Evaluating arealevel spatial clustering of SalmonellaEnteritidis infections and their socioeconomic determinants in the greater Toronto area, Ontario, Canada (2007 – 2009): a retrospective populationbased ecological study
 Csaba Varga^{1, 2}Email author,
 David L Pearl^{1},
 Scott A McEwen^{1},
 Jan M Sargeant^{1, 3},
 Frank Pollari^{4} and
 Michele T Guerin^{1}
DOI: 10.1186/14712458131078
© Varga et al.; licensee BioMed Central Ltd. 2013
Received: 4 September 2013
Accepted: 13 November 2013
Published: 15 November 2013
Abstract
Background
There have been only a few regionlevel ecological studies conducted in Canada investigating enteric infections in humans. Our study objectives were to 1) assess the spatial clustering of Salmonella enterica serotype Enteritidis (S. Enteritidis) human infections in the Greater Toronto Area, and 2) identify underlying arealevel associations between S. Enteritidis infection rates and socioeconomic status (SES) indicators that might explain the clustering of infections.
Methods
Retrospective data on S. Enteritidis infections from 2007 to 2009 were obtained from Ontario’s reportable disease surveillance database and were grouped at the forward sortation area (FSA)  level. A spatial scan statistic was employed to identify FSAlevel spatial clusters of high infection rates. Negative binomial regression was used to identify FSAlevel associations between S. Enteritidis infection rates and SES indicators obtained from the 2006 Census of Canada. Global Moran’s I statistic was used to evaluate the final model for residual spatial clustering.
Results
A spatial cluster that included nine neighbouring FSAs was identified in downtown Toronto. A significant positive curvilinear relationship was observed between S. Enteritidis infection rates and FSAlevel average number of children at home per census family. Areas with high and areas with low average median family income had higher infection rates than FSAs with medium average median family income. Areas with a high proportion of visible minority population had lower infection rates than FSAs with a medium proportion of visible minority population. The Moran’s I statistic was not significant, indicating that no residual spatial autocorrelation was present after accounting for the SES variables in the final model.
Conclusions
Our study demonstrated that FSAs with high and low average median family income, medium proportion of visible minority population, and high average number of children at home per census family had the highest S. Enteritidis infection rates. These areas should be targeted when designing disease control and prevention programs. Future studies are needed in areas with high S. Enteritidis infection rates to identify sources of environmental contamination of the local food supply, to assess food safety practices at local food markets, retail stores, and restaurants, and to identify novel individuallevel risk factors.
Keywords
Salmonella Enteritidis Socioeconomic status Spatial scan statistic GIS Choropleth map Negative binomial regression Moran’s I Ecological studyBackground
Nontyphoidal salmonellosis (NTS) is a major foodborne zoonotic infection that poses a significant public health risk [1–5]; on a global basis, it affects an estimated 93.8 million people and causes 155,000 deaths annually [1]. In Canada, NTS is the second most frequently reported enteric infection [6], and results in the largest number of hospitalizations and deaths among foodborne diseases [7]. Surveillance systems in Canada have identified an increasing trend of Salmonella enterica serovar Enteritidis (S. Enteritidis) infections in humans, with a threefold increase between 2003 and 2009 [8]. As a result of this increase, S. Enteritidis became the most common Salmonella serotype in Ontario and Canada.
There are an abundance of case–control studies worldwide that have evaluated individuallevel associations between S. Enteritidis infections and potential risk factors including: consumption of chicken [9, 10], consumption of raw or undercooked eggs [9, 11, 12], persontoperson transmission through infected food handlers [13–15], animaltoperson transmission [16, 17], and international travel [9, 18–20]. In contrast, there are a limited number of populationbased ecological studies that have evaluated arealevel associations between enteric infections and socioeconomic status (SES) indicators. Ecological and individuallevel studies from Canada and Europe [21–24] have demonstrated associations between enteric infections and SES determinants (e.g. household income, education level, unemployment rate, number of children per household, cultural group, population density). Two ecological studies from the United States of America (US) analyzed associations between salmonellosis and arealevel SES and sociodemographic factors. Younus et al. [25] found that areas with higher education attainment had a greater incidence of salmonellosis compared to areas with a lower education attainment, which was partly explained by the overrepresentation of highly educated persons in the surveillance system due to their better access to health care and their willingness to visit a doctor even for mild symptoms. It has also been hypothesized that groups with higher education tend to eat outside of the home more frequently and are more likely to own pets [25], which are considered to be sources of Salmonella for individuals [26, 27]. In another US ecological study, Chang et al. [28] showed that areas with a higher black, Hispanic, or Latino population had a higher likelihood of Salmonella infections than areas with a predominantly white population. However, S. Enteritidis infections might have different SES risk factors than other Salmonella serotypes.
