parents with low cultural and economic levels. In particular, more than half of our study participants were urban with a university education. As the problem of antimicrobial misuse is a social issue, it demands a relevant approach from that perspective [44]. Hence, a good level of knowledge on antibiotics use could be partially related to the high socioeconomic status of our participants. The involvement of social scientists in the field is crucial to address the multiple dimensions of the problem, such as socio-cultural, economic, and political. In addition to the above, given the dynamic and constantly changing nature of human behavior, social media could also play an important role in public health interventions to combat antimicrobial resistance, and their role is yet to be further established [44,45].

We acknowledge that our study as a pragmatic one has strengths and limitations. One of its strengths is that data were collected through self-administered questionnaires. This methodology was preferred more than face-to-face or telephone interviews in order to avoid the chance of the interviewer influencing the parent's response as well as the likelihood that interviewers respond in accordance with what is believed to be the expected answer. With regards to study limitations, firstly, the distribution of the questionnaires took place in the summer months of 2020 (during precautionary measures to prevent the spread of COVID-19) with a low number of primary care visits. Secondly, it was difficult for parents with low socioeconomic status or belonging to special population groups (e.g., Roma) to take part in the study because they might have been incapable of reading or understanding the questionnaire. Moreover, the study enrolled parents who only sought care at health care centers, which limits the generalizability of the results. Finally, data were collected from a large city in Greece and thus mainly represent the urban but not necessarily the rural population of the area.

#### **4. Materials and Methods**

We performed a cross-sectional study including parents of children aged 0–16 years and attending the primary pediatric healthcare services in the city of Patras during a period of 4 months (May–August 2020). Patras is the third-largest city in Greece after Athens and Thessaloniki. All parents/guardians of children attending the primary care clinics were informed in detail with a written information sheet about the study aims, as well as data confidentiality. Parents who agreed to take part in the study were asked to complete an anonymous self-administered questionnaire regarding their KAP on the use of antibiotics for childhood URTIs and COVID-19. The questionnaire was written in Greek and consisted of 2 parts. Part A included 13 questions on demographics (age, gender, nationality, level of education, occupation, health insurance status, single-family status, child's chronic disease, access to the healthcare system). Part B included 23 questions on the parent's knowledge, attitudes, and practices on antibiotic use for URTIs and COVID-19. Answers were given on a 5-point Likert scale ("strongly agree", "agree", "uncertain", "disagree", "strongly disagree", or "never", "sometimes", "often", "most of the time", "always"). Statistics were performed using the SPSS v.21 software (IBM Corp, Armonk, NY, USA). The level of significance was set to 0.05 for all analyses. Descriptive statistics were used to summarize data on demographic characteristics, knowledge, attitude, and practices towards antibiotic use. Chi-square test was used to test for significant association between demographic characteristics and parents' knowledge, attitude, and practices. Unsure responses were scored as incorrect. Factor analysis was used in order to identify groups of parents with common characteristics that were at high risk of inappropriate antibiotic use.

#### **5. Conclusions**

This study is an important step towards a better understanding of the knowledge, attitude, and practices regarding antibiotic use in URTIs in children during the COVID-19 pandemic. It revealed that there was a diversity of parental awareness about antibiotics and antimicrobial resistance on the basis of social-demographic factors. Of note, the pandemic does not seem to have influenced parental views on antibiotic use for their offspring. Educational campaigns targeting high risk groups to decrease misconceptions about antibiotic use and to increase awareness regarding the risks of inappropriate use are warranted. Targeted education on the particularities of COVID-19 is also vital for parents. In addition, continuous training of pediatricians about effective communication with parents and prudent prescribing of antibiotics is essential. Ideally, this study should be conducted with a large country-representative sample and during the consequent phases of the pandemic in order to assess for possible changes. It would also be useful to assess attitudes pre and post-training, specifically designed for that purpose, for both parents and pediatricians. Furthermore, studies assessing country-specific determinants of antibiotic misuse, such as country wealth and health care system particularities, would also be useful for the implementation of multilevel interventional programs aimed at limiting the spread of antibiotic resistance. Despite multiple efforts to achieve that in Greece during the last decades, there is still room for further development. Public health interventions at a national level should be constant and sustainable following the successful examples of other European countries. Finally, antimicrobial stewardship activities and interventions should not be neglected during the COVID-19 pandemic.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/antibiotics10070802/s1, Table S1: Association of demographic characteristics with knowledge statements about URTIs, Table S2: Association of demographic characteristics with attitude/practice statements about URTIs.

**Author Contributions:** Conceptualization, D.G., G.D.; methodology, D.G., G.D.; software D.G., M.-E.O. and G.D.; validation, D.G. and G.D.; formal analysis, S.F. and M.-E.O.; investigation, M.-E.O. and A.V.; resources, M.-E.O. and A.V.; data curation, M.-E.O., A.K., and A.V.; writing—original draft preparation, M.-E.O. and D.G.; writing—review and editing, M.-E.O., D.G., A.K., A.V., S.F., G.D.; visualization, D.G.; supervision, D.G., G.D.; project administration, M.-E.O. and A.V.; funding acquisition N/A. All authors have read and agreed to the published version of the manuscript.

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

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of Research and Ethics Committee of Patras Medical School (decision number 5919/25.05.2020).

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

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

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

#### **References**


### *Article* **Factors Associated with Antimicrobial Stewardship Practices on California Dairies: One Year Post Senate Bill 27**

**Essam M. Abdelfattah 1,2,3 , Pius S. Ekong <sup>1</sup> , Emmanuel Okello 1,3 , Deniece R. Williams <sup>1</sup> , Betsy M. Karle <sup>4</sup> , Terry W. Lehenbauer 1,3 and Sharif S. Aly 1,3,\***


**Abstract:** Background: The current study is aimed at identifying the factors associated with antimicrobial drug (AMD) use and stewardship practices on conventional California (CA) dairies a year after CA Senate Bill 27. Methods: Responses from 113 out of 1282 dairies mailed a questionnaire in 2019 were analyzed to estimate the associations between management practices and six outcomes including producer familiarity with medically important antimicrobial drugs (MIADs), restricted use of MIADs previously available over the counter (OTC), use of alternatives to AMD, changes in on-farm management practices, changes in AMD costs, and animal health status in dairies. Results: Producers who reported having a veterinarian–client–patient relationship (VCPR) and tracking AMD withdrawal intervals had greater odds of being familiar with the MIADs. Producers who began or increased the use of preventive alternatives to AMD in 2019 had higher odds (OR = 3.23, *p* = 0.04) of decreased use of MIADs previously available OTC compared to those who did not. Changes in management practices to prevent disease outbreak and the use of diagnostics to guide treatment were associated with producer-reported improved animal health. In addition, our study identified record keeping (associated with familiarity with MIADs), use of alternatives to AMD (associated with management changes to prevent diseases and decreased AMD costs), and use of diagnostics in treatment decisions (associated with reported better animal health) as factors associated with AMD stewardship. Conclusions: Our survey findings can be incorporated in outreach education materials to promote antimicrobial stewardship practices in dairies.

