Multinomial regression methodology is rarely used in foodborne disease investigations, and to our knowledge it was used only once previously in a U.S. case–control study where researchers compared various SE PT infections to healthy controls
. To the best of our knowledge, our study is the first study worldwide to use multinomial logistic regression and the case-case study approach to evaluate associations among SE infections with different PTs and various risk factors. Moreover, our study used a three level random intercept logistic regression model with the GLLAMM procedure. The advantage of the GLLAMM model is that it adjusts for the variation in risk factors for SE PT infections across exposure periods and health regions and it allows for the measurement of unexplained variation in SE PT infection risk factors at these levels. From the total unexplained variation, a moderate amount (10.3%) resided at the exposure period level indicating seasonal clustering of SE PT infection risk factors. Only a small proportion (0.5%) of the total variation resided at the health region level indicating no spatial clustering of cases. This may be explained by the possibility that risk factor variables were evenly distributed across the study area.
In the multivariable model, statistically significant associations were demonstrated between contact with dogs during the three days before disease onset and infection with SE PT8 when compared to the non-PT8/non-PT13a groups, while accounting for age groups and gender, and using exposure periods and health regions as random effects to account for clustering. Several studies have previously demonstrated associations between human salmonellosis and animal contact. Examples include contact with reptiles
, and pet birds
. Moreover, it is well documented that owning a dog can be a risk factor for acquiring salmonellosis. Behravesh et al. demonstrated that feeding a dog in the household with dry food was a risk factor for Salmonella Schwarzengrund infections
. Several studies showed that feeding raw foods to dogs increased their likelihood of Salmonella shedding and consequently the risk of infecting the environment and possibly humans
[24–28]. Findings from our study are consistent with the studies above. It should be noted, however, that these studies used different analytical methods and Salmonella Enteritidis was not among the identified Salmonella serotypes. More research is required to understand SE disease transmission from dogs to humans, and more specifically, to confirm whether there is a relation between SE PT8 infection and exposure to dogs in order to assist public health authorities with designing SE PT - specific prevention and control programs.
Consumption of peppers was significantly associated as having a protective effect for SE PT8 cases when compared to the non-PT8/non-PT13a cases. This result could also indicate that consuming peppers is a risk factor for SE infections with the non-PT8/non-PT13a cases. Historically foodborne disease outbreaks including outbreaks caused by various Salmonella serotypes have been most frequently caused by consumption of foods of animal origin, but more recently an increased number of foodborne outbreaks have been associated with consumption of fresh fruits and vegetables
[29–31]. In a U.S. multistate outbreak investigation, public health authorities identified a link between raw jalapeño pepper consumption and Salmonella Saintpaul infections
In the univariable unconditional model, the odds ratios for frozen chicken consumption and PT8 and PT13a infections compared to non PT8/non PT13a infections were relatively high (4.60 and 5.75, respectively) however not statistically significant. We believe the lack of statistical significance is likely due to a lack of power in the study or that chicken consumption was spread equally among the PT comparison groups. Our finding is not surprising knowing that PT13a and PT8 are among the most frequently SE PTs identified from retail chicken through the Canadian integrated surveillance systems
. Associations between chicken consumption and SE PT8 and SE PT13a infections have also been demonstrated in studies from the U.S.
[11, 19, 33].
Our study only identified two statistically significant risk factors at the multivariable level and these risk factors were both associated with SE infections caused by PT8. The inability to identify other statistically significant risk factors could be explained by these risk factors being distributed evenly across the PT comparison groups (i.e., PT8, PT13a, and non-PT8/non-PT13a). If this was the case, the case-case study design would not be able to demonstrate differences among the given risk factors and the various PT comparison groups. It is also possible that the two statistically significant findings occurred by chance alone, or there were confounder variables that we did not include in our analysis.
There are several advantages to using the multinomial case-case technique. One clear advantage is that for one particular pathogen (e.g., SE), the laboratory sub-type of interest (e.g., PT8) can be compared to other sub-types (e.g., PT 13a, non PT8/non PT13a) within the same model. Consequently, the model can analyze more than two outcomes simultaneously. A second advantage is that the methodology uses case-case comparisons that allow inferences to be made using only case exposure data. Thus, the effort required to collect control data is not required and, therefore, less time and fewer resources are required to complete the study. This could be particularly applicable in food borne outbreak investigations. If a food item is causing an above expected amount of illnesses of a frequently reported pathogen subtype (i.e., a PT or a pulsed-field gel electrophoresis pattern), the subtype could be compared to the “other” subtypes in a relatively efficient manner, assuming the food item is causing illness in expected amounts in the “other” subtypes and there is an adequate number of cases in the other subtype groupings to provide sufficient power. This method has been previously used efficiently in the U.S. during multistate listeriosis outbreak investigations
A third advantage of case-case methodology is that the study design reduces the selection and recall bias usually experienced in case–control studies.
There are several disadvantages in using a case-case multinomial methodology. One disadvantage is that the pathogen subtyping comparison groups (e.g., SE PT8 vs. the non-PT8s) may have similar risk factors
. If this is the case, the methodology would fail to identify the risk factor as associated with either PT group. As noted above, in our study the risk factors for SE may have been evenly distributed among the three PT groups and thus significant differences could not be detected. It is important to note that the case-case approach tested the difference between three SE PT groups as opposed to testing the difference between cases infected with SE and those not infected, which is assessed in a traditional case–control design. Therefore, different interpretations may occur for the same risk factor depending on which study design is being employed. Case-case analysis may only demonstrate that infections with certain PTs are associated with certain risk factors compared with other risk factors. Thus, one must carefully interpret the results of the two different study designs. If the case-case methodology is used more frequently, a greater understanding of the advantages and disadvantages in regard to interpreting the outcome of each methodology will be required.