Methods
Study population and study design
In Ontario, there are 36 PHUs mandated by the provincial ministry of health to manage health promotion and disease prevention programs. Our study area included three PHUs within the GTA, namely, the City of Toronto Health Unit, Peel Regional Health Unit, and York Regional Health Unit. The analysis was conducted at the FSAlevel, a welldefined zone within a larger geographic region, represented by the first three characters of the postal code. We excluded FSAs with less than 500 residents (n = 12) to obtain stable incidence rates, and FSAs with centroids outside of the three PHU boundaries (n = 7) to attain a welldelimited and uniform geographical study area. We included 153 FSAs containing a combined population of 4,570,151 that accounted for 40% of Ontario’s population in 2006 (Figure 1). The annual population sizes of the 153 FSAs ranged from 2,172 to 84,180 persons, with a mean of 29,870 persons.
Data sources
i) Ontario Ministry of Health and LongTerm Care, integrated Public Health Information System
In Ontario, a confirmed case of salmonellosis is defined as the isolation of Salmonella spp. (excluding Salmonella Typhi or Paratyphi) from an appropriate clinical sample (e.g. stool, urine, blood) with or without clinicallycompatible signs and symptoms [30]. Salmonellosis is a reportable disease under provincial legislation [31]; all infections confirmed by hospital, private, and public health laboratories must be reported to the local PHU, whose personnel are required to investigate each case and enter the case’s demographic information (age, sex, area of residence), clinical features, and exposure history to the MOHLTC through iPHIS. In addition, all clinical Salmonella isolates are sent to Toronto Public Health Laboratories for confirmation and serotyping using conventional methods [32]. Information pertaining to S. Enteritidis cases’ PHU, FSA, and date of onset of illness were extracted from iPHIS. Within the iPHIS system, individuals who experience a recurrent episode of a reportable disease receive the same case number. Therefore, we were able to use unique deidentified case numbers to identify recurrent episodes of S. Enteritidis infections. If duplicate case numbers were identified, only the first episode was included in the analysis. Our study data represent all S. Enteritidis infections from the three PHUs captured within the database between January 1, 2007 and December 31, 2009. During the study period, no major outbreaks were reported in the GTA. Reports obtained from iPHIS are useful for evaluating food source attribution and demographic risk factors, yet they lack SES indicators, such as family income, family size, or level of education, due to privacy legislation. Our data did not have any personal or health information that could be linked back to the original identifiers; therefore, the University of Guelph Ethics Review Board did not require ethics approval for this study.
ii) Census of Canada
The Census of Canada is administered every five years by Statistics Canada and collects selfreported demographic and socioeconomic characteristics on Canadians [33]. Census data are available for various geographic areas, including country, provinces and territories, health regions, and FSAs. Forward sortation area was selected as the unit of interest for our study because FSAs in the GTA have welldefined geographical boundaries; FSAlevel SES indicators for the three PHUs of interest within the GTA were obtained from the 2006 Census.
Statistical analysis
Descriptive statistics
The unadjusted S. Enteritidis incidence per 100,000 personyears per FSA over the study period was calculated by dividing the sum of the annual number of S. Enteritidis cases per FSA by the sum of the annual FSA population estimates obtained from the 2006 Census. A choropleth map of the unadjusted incidence was created using ArcGIS 10 (ESRI Inc., Redlands, CA, US); Jenk’s optimization classification method was used to define the critical intervals for mapping [34]. This method arranges data into classes based on their distribution by using an algorithm that reduces variance within groups and maximizes variance between groups.
In addition, the relative risk per FSA was calculated by dividing the observed number of cases per FSA by the expected number of cases per FSA (calculation described in the ‘Dependent variable’ subsection), and then mapped using ArcGIS 10 to illustrate FSAs with excess risk after adjusting for the age and sex distribution of the population.