**Keywords:** antimicrobial drug use; antimicrobial stewardship; dairy cattle; knowledge; logistic regression models; machine learning; survey

### **1. Introduction**

Antimicrobial drugs (AMD) are important compounds used in humans and livestock for the prevention, control, and treatment of bacterial infections. Medically important antimicrobial drugs (MIADs) are AMD considered as essential or otherwise important for therapeutic purposes in humans. In the United States, MIADs are AMDs that are important for treating human disease and includes all critically important, highly important, and important drugs according to the Federal Food and Drug Administration's Guidance for Industry #152 [1]. According to the U.S. Food and Drug Administration [1], for the 2019 calendar year, the cattle industry was involved with 41% of the total sales and distribution of MIADs approved for use in food-producing animals. Likewise, the FDA estimated for

**Citation:** Abdelfattah, E.M.; Ekong, P.S.; Okello, E.; Williams, D.R.; Karle, B.M.; Lehenbauer, T.W.; Aly, S.S. Factors Associated with Antimicrobial Stewardship Practices on California Dairies: One Year Post Senate Bill 27. *Antibiotics* **2022**, *11*, 165. https://doi.org/10.3390/ antibiotics11020165

Academic Editor: Marc Maresca

Received: 16 December 2021 Accepted: 25 January 2022 Published: 27 January 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/).

the 2019 sales and distribution data that 81% of cephalosporins, 65% of sulfas, 45% of aminoglycosides, and 42% of tetracyclines were associated with cattle. However, AMD use comes with the risk of antimicrobial resistance (AMR) in humans and livestock [2–5].

In January 2017, the FDA fully implemented the Veterinary Feed Directive (VFD) final rule requiring approval and oversight by a licensed veterinarian for MIADs administered for therapeutic purposes via animal feed to food-producing animals and eliminated non-therapeutic uses for both growth promotion and feed efficiency [6]. The FDA also concurrently applied the same requirement for MIADs administered in drinking water for food-producing animals. Beyond these national regulations, in January 2018, California (CA) implemented Senate Bill 27 (SB 27; codified as the Food and Agricultural Code, FAC 14400-14408, here onwards referred to as SB 27), becoming the first state to require a veterinary prescription, under a valid veterinarian–client–patient relationship (VCPR), for all other dosage forms (e.g., injectables and boluses) of MIADs used for livestock [7]. The SB 27 removed all MIADs from over the counter (OTC) sales for livestock. In addition, SB 27 mandated the development of antimicrobial stewardship (AMS) guidelines and resources that support the collection of information on livestock management practices, the monitoring of AMD sales, and AMD use to provide relevant information to producers and other stakeholders. Similar legislation was passed in other states to increase AMS through the judicious use of AMD, such as Maryland SB 471 [8] and Oregon SB 920 [9]. The implementation of effective AMS practices is critical to reducing the threat AMR poses to animal and public health [10]. Similarly, worldwide AMD use in food-producing animals is increasingly being regulated with the goal of adopting AMS practices. Such regulations have included the ban on the preventive use of AMD in the feed of food-producing animals in the European Union [11], benchmarking AMD use between farms in the Netherlands [12], documenting AMD use on individual farms through electronic connection with billing in Denmark [13] and setting up national targets for a reduction of AMD use in Belgium [14].

Ekong et al. [15] conducted a survey immediately after the implementation of SB 27 to identify the factors associated with the adoption of AMS practices in CA conventional dairies. Amongst the survey findings were that dairy producers who reported keeping written or computerized animal health protocols, keeping a drug inventory log, being aware that the use of MIADs required a prescription, involving a veterinarian in the AMD treatment duration and determination, and using on-farm diagnostics to guide AMD therapy were associated with good AMS practices. Our hypothesis was that factors that were associated with AMS immediately after the implementation of SB 27 were still important a year later. Our objective was to identify the factors associated with AMS practices in adult cows in conventional dairies one year post implementation of SB 27 in CA and to compare them to those identified immediately after SB 27.

#### **2. Results**

#### *2.1. Descriptive Statistics*

A total of 131 (10.2%) out of 1282 CA licensed grade A conventional and organic dairies responded to our survey. For the current study, AMS was investigated using responses from only conventional dairies, specifically a total of 113 survey responses. Herd demographics of the dairies included in the survey analyses are summarized in Supplementary Materials Table S1. The majority of dairies were from Northern San Joaquin Valley (NSJV; 46.9%) and Greater Southern California (GSCA; 46.0%) compared to Northern California (NCA; 7.1%). The distribution of breeds in respondent dairies was primarily Holstein (56.3%), mixed herds of Holstein and Jersey breeds (35.7%), crossbreed (5.4%), and Jersey (2.4%).

#### *2.2. Logistic Regression Models*

For each of the six logistic regression models, the NCA and NSJV regions were combined due to limited sample size and collectively compared to the GSCA region. With a few exceptions, region, herd size, and breed were not significantly associated with the model outcomes.

#### 2.2.1. Predictors Concerning Familiarity of Dairies with the FDA "MIADs" Term

Among the dairies that completed the survey, a total of 73 respondents (65.8%) reported familiarity with FDA's term 'MIADs', while 38 respondents (34.2%) were not familiar with this term. Table 1 summarizes the final logistic regression model for the association between the survey factors and familiarity with MIADs. Dairy producers who reported tracking both milk and meat withdrawal intervals, amongst other information such as date, dose, or route of AMD treatment, had greater odds (OR = 10.6, *p* < 0.01) of being familiar with MIADs compared to those who did not track either. Among the dairies in our study, 94 dairies (84%) selected milk and/or meat withdrawal intervals as information they tracked during AMD treatment and 18 dairies (16.1%) reported tracking either date, dose, or route of AMD treatment. Responses to the related question of "Do you keep track of AMD withdrawal intervals for treated cows?" confirmed a similarly high response rate with 99% of respondents reporting that they keep track of AMD withdrawal intervals, with more than half (58%) using computer software while the remaining (41%) use either paper records, white/chalk board, memory, or markings on animals. In addition, dairy producers who reported having a VCPR had greater odds (OR = 15.3, *p* = 0.03) of being familiar with MIADs than those who reported they did not have a VCPR.

