The Effect of Food Handler Certification on Food Premises in Ontario,Canada

Introduction

Foodborne illness constitutes a significant public health problem for Canada. Recent burden of foodborne illness estimates indicate about 4 million cases occur in Canada every year (Thomas et al., 2015), with associated cost estimates for gastrointestinal illness varying considerably, depending on costing methodology, from ∼$3.7 billion to $14 billion dollars annually (Keener et al., 2014; Young et al., 2015).

Food premises have been implicated in a significant proportion of foodborne disease outbreaks and sporadic gastrointestinal infections (Kassa et al., 2001; Angulo et al., 2006; Insulander et al., 2008; U.S. CDC, 2017). In Ontario, Canada, food premises were estimated to account for one-third of all gastrointestinal illness cases (Vrbova et al., 2012). It is for this reason that local public health authorities in Ontario, that is, health units, address these risks primarily through risk-based inspections by qualified environmental health professionals and provision of food handler training certification to operators of food premises (Campbell et al., 1998; Thompson et al., 2005; OMHLTC, 2018). Before 2018, several health units enacted municipal by-laws to enforce mandatory food handler training within their areas of jurisdiction in the absence of a province-wide legal requirement.

The efficacy of food handler training has been studied using several approaches. This includes measuring food safety knowledge and attitudes before and after training (Park et al., 2010; Soon et al., 2012; Garcia et al., 2013; Da Cunha et al., 2014; Nik Husain et al., 2016); comparing inspection scores of establishments that are chain versus independently owned (Kassa et al., 2010; Murphy et al., 2011), or where staff did or did not complete food handler training (Mathias et al., 1995; Jenkins-McLean et al., 2004; Thompson et al., 2005; Averett et al., 2011); and investigating the effect of mandatory training for all food handlers of establishments compared to training provided solely to managers (Cotterchio et al., 1998; Pilling et al., 2008).

However, it remains unclear whether knowledge attained is sufficient to translate into long-lasting positive behavior changes and subsequently to improved food safety outcomes detectable at the food premise level (Da Cunha et al., 2015; de Sousa Carvalho Rossi et al., 2017). In addition, previous studies have not measured the impact of public health inspector-level (PHIs) differences on food safety outcomes. Consequently, evidence on the effectiveness of food handler training to improve inspection outcomes to date remains inconclusive.

Scarcity of resources, opportunity cost, necessity for governments to deliver health programs more efficiently, and 2018 Ontario legislation amendments making food handler training mandatory provide further reasons for examining the effectiveness of food handler training to improve premise food safety and ultimately reduce the burden of foodborne illness.

The primary object of this study was to investigate the effect of food handler training programs on food safety inspections of premises located in a Southern Ontario health unit. The second objective was to investigate the influence of inspectors on food safety inspection results by looking at percentage variability that is explained at the establishment and inspector levels.

Materials and Methods

We extracted data from the health unit's database containing information on food premise inspections, premise type, location, and risk category (high, medium, or low). Twelve full-time PHIs each assigned to geographic areas in the health unit conducted inspections in the 2017 calendar year. The dataset captured information on premise conditions summarized into six food safety inspection domains as shown in Supplementary Table S1 and in accordance with provincial legislation. The health unit's information system used these domain variables to generate a sanitation index score, which also accounted for severity and frequency of non-compliance items noted at the end of each inspection.

For the purposes of this study, food premises were considered to have failed the inspection if the health unit granted yellow signage, indicating premises had failed to comply with regulatory requirements, or when premises were issued a closure notice indicative of an imminent health hazard (Supplementary Fig. S1). PHIs verified the presence of at least one or absence of a certified food handler in the facility at the time of inspection without knowing the study's hypothesis or purpose.

Determinants of premise risk category in order of importance included: the population served by the establishment, complexity of food preparation, volume and types of foods served, historical compliance record, foodborne illness/outbreak history, existence of a food safety management system, and food handler training certification status of persons employed by the premise (OMHLTC, 2015). Only food premises categorized as medium or high-risk were included in the study since these usually require preparation of foods that can support the growth of disease-causing organisms also known as potentially-hazardous foods.

We classified food premises as institutional if its primary business was care provision of high-risk populations (e.g., children, the elderly); and routine preparation of potentially-hazardous foods. Food premises were further classified based on ownership status (e.g., independently-owned, chain), predominant type of cuisine served, and degree of rurality. The original dataset only provided establishment names and postal codes. As a result, we ascertained these factors by web-searching locations, and reviewing establishment menus found on each facility's website. Supplementary Table S2 depicts the cuisine categories identified. Establishments were screened manually through identification of those belonging to the Top 250 chain restaurant rankings in 2018, published by Restaurant Business and Technomic Inc. Premises identified to be part of a franchise network consisting of four or more locations across Canada and/or the United States was considered a chain establishment.