Spatial analysis
Date of illness onset was used to assign each case to a particular month and year. Each case was categorized into one of five age groups: 0–9, 10–24, 25–34, 35–49, and 50+ years. Children 0–9 were grouped into one category to account for children’s higher risk of enteric infections [16, 21]. Adolescents and young adults were grouped together because of their similar rate of S. Enteritidis infections [29]. Adults 50 years of age and older were pooled into one category because of the small number of cases in this age group. Cartesian coordinates of latitude and longitude for each FSA centroid were calculated in ArcGIS 10. A database was created in a Microsoft Excel 2010 (Microsoft Corporation, Redmond, WA, US) spreadsheet containing each case identifier and its corresponding information (FSA, FSA population estimate, Cartesian coordinates of the FSA centroid, illness onset month and year, age, and sex), which was subsequently imported into the SaTScan™ software [35] for analysis.
To identify clusters with high infection rates we employed the retrospective, purely spatial scan statistic that used the discrete Poisson model [36]. Age and sex were added to the model to account for possible confounding effects. The model assumes that the number of cases in each FSA are Poissondistributed, based on a known underlying population at risk. The scan statistic uses a circular scanning window of variable radii. The null hypothesis assumes equal infection rates inside the scanning window compared to outside the scanning window, whereas the alternative hypothesis expects a higher infection rate inside the scanning window compared to outside [37]. The scan statistic uses a maximum likelihood function that identifies high rate clusters that are least likely to have happened by chance alone. A pvalue was obtained through Monte Carlo hypothesis testing [38] using 999 replications and a relative risk is allocated to this cluster. For our analysis, the maximum spatial circular window size was set to include up to 50% of the population at risk [37]. Secondary clusters were also reported if they did not overlap with the primary cluster. Spatial clusters significant at α = 0.05 were illustrated in a map using ArcGIS 10.
Regression analysis
A regression analysis (described below) was employed to identify arealevel associations between SIRs of S. Enteritidis infections and SES variables in order to better understand the underlying socioeconomic factors that might contribute to spatial clustering of S. Enteritidis infections within the three PHUs.
Variable selection and creation
i) SES variables
According to the Census of Canada, a census family is a married couple (with or without children of either or both spouse), commonlaw couple (with or without children of either or both partner), or loneparent family regardless of sex of parent. A couple may be of opposite or same sex. All members of a specific census family must live in the same household.
 1)
Average number of children at home per census family, which was calculated by summing the number of children per family then dividing by the total number of families.
 2)
Average number of persons per census family, which was calculated by summing the number of persons per family then dividing by the total number of families.
 3)
Average number of rooms per home, which was calculated by summing the number of rooms in a home then dividing by the total number of homes.
 4)
Immigrant population proportion, which was calculated by dividing the immigrant population by the total population and multiplying by 100. The term immigrant refers to a person who is, or has ever been, a landed immigrant in Canada. A landed immigrant is a person born outside of Canada who has been granted the right to live in Canada permanently by immigration authorities.
 5)
Average median family income, which was calculated by dividing the sum of the median income of families in a household by the total number of households. The Census dictionary defines the median income of families in a household as “that amount which divides family member’s income size distribution, ranked by size of income, into two halves. Median incomes of families in a household are normally calculated for all members of the family, whether or not they reported income”. The total income of the family members was defined as the sum of the income obtained from wages, salaries, bonuses, interest, and any provincial  federal benefits and welfare payments.
 6)
Proportion of university graduates of persons between 25 and 64 years of age, which was calculated by dividing the number of persons between 25 and 64 years of age with a university degree by the total population between 25 and 64 years of age and multiplying by 100.
 7)
Unemployment rate of persons 15 years of age or older, which was calculated by dividing the number of unemployed by the number of people in the labour force and multiplying by 100.