**Table 1.** Estimated coefficients and odds ratios from a multilevel mixed-effects logistic regression model for the association between survey factors and familiarity with the U.S. Food and Drug Administration's term "medically important antimicrobial drugs".


<sup>1</sup> Northern California (NCA), Northern San Joaquin Valley (NSJV), and Greater Southern California (GSCA). <sup>2</sup> Factors are statistically significant at *<sup>p</sup>* <sup>≤</sup> 0.05.

#### 2.2.2. Predictors Concerning the Use of MIADs That Were Restricted from OTC Sales Beginning in January 2018

In our survey, 81 of the 113 respondent dairies reported use of OTC MIADs before 2018. However, of the 81 respondents, only 78 dairies responded to a survey question regarding the changes (decreased, no change, or increased) they made for use of OTC MIADs after SB 27 became effective. Of those 78 dairies, 37 (47.4%) reported decreased use of MIADs federally labeled for OTC sale, while 41 dairies (52.6%) reported increased or no change of this category of MIADs. The final model for the associations between the survey factors and the change in use of MIADs that were previously available OTC is summarized

in Table S2. Large dairies (≥1304 cows) had six-times greater odds for the decreased use of OTC MIADs compared to small dairies (<1304 cows; *p* < 0.01). In addition, our model showed that producers who began using or increased the use of preventive alternatives to AMD in their dairies in 2019 had higher odds (OR = 3.2, *p* = 0.04) of a decreased use of MIADs that were previously available OTC compared to those who did not. For producers who reported the use of preventive alternatives to AMD in our study, 58.3% reported the use of vaccines, herbal remedies (36.1%), vitamins (66.7%), and/or minerals (38.9%).

An analysis of the survey responses from producers who began or increased their use of preventive alternatives to AMD in 2019 showed that a higher percentage of such producers made changes in management to prevent disease outbreaks or spread compared to those who neither began nor increased their use of such alternatives (44.8% ± 9.4 vs. 23.8% ± 4. 8; *p* = 0.03). Furthermore, a higher percentage of such producers vaccinated their animals against coliform mastitis compared to those who neither began nor increased their use of AMD alternatives (92.6% ± 5.13 vs. 74.3% ± 5.11; *p* = 0.04).

2.2.3. Predictors Concerning the Use of Preventive Alternatives to AMD on Dairy Farms

Among the dairies who completed the survey, a total of 30 respondents (26.8%) reported an increased use of preventive alternatives to AMD in their dairies during 2019. Table 2 summarizes the final model for the survey factors associated with the use or increased use of preventive alternatives to AMD.

**Table 2.** Estimated coefficients and odds ratios from a multilevel mixed-effects logistic regression model for the association between survey factors and use or increased use of preventive alternatives to AMD.


<sup>1</sup> Northern California (NCA), Northern San Joaquin Valley (NSJV), and Greater Southern California (GSCA). <sup>2</sup> Factors are statistically significant at *<sup>p</sup>* <sup>≤</sup> 0.05. <sup>3</sup> Breed was categorized into two levels, namely Holstein and others (Jersey, crossbreed, and mixed herds), due to small sample size.

Our model showed that producers who received training or participated in any dairy quality assurance programs during 2019 had greater odds (OR = 3.6, *p* = 0.05) of using or increasing the use of AMD preventive alternatives compared to those who did not participate in such programs. In addition, our survey revealed that conventional dairies that reported a decreased use of MIADs previously available OTC had higher odds (OR = 6.0, *p* = 0.01) of usage or an increase in the use of AMD preventive alternatives compared to dairies that reported increased or no change in the use of MIADs previously available OTC. 2.2.4. Predictors Concerning Changes in Management Practices to Prevent Spread or Outbreaks of Disease on Dairies

Among the dairies who completed the survey, a total of 32 respondents (29.4% ± 4.4) reported they had made changes in management to prevent disease outbreaks or spread in their dairies, while 77 respondents (70.6% ± 4.4) reported they had made no changes in management. In our study, 22% of respondents reported a change in management in the form of vaccination programs to prevent disease, 4.6% reported a change in management in the form of improved biosecurity (e.g., restricted traffic on operation, better isolation of sick animals, or designated separate equipment for feed and manure handling), 1.8% reported a change in management in the form of prepurchase testing of animals before being added to the herd, and 0.9% reported a change in management in the form of quarantining of purchased/returning animals from offsite locations.

Table S3 summarizes the final model for the survey factors associated with changes in management to prevent outbreaks. Our model showed that producers who included veterinarians in the revision of animal health protocols for cows had greater odds (OR = 13.5, *p* = 0.01) of reporting changes in management practices in their dairies compared to those who did not include veterinarians in protocol reviews. In addition, producers who reported better animal health on their farms during 2019 had greater odds (OR = 4.7, *p* < 0.01) of having made management changes to prevent disease spread or outbreaks compared to those who reported no changes to their animal health. Finally, producers who reported the use or increased use of alternatives to AMD in their dairies, such as vitamins, minerals, and vaccines, to reduce the use of MIADs were at greater odds (OR = 2.7, *p* = 0.04) of having made management changes to prevent disease spread or outbreaks compared to those who did not use any alternatives to AMD.

#### 2.2.5. Predictors Concerning Change in a Dairy's AMD Costs

Changes in a dairy's drug costs were dichotomized as "decreased AMD costs" vs. "increased/no change" in AMD costs. Among the dairies who completed the survey, a total of 29 respondents (26.6% ± 4.3) reported decreased farm AMD costs since January 2018, while 80 respondents (73.4% ± 4.3) reported increased or no change in farm AMD costs. Table S4 summarizes the final model for the survey factors associated with decreased farm AMD costs.

Our model showed that producers who reported the use or increased use of AMD preventive alternatives reported decreased AMD costs in their dairies compared to those who did not use AMD preventive alternatives (OR = 7.89; *p* ≤ 0.01). In addition, our study showed that producers who indicated better animal health in their farms also reported decreased AMD costs in their dairies compared to those who reported no change or worse animal health (OR = 5.48; *p* ≤ 0.01).

#### 2.2.6. Predictors Concerning Change in Reported Farm Animal Health

The response to a survey question regarding the farm animal health status as an outcome was dichotomized as "better animal health" vs. "worse/no change". Among the dairies who completed the survey, a total of 47 respondents (44.3% ± 4.9) reported better animal health since January 2018, while 59 respondents (55.7% ± 4.9) reported worse or no change on the farm's animal health. Table 3 summarizes the final model for the survey factors associated with better animal health in the dairy farms.