We used the Rurality Index for Ontario, published by the Ontario Medical Association (OMA) to assess a premise's degree of urbanization; classifying locations as either highly urbanized or mixed-urban environments (Kralj, 2000, 2008).

Results

The dataset contained a total of 1795 observations from 892 food premises inspected by 12 PHIs. All variables investigated were either dichotomous or categorical in nature. Most predictor variables displayed substantial variability and had few missing values with the exception of cooking temperature and food contamination food safety domain variables. As a result, these were removed from further analyses (Supplementary Table S3). The prevalence of inspection failure for the study was 3.62% (95% confidence interval: 2.81–4.59) with a total of 65 failed inspections. Correlation analyses between all pairwise dichotomous and categorical predictors indicated no signs of multicollinearity. Ten observations were removed due to ignorable missing attributes. Consequently, 1785 observations were retained for statistical modeling. Univariable analyses between inspection outcome and independent variables using a liberal p-value of 0.2 resulted in handwashing (LRT p ≤ 0.2364) and chain (p ≤ 0.2925) being the only non-statistically significant independent variables (Supplementary Table S4).

A fixed effects logistic regression multivariable model was built to assess confounding and interaction effects. We excluded three variables (handwashing, degree of urbanization, and institutional operations) from final multivariable analyses since they were not confounders or part of a significant interaction term. The final multivariable model chosen to fit the data contained only statistically significant explanatory variables as depicted in Table 1. The LRT performed on premise cuisine type indicated that the categories for this variable were significantly different from each other (p ≤ 0.0047). The final mixed effects multivariable logistic regression model with inspector as a random intercept produced similar results as the fixed effects nested model with all independent variables remaining statistically significant (Table 2). Inspector was a significant random effect but only explained 2.1% of the variation in the data (Supplementary Table S5). In addition, the final mixed effects logistic regression model had 95% confidence intervals that were noticeably wider; with larger standard errors for all independent variables when compared to the fixed effects model with the exceptions being premise risk level and food storage. The final fixed effects model produced a smaller BIC value, indicating that it had comparatively better fit and parsimony.

The final fixed effects model fitted the data based on the Hosmer-Lemeshow GOF test result (p ≤ 0.2557) (Supplementary Table S6) and was adequately capable of predicting inspection failure (Supplementary Fig. S2). Removal of influential or poorly fitting covariate patterns from the model did not produce a change in the significance, direction, or the overall interpretation of the coefficients in the model. BLUPs of the final mixed effects logistic regression model, with inspector as a random effect, were noted to have constant variance, but were not normally distributed (Supplementary Fig. S3, S4, S5, S6) (Supplementary Fig. S7), indicative that model assumptions were not met. Repeated measures had minimal impact on other multivariable model results (Supplementary Table S7).

Discussion

Our final logistic regression models are unequivocal in the estimation of the effects of food handler certification on inspection results. Premises without certified food handlers at the time of inspection were significantly more likely to fail when compared to premises that had a certified food handler on-site (Tables 1 and 2). This finding is congruent with that of several studies (Campbell et al., 1998; Viedma et al., 2000; Thompson et al., 2005), and coherent with the understanding that increasing a food handler's food safety knowledge is an important factor in preventing the onset of a series of interconnected events; beginning with food safety lapses, failed inspections, and ultimately foodborne illness (Chapman et al., 2010; Burke et al., 2014; Da Cunha et al., 2014; Andrew et al., 2017).

Some of these lapses relate to the food safety inspection domain variables included in the study such as inadequate holding temperature of potentially hazardous foods (Panchal et al., 2013), improper storage of foods (Todd et al., 2010), and insufficient premise protection against pests, perhaps because untrained food handlers may underestimate the importance of these factors in shaping the overall food safety performance of their operation.

Similarly, it is unsurprising that our final models estimate that the odds of a food premise failing an inspection is dramatically increased when inadequate food temperatures, improper food storage, or lack of premise protection against pests was noted during inspections; the latter possessing the greatest strength of association with substandard inspection findings. Our results suggest food handler training programs with a particular focus on these three domain variables have potential to significantly improve premise food safety. This includes emphasizing the importance of keeping foods that can support the growth of pathogenic microorganisms away from the time/temperature danger zone of between 4°C and 60°C for over 2 h; proper cooling of foods given time and temperature considerations (FPTFSC Secretariat, 2016) and proper food storage practices to protect foods from potential sources of contamination.

The final mixed effects multivariable model indicated that inspector was a significant random effect (p ≤ 0.0001), but at 2.1%, it only explained a very small amount of variation in the data. This random effect could represent unmeasured predictor variables that influence the prevalence of inspection failure in the study population. Since inspectors are assigned to geographic areas in the health unit, this could include predictor variables associated with premise location. A possible explanation for the small variability explained at the inspector level is that the study's health unit implemented a comprehensive quality assurance program that trains and assesses PHIs to improve consistency of inspection methodologies.