 8)
Visible minority population proportion, which was calculated by dividing the visible minority population by the total population and multiplying by 100. Visible minority, according to the Employment Equity Act, refers to a person, other than an Aboriginal person, who is nonCaucasian in race or nonwhite in colour.
ii) Dependent variable
The dependent variable (observed number of cases) was the number of laboratoryconfirmed S. Enteritidis human infections reported in iPHIS from January 1, 2007 through December 31, 2009 from the three PHUs within the GTA, aggregated to the FSAlevel. Using ageandsexbased population estimates from the 2006 Census, the expected number of cases for each FSA was calculated using indirect standardisation [39].
Modeling approach
A database containing the SES variables, dependent variable, and expected number of cases for each FSA was constructed in Microsoft Office Excel 2010, and subsequently imported into STATA Intercooled 10.1 statistical software (Stata Corporation, College Station, TX, US) for analysis. We initially analyzed the data using a Poisson regression model. To estimate SIRs of S. Enteritidis infections, the natural logtransformed expected number of cases was used as the offset to account for differences in FSAlevel ageandsexbased population that might affect FSAlevel infection rates [40].
We assessed the relationship between the natural log of the SIR of S. Enteritidis infections and each of the continuous SES variables by a locally weighted regression method using lowess curves. If nonlinearity was observed, we introduced a quadratic term. If the quadratic term was statistically significant (p ≤ 0.05), collinearity between the linear and quadratic terms was reduced by subtracting the mean value from each individual value of the linear term (centering), and then squaring each centered value [41]. If the nonlinear relationship was not adequately modelled with a quadratic term, the variable was categorized into three equal groups  high, medium, and low  based on its distribution over all FSAs. The medium category was chosen as the reference category.
Univariable screening (α = 0.10) consisted of individually regressing each SES variable on the dependent variable. Pairwise correlation coefficients using the Spearman’s rank test among all significant independent variables (p < 0.10) were examined to prevent the inclusion of collinear variables in the multivariable model. If two variables were highly correlated (rho > 0.70), the variable with the smallest pvalue on univariable analysis was considered for the multivariable model.
All unconditionally significant variables identified on univariable screening, and a categorical variable signifying the three PHUs, were offered to a multivariable model, and a manual backward elimination process was performed. Variables with p > 0.05 were removed unless there was evidence of confounding (when removal of a nonintervening variable changed the coefficients of remaining variables by more than 25%) [41]. Once a main effects model was created, twoway interaction terms between significant SES variables were tested and retained in the final model if they were significant (p ≤ 0.05).
The fit of the Poisson model was assessed using the Pearson χ^{2} goodnessoffit test [42]. The test was significant (χ^{2} = 178.84, P = 0.03) indicating that the model did not fit the data; therefore, the analysis was repeated using a negative binomial regression model.
The final negative binomial model was evaluated to assess model fit and to measure spatial autocorrelation among residuals. First, we assessed Ansombe residuals for each FSA to identify outliers, and then a normal quantile plot (quantiles of the residuals against the quantiles of the normal distribution) was used to assess model fit [43]. In addition, we visually assessed Ansombe residuals in a choropleth map using ArcGIS 10 to identify FSAs with over or underpredicted values (> 3 or < − 3). Secondly, spatial autocorrelation of FSAlevel residuals was evaluated by applying the global Moran’s I statistic [44]. A significant global Moran’s I (p ≤ 0.05) would indicate that FSAlevel residuals are not randomly distributed across the study area and accounting for spatial clustering among FSAs would be required.
For each SES variable from the final model with a significant quadratic term, we plotted the variable (linear and quadratic terms) against the predicted log SIR of S. Enteritidis infections in order to describe the curvilinear relationship while controlling for the other variables in the model. For each SES variable from the final model, choropleth maps were created in ArcGIS 10 to visualize the FSAlevel distribution of the data.
Results
Descriptive statistics
Spatial clustering
Spatial cluster of high Salmonella Enteritidis infection rates in the Greater Toronto Area, Ontario, Canada ^{ a) }
Location Ids (Forward Sortation Areas)  Observed cases  Expected cases  Observed / Expected  Relative risk  Log likelihood ratio  Pvalue 