Our results showed that producers who included the veterinarian in the decision as to which AMD were used to treat sick cows had lower odds of reporting better animal health in their dairies compared to those who did not include the veterinarian in their treatment decision (OR = 0.31; *p* = 0.01).


**Table 3.** Estimated coefficients and odds ratios from a multilevel mixed-effects logistic regression model for the association between antimicrobial drugs (AMD) survey factors and better farm animal health.

<sup>1</sup> Northern California (NCA), Northern San Joaquin Valley (NSJV), and Greater Southern California (GSCA). <sup>2</sup> Factors are statistically significant at *<sup>p</sup>* <sup>≤</sup> 0.05. <sup>3</sup> Breed was categorized into two levels, namely Holstein and others (Jersey, crossbreed, and mixed herds), due to small sample size.

Producers who indicated using on-farm diagnostic techniques to guide AMD treatment decisions had greater odds of reporting better animal health on their dairies compared to those who did not use diagnostic techniques (OR = 4.53; *p* < 0.01).

In addition, the current model indicated that producers who reported management changes in their dairies to prevent disease spread or outbreaks had greater odds (OR = 2.91, *p* < 0.01) of having better animal health in their farms during 2019 compared to those who reported no changes in management practices. Similarly, producers who reported decreased AMD costs in their dairies had greater odds (OR = 3.68, *p* = 0.01) of having better animal health in their farms during 2019 compared to those reported increased/no change in their AMD costs.

2.2.7. Predicting Factors Associated with Dairy Producers' Perceptions Regarding the Importance of AMS Practices on Dairies Using Machine Learning Classification Models

The distribution of CA dairy producers (*n* = 113) with respect to five statements on the importance of AMS practices is presented in Figure 1. Based on the number of the AMS practices that the dairy producers scored as very important, we found that 41.3%, 37.6%, 19.3%, 0.9%, and 0.9% of producers were given a score of 5, 4, 3, 2, and 1, respectively. By classifying producers as having "good to excellent = score of 4 and 5" AMS knowledge or as having "limited-moderate = score of 3 or less" AMS knowledge, we found that 78.9% (86/109) of producers had "good to excellent" knowledge, while 21.1% (23/109) of producers were classified as having "limited-moderate" knowledge based on their responses.

their responses.

**Figure 1.** Distribution of responses from Californian conventional dairy producers to five statements on importance of antimicrobial stewardship practices. Bars represent proportion of responses by level of importance. The plot summarizes survey responses from 113 conventional dairy respondents. **Figure 1.** Distribution of responses from Californian conventional dairy producers to five statements on importance of antimicrobial stewardship practices. Bars represent proportion of responses by level of importance. The plot summarizes survey responses from 113 conventional dairy respondents.

<sup>1</sup> Northern California (NCA), Northern San Joaquin Valley (NSJV), and Greater Southern California (GSCA). <sup>2</sup> Factors are statistically significant at *p* ≤ 0.05. <sup>3</sup> Breed was categorized into two levels, namely Holstein and others (Jersey, crossbreed, and mixed herds), due to small sample size.

2.2.7. Predicting Factors Associated with Dairy Producers' Perceptions Regarding the Importance of AMS Practices on Dairies Using Machine Learning Classification Models

The distribution of CA dairy producers (*n* = 113) with respect to five statements on the importance of AMS practices is presented in Figure 1. Based on the number of the AMS practices that the dairy producers scored as very important, we found that 41.3%, 37.6%, 19.3%, 0.9%, and 0.9% of producers were given a score of 5, 4, 3, 2, and 1, respectively. By classifying producers as having "good to excellent = score of 4 and 5" AMS knowledge or as having "limited-moderate = score of 3 or less" AMS knowledge, we found that 78.9% (86/109) of producers had "good to excellent" knowledge, while 21.1% (23/109) of producers were classified as having "limited-moderate" knowledge based on

A descriptive analyses of responses based on their knowledge of AMS practices (Table S5) showed that a greater percentage of respondents classified as having good to excellent AMS knowledge kept drug inventory logs compared with respondents classified as having limited–moderate knowledge (81.4% ± 5.9 vs. 18.6% ± 5.9; *p* < 0.01). Similarly, results showed that a greater percentage of respondents classified as having good to excellent AMS practices knowledge were familiar with FDA's term "MIADs" compared with respondents classified as having limited–moderate knowledge (90.1% ± 3.5 vs. 9.85% A descriptive analyses of responses based on their knowledge of AMS practices (Table S5) showed that a greater percentage of respondents classified as having good to excellent AMS knowledge kept drug inventory logs compared with respondents classified as having limited–moderate knowledge (81.4% ± 5.9 vs. 18.6% ± 5.9; *p* < 0.01). Similarly, results showed that a greater percentage of respondents classified as having good to excellent AMS practices knowledge were familiar with FDA's term "MIADs" compared with respondents classified as having limited–moderate knowledge (90.1% ± 3.5 vs. 9.85% ± 3.5; *p* < 0.01).

*Antibiotics* **2022**, *11*, x. https://doi.org/10.3390/xxxxx www.mdpi.com/journal/antibiotics ± 3.5; *p* < 0.01). Table S6 summarizes the performance of three different classification algorithms. The three models had similar specificity; however, the highest sensitivity was achieved by the Table S6 summarizes the performance of three different classification algorithms. The three models had similar specificity; however, the highest sensitivity was achieved by the gradient boosting (GB) model. The three classification models identified herd size and familiarity with MIADs as the top two important predicting variables for classifying dairy producers' AMS knowledge. The decision tree (DT) model showed that a higher percentage of dairy producers who were familiar with the FDA term MIAD were classified as producers with "good-excellent" AMS knowledge compared to producers with "limitedmoderate" AMS knowledge. In addition, the DT model further classified dairies of a herd size >1737 milking cows/herd as dairies with "good-excellent "AMS practices compared with dairies of a herd size ≤1737 cows/herd.

Despite the moderate sensitivity (46.5%), random forest had the greatest precision (88.9%) compared to the remaining ML models. In general, the best classification performance as measured using the AUC was obtained using the GB model (AUC = 96). The top ten GB-identified predicting variables included herd size, familiarity with the FDA term "MIADs", annual RHA for milk production, location of herd in CA, veterinarian input on parenteral AMD purchases, willingness to treat animals with AMD alternatives, use of on-farm diagnostic techniques to guide AMD treatment decisions, producer agreement with the statement that antibiotics use in livestock may cause problems in humans, and the

basis for the mastitis treatment decision (abnormal milk, California Mastitis Test, microbiologic culture of milk samples, drug sensitivity testing or treatment while culture results are pending) and producer agreement that current antibiotic use in livestock will make it harder to treat future infections (Supplementary Materials Figure S1).