A premise's risk level is another important predictor of food safety performance according to our final models. Premises classified as medium risk had significantly lower odds of failing an inspection compared to high-risk food premises (Tables 1 and 2); possibly because PHIs may hold high-risk food premises to a higher standard than lower risk establishments. The inherent nature of medium as opposed to high-risk operations is another explanation for the observed effect; with the latter being more complex in terms of volume and extent of food preparation steps. This is further evidence of the application of food premise risk assessments as a way for health units to prioritize resources both in terms of stipulating inspection frequencies and for targeting food handler training certification as an intervention.

The odds of inspection failure were also significantly different depending on the cuisine type category of a premise with those classified as South Asian having the greatest odds of inspection failure when compared to premises serving unspecialized foods (Tables 1 and 2). There are various possible reasons for this, including the possibility of inspector differences in risk perception when unfamiliar with a particular type of cultural cuisine during an inspection.

Another explanation is that establishments with cultural cuisines may have a greater proportion of food handlers who are recent immigrants to Canada from countries where food safety regulations and cultural practices differ considerably from that of North America. Studies have shown that ethnic and cultural groups may have unique food safety knowledge and practice needs (Henley et al., 2012; Liu et al., 2013), and this may be an intervening factor in the relationship between premises with cultural cuisines and increased food safety lapses (Kwon et al., 2010; Harris et al., 2015). Food handler certification classes offered by the study's health unit are taught only in English and this language barrier adds another layer of complexity in transferring food safety knowledge. Consequently, the provision of culturally relevant food handler training education is supported by the results of this study. This includes ensuring the availability of food handler training courses in a variety of languages.

Further to this, previous studies have found chain establishments have significantly fewer critical infractions than independent ones (Kassa et al., 2001; Murphy et al., 2011). Our final multivariable models detected the opposite relationship, with independently-owned establishments having lower odds of inspection failure when compared to chain operations. Although not statistically significant, univariable analysis of this effect in our study (Supplementary Table S4) agreed with the findings of previous studies, perhaps because in univariable analysis the effect of unmeasured predictor variables is not taken into consideration. In our study, the inclusion of other important predictor variables in the final multivariable models revealed a distorting effect between the relationship of chain establishment and inspection outcomes.

Future studies should further examine factors pertaining to independent premise ownership that could explain this relationship. This may include measuring the effect of length of business ownership, behavioral factors associated with pride of ownership (McIntyre et al., 2013), or the relative magnitude of chain operations (i.e., multinationals, domestic chain, or independent) that may impact a premise's food safety performance; further evidence of the multifactorial nature of causes that lead to substandard inspection results.

Other possibilities for future research include investigating the number of certified food handlers in a premise that result in optimized food safety performance at the time of inspection. Future studies could attempt to measure other explanatory variables at the inspector level that may explain prevalence variability in inspection findings. This could include premise location, measuring years of professional environmental health experience, professional development, food safety inspection methodology, or inspector differences in food safety risk perception.

Several limitations restricted our ability to make inferences beyond that of the source population of this study. One of them is the possibility of non-differential misclassification of premise level risk factors, including ascertainment of chain versus independent establishments. For example, our study did not account for ownership control differences between corporate-owned and franchisee type operations. It is possible that owners of franchisee operations with substantial control over the establishment function much like an independently owned one. The impact of this type classification error depends on their magnitude and the actual prevalence of classification (Dohoo et al., 2012). It is usually toward the null with a few exceptions, including in cases where sensitivity and specificity of exposure and outcome average less than 50% (Dosemeci et al., 1990).

Further to this, our study analyzed data obtained from 12 PHIs, working in a primarily urban health unit in a single calendar year. As a result, the study data may not be representative of all food premises in Ontario. Likewise, due to the small number of random effects at the inspector level (i.e., only 12 PHIs), it was difficult for the model to estimate BLUPs along the normality curve; which explains the reasons why BLUPs for the final mixed effects model lacked normality. Future studies should also consider this factor if the intent is to measure inspector variation as a random effect.

Conclusion

Our study provides supporting evidence on the effectiveness of food handler training certification as a public health policy; and that increseasing the prevalence of establishments with certified food handlers can improve overall premise food safety. Ensuring that food handler training programs are readily available to food handlers is one way health units can promote this, but a multipronged policy approach in addressing this complex issue would also include the use of regulation, exhortation, and authority as policy instruments. A move toward Ontario standardization of food handler training requirements for food premise operators is a positive change. Nevertheless, further efforts should ensure that training programs are accessible and relatable to premise operators; particularly those serving all types of cultural cuisines.