M5T, M5G, M5S, M5B, M6J, M5V, M5C, M4Y, M5E  40  21.56  1.85  1.90  6.49  0.050 
Negative binomial regression
Results of univariable negative binomial regression models (n = 846 cases from 153 FSAs)
Variable ^{a)}  Type  IRR ^{c)}(95% CI)  P  value  

Average number of children at home per census family  Linear (X)  1.03 (0.71, 1.50)  0.881  
Quadratic (X^{^2}) ^{b)}  4.45 (1.81, 10.92)  0.001  
Average number of persons per census family  Linear (X)  0.89 (0.64, 1.25)  0.515  
Quadratic (X^{^2})  4.47 (1.82, 10.98)  0.001  
Average number of rooms per home  Linear (X)  0.97 (0.91, 1.04)  0.398  
Quadratic (X^{^2})  1.08 (1.03, 1.13)  0.002  
Immigrant population proportion  Linear (X)  0.99 (0.99, 1.00)  0.284  
Quadratic (X^{^2})  0.99 (0.99, 0.99)  0.014  
Average median family income  Categorical  Low (< CAD ^{d)} 65,000)  1.30 (1.08, 1.57)  0.005 
Medium (CAD 65,000  85,000)  Reference    
High (> CAD 85,000)  1.31 (1.08, 1.58)  0.006  
Proportion of university graduates of persons between 25 to 64 years of age  Categorical  Low (10.4  28.6)  0.93 (0.77, 1.12)  0.446 
Medium (28.7 – 40.0)  Reference    
High (40.4 – 72.5)  0.95 (0.78, 1.16)  0.640  
Unemployment rate of persons 15 years of age or older  Categorical  Low (0  6.0)  1.06 (0.87, 1.29)  0.544 
Medium (6.1  7.5)  Reference    
High (7.6  11.0)  1.08 (0.89, 1.32)  0.423  
Visible minority population proportion  Categorical  Low (2.1  29.4)  0.78 (0.64, 0.94)  0.009 
Medium (29.5  51.8)  Reference    
High (51.9  93.4)  0.77 (0.64, 0.92)  0.004 
Results of the final multivariable negative binomial regression model (n = 846 cases from 153 FSAs)
Variable ^{a)}  Type  Estimate (95% CI)  IRR ^{c)}(95% CI)  Pvalue  