#### **3. Discussion**

The response rate obtained in this survey was relatively low; however, our response rate is consistent with other mailed surveys conducted in CA [16,17] and with other surveys of both Canadian [18] and UK dairies [19]. Furthermore, response rates stratified by region in our survey were like the regional distribution of herds as reported in a California Department of Food and Agriculture report [20]. The current survey respondents reported a greater proportion of mixed-breed herds compared to a previous survey, indicating either a different population of respondents or a shift in the state dairy herd breed make-up [21]. Earlier surveys also reported higher estimates for the Holstein herd composition (65% in Aly et al. [22]; 77% in Love et al. [16]; and 66% in Ekong et al. [23]) compared to our survey. Such variability may be due to inherent differences in producers responding to such a wide range of surveys (beef quality assurance, bovine respiratory disease, and antimicrobial stewardship), which could be influenced by their interests.

#### *3.1. Predictors Concerning Familiarity of Dairies with the FDA "MIADs" Term*

Our results showed that dairy producers who reported tracking both milk and meat withdrawal intervals during AMD treatment or those who reported having a VCPR were more familiar with the FDA's term "MIADs" than those who did not report. These findings agree with findings from the survey conducted the year SB 27 was implemented [15]. However, our survey showed that 34.2% of the dairy respondents were not familiar with or not sure how the FDA's term "MIADs" related to their dairies. Similarly, a survey that was conducted to identify the common perceptions of Tennessee cattle producers regarding the VFD [24] found that 13% and 25% of dairy producers were not familiar at all or were slightly familiar with VFD, respectively. Hence, more educational outreach is needed to increase producers' knowledge and familiarity with the judicious use of MIADs in maintaining the health of cattle and reducing the pressure for AMR. The veterinarians can play an important role in communicating and educating farmers on concepts of AMR and the judicious use of AMD at the farm level. The survey conducted in the UK [19] found that the dairy farmers who had greater awareness of AMR were those that had more visits from and contact with their vets.

The majority of respondents (94.6%) in our survey confirmed they had a VCPR, while the remaining respondents (5.4%) indicated they did not; however, the latter also indicated that a veterinarian was involved in AMD treatment decisions, which may refer to the need for outreach on what establishes a valid VCPR. Veterinarians can play a major role in educating dairy producers to ensure that they have knowledge of AMR threat and AMS. Alternatively, producers may gain knowledge of MIADs independent of their herd veterinarian due to their own education and inquisition. Nevertheless, under the recently adopted VFD final rule and SB 27 regulations, veterinarians are expected to work alongside livestock producers as well as feed manufacturers and distributors to assume a greater role and increased responsibility for the use of MIADs. The VFD final rule specifies that the veterinarian must work with their client and assume responsibility for making clinical judgments about animal health; must have sufficient knowledge of the animals by virtue of examinations and/or visits to the farm where the animals are located; and must provide for any necessary follow-up evaluation or care, which are necessary components of a VCPR [6].

#### *3.2. Predictors Concerning the Use of MIADs That Were Restricted from OTC Sales Beginning in January 2018*

On January 2018, SB 27 moved all MIADs that were federally labeled for OTC sale, such as penicillin, oxytetracycline, and tylosin, to prescription status in CA. Our model showed

that herd size and the use of preventive AMD alternatives were important predictors for dairy producers reporting a decreased use of OTC AMD in their dairies. The reported decrease in the use of OTC MIADs by large dairies compared to smaller dairies could be attributed to the differences in management practices, presence of computerized health protocols, and familiarity with MIADs or the use of preventive alternatives to MIADs. Our machine learning models also showed that the majority (84%) of the large dairy producers were classified as producers with "good to excellent" AMS knowledge compared with producers of small dairies. Similarly, the survey by Ekong et al. [15] found that large dairies producers were more familiar with MIADs, reported a greater use of alternatives to AMD, and reported better animal health compared to small dairy producers after implementation of SB 27.

The use of preventive alternatives to AMD may have filled the gap in the decreased use of OTC MIADs; however, it is not known whether the increased use of alternatives preceded or followed the decrease in the use of AMD previously available OTC. Similarly, Ekong et al. [15] found that producers of large dairies and producers who began using AMD alternatives had higher odds for reporting a decreased use of MIADs that were previously available OTC compared to producers of small dairies or producers who did not use alternatives to AMD.

In addition, most producers (53%) who reported a decreased use of OTC MIADs also reported that they strongly agreed that current AMD use practices in animal agriculture will make it harder to treat future livestock infections compared to producers who did not report a decreased use of AMD. Such findings indicate that producers who believe that current AMD use practices will make it harder to treat future livestock infections may have also employed good management strategies that include disease prevention and outbreak investigations, and the use of preventive alternatives, which may result in decreased AMD use on dairy farms. A study that surveyed the New York State dairy veterinarians' perceptions of AMD use [25] reported that veterinarians in their study believed that AMD use could be reduced through improved herd management strategies. However, the veterinarians also stated that the biggest barrier to implementing these changes were financial. Gerber et al. [26] found that dairy producers who implemented preventive management changes for udder and uterine health significantly reduced the use of systemic AMD for both udder and uterine health while maintaining animal health compared to producers that did not implement improvements in herd health management.

#### *3.3. Predictors Concerning the Use of Preventive Alternatives to AMD on Dairy Farms*

Our results showed that dairy producers who reported a decreased use of OTC AMD and reported participation in dairy quality assurance programs were more likely to report the use or increased use of preventive alternatives to AMD in their dairies. Dairy quality assurance programs are voluntary programs that promote quality animal care practices, food safety and quality assurance, as well as enhanced consumer confidence in dairy products (CDRF, 2011). These programs, including CDQAP (California Dairy Quality Assurance Program), the National Dairy FARM (Farmers Assuring Responsible Management) program, cooperative extension outreach education, creamery-led programs, and on-farm training, provide training and standards for quality animal care to promote best management practices for AMS and public health. Most of these programs have specific modules about AMS which provide ongoing education for the dairy community on the judicious and responsible use of MIADs, including avoidance of drug residues in milk and meat.

Similar results were obtained by Ekong et al. [15] who found that producers who reported a decreased use of OTC MIADs had 5.2-times greater odds of having initiated the use or increased use of alternatives to AMD in their dairies compared to producers who made no change or increased use of OTC MIADs. Antimicrobial drug preventive alternatives may have replaced MIADs that were previously available OTC, as evident by the reported decrease in the use of AMD for therapeutic purposes by producers who used or increased the use of alternatives. Alternatively, producers could have increased their use of AMD alternatives due to other reasons including changes in costs of MIADs that may have favored a shift in use of such alternatives. Therefore, further research is needed to verify our findings and determine the success of such preventive measures as alternatives to AMD.