Average number of children at home per census family  Linear (X) ^{b)}  0.22 (−0.20, 0.64)  1.24 (0.82, 1.89)  0.313  
Quadratic (X^{^2})  1.69 (0.84, 2.53)  5.40 (2.32, 12.54)  <0.001  
Average median family income  Categorical  Low (< CAD ^{d)} 65,000)  0.29 (0.11, 0.47)  1.34 (1.12, 1.61)  0.002 
Medium (CAD 65,000  85,000)  Reference      
High (> CAD 85,000)  0.22 (0.03, 0.40)  1.24 (1.03, 1.49)  0.020  
Visible minority population proportion  Categorical  Low (2.1  29.4)  −0.16 (−0.35, 0.036)  0.85 (0.70, 1.04)  0.110 
Medium (29.5  51.8)  Reference      
High (51.9  93.4)  −0.27 (−0.44, 0.10)  0.76 (0.64, 0.91)  0.002  
Intercept      −0.12 (−0.29, 0.05)    0.178 
Discussion
Our study used a spatial scan statistic to identify high rate clusters of S. Enteritidis infections in three PHUs within the GTA in Ontario. Multivariable negative binomial regression was employed to identify associations between S. Enteritidis infections and SES variables to better understand the underlying SES risk factors that might contribute to the spatial clustering of S. Enteritidis infections. Our unit of analysis was FSAs from these PHUs, where the S. Enteritidis case data acquired from the Ontario Ministry of Health and LongTerm Care (MOHLTC) integrated Public Health Information System (iPHIS) database, and the SES information obtained from the 2006 Census of Canada, were aggregated, merged, and analyzed. We combined traditional statistical methods with GIS technology (choropleth maps) for the statistically significant high rate cluster and the distribution of statistically significant SES indicators, which was a useful approach that assisted with visualizing FSAlevel associations across the GTA.
We found that FSAs with high and FSAs with low average median family income had a higher incidence of S. Enteritidis infections compared to FSAs with medium average median family income (the reference category). An ecological study conducted in the US demonstrated higher salmonellosis rates in areas with high family income population compared to areas with low family income population [25]. The researchers hypothesized that residents from higher income areas had greater access to healthcare, healthseeking behaviour, and pet ownership, and had eating behaviours that increased resident’s salmonellosis risk. While there could be any number of underlying individuallevel factors to explain our findings, one additional hypothesis for increased salmonellosis rates in high income neighbourhoods might be residents’ more frequent international travel, which has been shown to be an important risk factor for salmonellosis [9, 18–20]. In Ontario, international travel, eating at a food establishment, and eating at home were the top three exposure settings for S. Enteritidis infections at the individual level during the study period (unpublished data). In contrast to our study, the US study had the lowest salmonellosis rates in low median family income areas. This discrepancy might be partly explained by the fact that all Canadian citizens, permanent residents, and landed immigrants have access to provinciallyfunded healthcare, which likely eliminates or reduces bias related to healthcare accessibility, compared to the US, where a large proportion of the population does not have access to funded healthcare [45]. Moreover, the US study combined all Salmonella serotypes. The pathogenicity and source attribution of Salmonella serotypes can differ [20, 46, 47], which might have led to non serotypespecific associations in the US study that are not representative of S. Enteritidis infections. Additionally, the higher rates of infection from low income areas in our study might be explained by poorer microbial quality of foods consumed [48], or by more frequent retail food safety violations [49, 50] that could increase residents’ risk of foodborne diseases, including S. Enteritidis infections. Future research should consider public health inspection violations in different retail settings (e.g. stores, restaurants, food suppliers) to determine if there is an FSAlevel association between S. Enteritidis infection rates and health inspection violations, and whether the association depends on average median family income or other FSAlevel SES factors.
We found a positive curvilinear association between the FSAlevel average number of children at home per census family and the SIR of S. Enteritidis infections. This finding could be explained by children’s higher susceptibility to salmonellosis, which has been demonstrated by individuallevel studies [29, 51, 52]. Risk factors associated with S. Enteritidis infections in children include international travel [53, 54], riding in shopping carts and exposure to raw meat and poultry products [55], persontoperson transmission in daycare centres and in private homes [56], and contact with reptiles [54, 57, 58] and cats [57]. In addition, an ecological study conducted in the US identified a positive association between salmonellosis and counties with a high proportion of young children, which appears to be a consistent risk factor for all Salmonella serotypes [28]. This finding might be explained by the higher severity of salmonellosis in this age group, and consequently an increased likelihood of visiting a physician and getting tested and reported [28]. However, we acknowledge that the average number of children at home per census family and the average number of persons per census family could be substitutes for each other due to their high correlation, which might suggest that home density and eating habits of larger families might have a role in the positive association with the SIR of S. Enteritidis infections. Future individuallevel research is needed in Ontario to elucidate our finding.
Forward sortation areas with high visible minority population proportion had a lower SIR of S. Enteritidis infections compared to FSAs with medium visible minority population proportion. This finding is in contrast with a previous US ecological study, in which researchers demonstrated increased salmonellosis rates in areas with higher black, Hispanic, or Latino populations compared to areas with a predominantly Caucasian population [28]. However, comparing nonCaucasian populations between Canada and the US should be done with caution, because in the GTA, a high proportion of the nonCaucasian population is South Asian or Chinese [59]. Of note, in our study no significant difference was found between the low and high visible minority population proportion FSAs, which warrants further individuallevel studies to better understand the effect of ethnicity on the incidence of S. Enteritidis infections.
A single spatial cluster of higher than expected incidence rates of S. Enteritidis infections was identified in the southcentral area (downtown) of the City of Toronto Health Unit that included nine neighbouring FSAs. The majority of these FSAs had SES characteristics that were positively associated with the SIR of S. Enteritidis infections in the regression model (low average median family income, medium visible minority population proportion), which is consistent with these SES indicators playing a significant role in the clustering of S. Enteritidis infections. However, many of the FSAs in this cluster had one SES characteristic that was inconsistent with the regression model (low average number of children at home per census family), which warrants further spatial epidemiological assessment. Applying GIS technology to visualize the FSAlevel spatial cluster of high incidence rates of S. Enteritidis infections and the FSAlevel distribution of significant SES variables was an useful technique to highlight areas where prevention and control programs should be targeted, and where future studies should be conducted to understand the underlying individuallevel risk factors.
First and secondorder spatial effects must be considered when analyzing spatial epidemiological data [39]. Firstorder effects in our analysis could be defined as variation in the mean value of the S. Enteritidis infection rate in the study area (i.e. a global or largescale trend). We addressed these effects by using a multivariable negative binomial regression model. Secondorder effects in our study could be defined as spatial dependence in the S. Enteritidis infection rates (i.e. local or smallscale effects). It could be expected that neighbouring FSAs have more common infection rates and SES indicator features than distant FSAs. We assessed the presence of secondorder effects by evaluating the spatial autocorrelation of FSAlevel Anscombe residuals using the global Morans’s I statistic. The statistic was not significant indicating that the SES variables explained the variation in S. Enteritidis infection rates across the study area well and no secondorder effects remained once these fixed effects were accounted for in our model.
Before generalizing our findings, a few limitations need to be considered. Passive laboratorybased surveillance programs underestimate the true burden of enteric infections. In Canada, it was estimated that for every reported salmonellosis case there were 26.1 cases in the general population that remained unreported [60]. There might also be geographic variations in underreporting of S. Enteritidis infections due to differences in utilization of health care providers or the sensitivity of testing methods used at different laboratories [61, 62]. However, in our study, these factors might not be as influential because our study area enclosed three neighbouring PHUs with similar healthcare providers.
A second limitation of our study is that we were unable to adjust our estimates for the seasonal variation in the incidence rate of S. Entertidis infections [29]. To account for month and year in our analysis we would need to have arealevel ageandsexbased monthly and yearly population estimates, and monthly and yearly socioeconomic status estimates. Socioeconomic data are collected every five years in Canada and the data are not subdivided into smaller time units (e.g. month or year).