#### *3.4. Predictors Concerning Changes in Management Practices to Prevent the Spread or Outbreak of Disease in Dairies*

Our results showed that dairy producers who reported the inclusion of veterinarians in the revision of health protocols were more likely to report changes in management practices. Herd veterinarians are professionals with experience in animal health and direct knowledge of their herds, and hence are capable of identifying the successful interventions to control and prevent diseases. The survey conducted in New York exploring dairy farmers' attitudes [27] indicated that dairy farmers are more receptive to the opinions of fellow farmers and veterinarians regarding AMD use compared to scientists/researchers and government regulators. A valid VCPR between the dairy producer and herd veterinarian can improve the health and welfare of animals by identifying shared concerns and adopting action plans [28]. For example, Jansen et al. [29] found that the communication between farmers and veterinarians helped in the adoption of practices that reduced mastitis.

Our survey findings also showed that changes in management practices were associated with better animal health. Implementing a disease prevention plan that includes good hygiene, isolation of sick or new animals, and a regular vaccination program are associated with improved animal health and productivity [30]. Good management practices help in reducing disease incidence through inhibiting the proliferation of, exposure, or susceptibility to pathogens [31]. Good management practices for biosecurity focus on efforts to prevent the entry of diseases onto the farm (external biosecurity) as well as to prevent disease transmission within the farm (biocontainment) [32,33]. Ohlson et al. [34] found an association between the lower prevalence of BRD infections with better biosecurity at the herd level. There is strong evidence that high standards of both biocontainment and external biosecurity may lead to improved animal health and, in turn, to a reduction of AMD use [30,35].

Producers who reported making changes in management to prevent disease outbreaks or spread in their dairies also reported an increased usage of alternative preventive measures to reduce the use of AMD. Preventive alternatives may work by improving animal health and hence reducing the need for AMD. However, more research is needed to identify alternatives with the same effectiveness and safety for dairy cattle in comparison to AMD [36].

#### *3.5. Predictors Concerning Change in Dairy's AMD Costs*

Producers reported that better animal health and use or increased use of AMD alternatives were important predictors of decreased AMD costs in their dairies. Increased use of preventive alternatives to AMD in livestock production has been associated with improved animal health and reduction in both AMD use and AMR [31,37]. The use of vaccines and other preventive measures can help minimize the need for AMD by preventing and controlling infectious diseases in animal populations [31]. Several studies have demonstrated that the use of various bacterial as well as viral vaccines in animals can result in a significant reduction in AMD consumption [31,38,39]. Furthermore, improved animal health may be associated with decreased AMD use and consequently reduced AMD cost. Maximizing the management practices that promote animal health and reduce the incidence of diseases may decrease the use of MIADs in dairy cattle, which is an important factor in reducing the pressure for AMR [40]. In agreement with our results, Ekong et al. [15] found that the reported use or increased use of AMD preventive alternatives was a predictor for a reported decrease in farm AMD cost, their study also reported that inclusion of veterinarian in decision to use AMD, decreasing use of MIADs previously available OTC, participation

in dairy quality assurance programs were positively associated with reporting decreased farm AMD cost.

#### *3.6. Predictors Concerning Farm's Animal Health Compared to 2018*

Our findings showed that the use of on-farm diagnostic techniques reduced AMD cost and adopting changes in management to prevent disease outbreak/spread were positively associated with better animal health. In agreement with the findings reported by Ekong et al. [15], our survey producers who reported decreased AMD costs on their dairies had greater odds (OR = 4.57, *p* = 0.01) of having better animal health in their farms during 2019 compared to those who reported increased/no change in their AMD costs. Both surveys also identified the use of on-farm diagnostic techniques to guide AMD treatment and improved management practices to prevent both disease outbreak and spread as predictors of producers who reported better animal health. The use of diagnostic techniques such as laboratory culture, auscultation, and lung ultrasound to guide treatment decisions for cows is an important practice to facilitate the judicious use of AMD [41,42]. The availability of affordable diagnostic tools that can detect animals at early stages of disease is important to prevent the high financial costs derived from lost productivity and the treatment of diseased animals [43]. Testing animals for disease can boost herd health and cut costs associated with AMD treatment [44]. A UK survey of dairy herds reported that both farmers and veterinarians recognized there was substantial room for improvement of current diagnostic tools for the detection of mastitis and metabolic disease in dairy cows, particularly with regard to early disease detection [44].

#### *3.7. Comparing Survey Findings Immediately Post SP 27 (2018) and One Year Later (2019)*

A detailed comparison of the associations explored over the two AMS surveys conducted in CA post SB 27 shows specific shared predictors with similar magnitudes of associations (Figure S2). Predictors common between the current survey (2019) and a year prior (2018) include the reported decreased use of AMD, increased use of AMD alternatives, having a VCPR, tracking AMD treatment information, on-farm record keeping, the use of on-farm diagnostic techniques to guide AMD treatment, participation in dairy quality assurance programs, the inclusion of veterinarians in decisions for the selection of AMD to treat sick cows, and implementation of management practices to prevent disease introduction and spread. In agreement with the current survey (2019), surveys conducted on Illinois dairies [45] and Ohio dairies [46] identified appropriate antimicrobial treatment selection; the use of health protocols; on-farm record keeping; and the application of herd-specific veterinary written protocols, education, and training of farm personnel on diagnostic criteria for the initiation of AMD treatments as important areas for the improvement of AMS in dairies.

#### *3.8. Factors Associated with Dairy Producers' Perceptions Regarding the Importance of AMS Practices on Dairies*

Our main goal for machine learning (ML) classification models was to identify factors that can assign dairy producers with good–excellent AMS knowledge. Machine learning models have been used in human medicine to predict AMS intervention in hospitals [47]; however, their use lacks prediction of AMS in veterinary medicine. Using such tools can be helpful for the prediction of AMS interventions in the dairy industry. Machine learning models can be developed to predict which dairies require a stewardship intervention. However, further work is required to develop models with adequate discriminatory power to be applicable to the real-world dairy industry. Our ML models classified producers of large dairies (>1737 milking cows/herd) as producers with "good-excellent" AMS knowledge compared with producers of small dairies (≤1737 cows/herd). A recent Australian study [48] surveyed livestock veterinary practices and identified a lack of access to education, training, and AMS resources as key barriers to the implementation of AMS practices in veterinary practices. Jones et al. [49] suggested that the scientific advice to convey to dairy

farmers to achieve responsible AMD use would include advice on best farm practice to minimize the risk of disease and data on cost savings that might be obtained from reduced AMD. In agreement with Ekong et al. [15], our classification ML models also showed that the knowledge of dairy producers about AMS and the adoption of AMS programs may be improved using continuing education and outreach specifically to producers of small and medium-sized dairies. Our results identified that herd size, familiarity with MIADs, RHA, the location of the dairy in CA, tracking AMD treatment, having a written computerized protocol, the somatic cell count, vaccination of animals against mastitis, keeping a drug inventory log, and awareness that MIADs require prescription were the common predictors for classifying dairy producer knowledge regarding AMS practices.

The majority of conventional dairy producers in our study (>95%) indicated that the administration of the appropriate AMD dose, the route, good record keeping on AMD treatment, and observing AMD withdrawal periods are very important AMS practices. This agrees with a previous study that has suggested that improvement in dairy farmers' AMD use practices can be achieved by using written treatment protocols [50] and the appropriate use of AMD through antimicrobial testing to guide the AMD treatment [51]. Therefore, more extension and outreach efforts should focus on those components to improve antimicrobial stewardship in California dairies.

#### *3.9. Study Limitations*

The response rate of the current survey was relatively low; however, it is similar to other surveys conducted in California. Furthermore, it is possible that only the producers most interested in AMS responded to this survey, therefore our survey could be subject to different selection bias. As with any survey, our findings are limited by the responses obtained through a questionnaire mailed to CA conventional dairy producers, which may be subject to information bias. However, we piloted the questionnaire using in-person interviews with extension and outreach specialists and veterinarians in CA [21] and used both multiple-point scales and ordinal Likert scales. The actual AMD usage or treatment practices and the health status of the cows on respondent dairies were not measured and hence findings based on related questions should be interpreted with caution. Finally, AMS practices cannot be characterized by the current survey's responses alone. Therefore, further studies are needed to directly measure the associations between AMD use and AMS practices based on management protocols and an evaluation of both health and production records.

#### **4. Materials and Methods**

A survey instrument was developed to collect information on AMD use in adult cows in CA dairies during 2019, one year post full implementation of SB 27. The survey was mailed to 1282 grade A dairies in CA during the period from May to December 2019. A list of all licensed grade A dairies in California in 2017 was obtained from the California Department of Food and Agriculture and served as a sampling frame for our survey. The survey development, descriptive statistics, and cluster analyses of the 2019 survey are described in detail in Abdelfattah et al. [21]. Briefly, the survey questionnaire consisted of 44 questions partitioned into three main sections. The first section included variables about herd demographics including the respondent's role, the county where the dairy was located, the herd's breed(s), number of milking cows, annual rolling herd average (RHA) milk production, and previous month's average bulk tank somatic cell count. The second section included questions about dairy cow health management and AMD use including protocols for dry-off, vaccination, disease prevention and diagnosis, sources of information on AMD, who makes decisions on the AMD purchased and used, whether producers had written or computerized animal health protocols, use of a drug inventory, and the presence of a VCPR in dairies. The third section inquired about the dairy practices including enrollment in the animal welfare audit and/or dairy quality assurance programs, producer's familiarity with MIADs and changes made since 2018 with regard to MIADs

previously available OTC, AMD costs, the use of AMD alternatives, disease prevention, and the herd's health status. An optional comments section was included in the survey to allow respondents to provide feedback about their concern regarding AMD use and AMR in dairies. Multiple-point scales and ordinal Likert scales were used to capture the participant responses to the survey questions.

#### *4.1. Statistical Analyses*

#### 4.1.1. Descriptive Statistics

For the 113 conventional survey responses, proportions and their standard errors were computed for categorical and ordinal variables, while means and standard errors were computed for continuous variables. Data on the location of dairies in CA were categorized into three regions, namely Northern California (NCA), Northern San Joaquin Valley (NSJV), and Greater Southern California (GSCA), based on the distinct differences among the three regions in dairy infrastructure and management practices [16]. For the purposes of model building, milking herd size was categorized as ≤1304 or >1304 milking cows based on the CA mean herd size while RHA was categorized as <10,880 kg/cow or >10,880 based on the CA mean herd milk production [20].

#### 4.1.2. Logistic Regression Models

Six survey questions related to AMD use and AMR in CA dairies were selected and analyzed as outcome (dependent) variables. These outcome questions included:


Logistic regression models were specified for each outcome. Univariate models were used to assess the association between the survey variables described in Abdelfattah et al. [21] and each of the six outcomes of interest. Predictors associated with an outcome of interest at *p* ≤ 0.20 were considered for further modeling. A manual forward model building approach was used while assessing confounding by the breed, herd size, RHA, and region using the method of change in estimates [52]. Previously excluded variables were offered into the model again and retained at *p* ≤ 0.05. All biologically meaningful interaction terms were explored using significance testing. Collinearity of all the potential explanatory variables was checked using the spearman rank correlation statistic. The diagnostics were

performed, and plots of residuals were examined, confirming the goodness of fit of each model. Final model selection and fit were assessed using the Akaike Information Criterion (AIC). The coefficients, odds ratios (OR), and their associated 95% confidence intervals were estimated in the final logistic model for factors statistically significant (*p* ≤ 0.05) with the outcome. All statistical analyses were performed using Stata 15 (Stata Corp, College Station, TX, USA).

#### 4.1.3. Machine Learning Classification Models

Machine learning algorithms were used to predict dairy producers who considered AMS practices as important to preserve the efficacy of AMD and reduce AMR in dairy farms. In this study, three widely used machine learning algorithms were evaluated, including decision tree (DT), random forest (RF), and gradient boosting (GB) algorithms. The responses of 113 conventional dairy producers to survey questions about AMS practices were used as a target for the three predictive models. The question requested respondents to classify the following five AMS practices as very important, somewhat important, or not important: (1) administration of appropriate AMD dose, route, and duration; (2) good record keeping on treatment and treatment dates; (3) having a current VCPR; (4) observing withdrawal periods and drug residue avoidance; and (5) using alternatives to AMD such as vaccines and supplements. The respondents were given a score of one to five based on the number of the AMS practices that they scored as very important. For example, if one respondent indicated that all five previously mentioned AMS practices are very important, they would be given a score of 5 and so on. A score of 5 was ranked "excellent", 4 as "good", 3 as "moderate", and 2 or 1 as "limited" knowledge of AMS practices. Then, dairy respondents were reclassified as having "good to excellent" AMS knowledge based on a score of ≥4 or as having "limited-moderate" knowledge based on a score of ≤3. Having "good to excellent" vs. "limited-moderate" AMS knowledge was the outcome of the three predictive models (target variable). Each model was offered a set of 27 survey factors (predictor variables) that contributed to most of the variability in the survey responses, as identified using multiple factor analysis, as reported by Abdelfattah et al. [21]. The predictor variables included herd demographics (herd size, location, annual rolling herd average milk production, and somatic cell count); good general practices (feed newborn calves colostrum from fresh cows, have a separate calving pen, and vaccinate against different diseases); AMD usage information (sources of information about AMD, inclusion of veterinarian in the decision to purchase and treat cows with AMD, tracking of AMD withdrawal periods, and having a written/computerized health protocol); mastitis management practices (basis for treatment of mastitis with AMD, AMD treatment choice, and class of AMD used to treat mastitis); metritis management practices (basis for treatment of metritis and AMD treatment choice); treatment choice for pneumonia in adult cows (oral or injectable AMD); familiarity of producers with the FDA's term "MIADs" and that MIADs require prescription; and the level of agreement (strongly agreed/agreed, neutral, or strongly disagree/disagree) of dairy producers on the following statements: current antibiotic use practices in animal agriculture will make it harder to treat future livestock infections, antibiotic use in livestock does not cause problems in humans, antibiotic use in livestock leads to bacterial infections in people that are more difficult to treat, any use of antibiotics may result in infections that are more difficult to treat in the future, and willingness to treat animals with alternatives to antibiotics if they were equally effective and comparable in price. The original dataset was partitioned into training and testing data sets using a random split ratio of 70: 30 (training: test). Each model was trained with the training dataset and evaluated by assessing their predictive performance on the testing dataset using Salford Predictive Modeler 8.0 software (https://www.minitab.com/en-us/products/spm/ accessed on 15 July 2021). For the DT algorithm, we used a 10-fold cross-validation method for testing, Gini as the optimization method, and the minimum cost tree as the choice for the best tree [53]. The RF method is based on multiple decision trees: it builds several individual classification trees using a random subsample of the data and then selects the most popular class [54]. For the RF

model, we used the out-of-bag testing method with 500 classification trees. Meanwhile, for the GB model, we used a 10-fold cross-validation method, a tree size of 500, balanced sample weights, and the best model chosen by cross-entropy [54]. Evaluation of model performance was based on accuracy, sensitivity, specificity, F1 score, and the Area Under Curve (AUC) estimated from receiver operator characteristic (ROC) curve analyses [55]. The probability threshold at which a classification was made was initially set at a standard 0.5. Then, the obtained results were explored, and the optimal predictive threshold was determined to select the highest specificity [56].

#### **5. Conclusions**

The majority of survey producers agreed that the administration of the appropriate AMD dose, the route, keeping good records on AMD treatment, and observing AMD withdrawal periods are very important AMS practices. Our results showed that better cow health was positively associated with management changes to prevent disease spread and the use of on-farm diagnostic techniques to guide AMD treatment decisions. In addition, the use or increased use of preventive alternatives to AMD was positively associated with the decreased use of OTC AMD and decreased farm AMD cost. Our findings identified that factors that were associated with antimicrobial stewardship practices in conventional CA dairies immediately after implementation of SB 27 are still important one year later. These factors included a valid veterinarian–client–patient relationship, a reduction in the use of MIADs that were previously available OTC before implementation of SB 27, a decreased farm AMD cost, and the use or increased use of AMD alternatives. In addition, herd size, familiarity with MIADs, location of the dairy in CA, tracking AMD treatment, having a written computerized protocol, keeping drug inventory logs, and awareness that MIADs require prescription were the common predictors between the survey of 2018 and 2019 that identified dairy producers with "good to excellent" antimicrobial stewardship knowledge using machine learning. The findings of our survey should be interpreted with caution due to biases related to surveys.

Findings from this survey may benefit extension outreach efforts by offering education and training on identified areas associated with improved antimicrobial stewardship practices in CA dairies. Future research is needed to study the association between the implementation of antimicrobial stewardship practices and the reduction in the prevalence of AMR in food-producing animals. In addition, further research is needed to identify the barriers that prevent the implementation of the identified components of antimicrobial stewardship practices in CA dairies.

**Supplementary Materials:** The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/antibiotics11020165/s1. Table S1: Summary of herd characteristics from 113 responses received from conventional dairies to a mailed questionnaire on antimicrobial drug (AMD) use in adult cows in California during 2019; Table S2: Estimated coefficients and odds ratios from a multilevel mixed-effects logistic regression model for the association between survey factors and decreased use of antimicrobial drugs (AMD) that were previously available over the counter; Table S3: Estimated coefficients and odds ratios from a multilevel mixed-effects logistic regression model for the association between survey factors and having farm-made changes in management to prevent spread of disease; Table S4: Estimated coefficients and odds ratios from a multilevel mixedeffects logistic regression model for the association between survey factors and decreased AMD cost compared to 2018; Table S5: Summary of descriptive analysis of dairy producers' respondents classified as having "limited-moderate knowledge" or "good-excellent knowledge" in antimicrobial stewardship practices based on responses obtained from 113 conventional California dairies; Table S6: Performance metrics of three classification models used to study the association between good– excellent antimicrobial stewardship practices and survey responses in 113 conventional California dairies during 2019; Figure S1: Predictive variables importance plot showing the ranking of variables' relative importance for predicting dairies with good to excellent antimicrobial stewardship practices based on survey responses from 113 conventional California dairies during 2019. "MIADs" are medically important antimicrobial drugs. "AMD" is antimicrobial drug; and Figure S2: Web diagram

showing the magnitude and direction of the associations between six antimicrobial stewardship outcomes and predictors from logistic regression models based on surveys of conventional California dairies during 2018 (black arrows) and 2019 (red arrows). Associations with both black and red arrows identify common associations between surveys.

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

**Funding:** This research study was funded by the California Department of Food and Agriculture, the University of California Davis's School of Veterinary Medicine, and the Office of Research's Principal Investigator Bridge Program. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

**Institutional Review Board Statement:** The study was reviewed by the University of California, Davis Institutional Review Board, and was granted exemption approval as our research did not involve human subjects (IRB number: 1537295-1; approval date: 9 January 2020).

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** This study was sponsored by the California Department of Food and Agriculture, and is subject to California Food and Agriculture Code (FAC) Sections 14400 to 14408. FAC Section 14407 requires that the data collected be kept confidential to prevent individual identification of a farm or business; as such, raw data from this study are not able to be shared.

**Acknowledgments:** The authors are grateful for survey producers, herd managers, and veterinarians.

**Conflicts of Interest:** The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


#### *Review*