Another possible limitation is that ecological studies only consider variation between groups and not within groups [63], making them unsuitable to relate population level risk factors to the individual level. However, ecological studies can be considered as a costeffective and useful alternative to individuallevel studies, where regional differences in SES risk factors can be evaluated and identified, which can be used by public health authorities to further assess these risk factors and target prevention and control programs [25, 28]. It is also recognized that analyzing associations among S. Enteritidis infections and SES variables at higher geographic levels (e.g. health unit, province) might have given different results [64] due to the Modifiable Areal Unit problem. However, by analyzing the data at a lower geographic level (in our case, the FSAlevel), we have likely decreased the ecological bias (by reducing the variation of SES factors within each region), and obtained more conservative estimates of our regression coefficients [65].
Our main objective was to identify arealevel associations between S. Enteritidis infection rates and SES indicators. Several variables that we investigated were based on census families (average number of children per census family, average number of persons per census family, average median family income); consequently, for these three SES indicators, single person homes were not investigated in our study, which could limit the external validity of our estimates. However, within the study area, the proportion of single person households was considerably lower than census family households [33].
Finally, we analyzed S. Enteritidis case data from 2007 to 2009; however, the only available socioeconomic data were from the 2006 Census of Canada. Population changes over the study period could have introduced bias into our study. However, it is likely that SES indicators and population sizes did not change drastically over this relatively short time period. We excluded FSAs with a population of less than 500 residents; however, these areas accounted for less than one percent of the total population, and therefore it is unlikely that these exclusions affected our results.
Conclusions
Our study demonstrated the usefulness of combining GIS technology and conventional and spatial statistical methods to identify high rate clusters and to analyze arealevel associations between S. Enteritidis infection rates and SES variables in three PHUs within the GTA by using an ecological study approach. We found a higher incidence rate of S. Enteritidis infections in FSAs with low and high average median family income compared to FSAs with medium average median family income that might be explained by the poorer quality of food consumed in the low income areas, in differences in food preparation and consumption practices among various areas, and in high income area residents’ more frequent international travel that increases their risk of salmonellosis. A positive curvilinear relationship was observed between the FSAlevel SIR of S. Enteritidis infections and average number of children at home per census family that might be attributable to children’s greater susceptibility to enteric infections. We did not detect any significant spatial dependency of residuals indicating that our fixed effects model explained the spatial dependency well.
The high rate cluster of S. Enteritidis infections detected by the scan statistic contained mainly FSAs in which the majority of the SES variables identified in the regression model were positively associated with S. Enteritidis infections, suggesting that these SES indicators (low average median family income, medium visible minority population proportion) significantly contributed to the spatial clustering of S. Enteritidis infections.
These findings will aid public health policy makers and practitioners to further evaluate individuallevel SES indicators. Areas with low and high average median family income, medium proportion of visible minority population, and high average number of children at home per census family should be targeted when designing disease control and prevention programs within these PHUs. Further studies are needed in areas with high S. Enteritidis infection rates to identify sources of environmental contamination of the local food supply, to assess food safety practices at local food markets, retail stores, and restaurants, to identify novel individuallevel risk factors, and to detect spatiotemporal trends of S. Enteritidis infections.
Abbreviations
 CI:

Confidence interval
 FSA:

Forward sortation area
 GIS:

Geographic Information System
 GTA:

Greater Toronto Area
 iPHIS:

integrated Public Health Information System
 IRR:

Incidence rate ratio
 LR:

Likelihood ratio
 MOHLTC:

Ontario Ministry of Health and LongTerm Care
 PHU:

Public Health Unit
 S. Enteritidis:

Salmonella enterica subspecies enterica serotype Enteritidis
 SES:

Socioeconomic status
 SIR:

Standardized incidence rate
 US:

United States of America.
Declarations
Acknowledgements
The authors acknowledge the MOHLTC for providing the data. We thank the staff of all PHUs and public health laboratories that tested samples, followed up with cases, and entered information into the public health surveillance database. We also acknowledge the Data Resource Centre at the University of Guelph library for their GIS and census support. The views expressed in this study are the views of the authors and do not necessarily reflect those of the MOHLTC.
Authors’ Affiliations
References
 Majowicz SE, Musto J, Scallan E, Angulo FJ, Kirk M, O’Brien SJ, Jones TF, Fazil A, Hoekstra RM: International collaboration on enteric disease ‘burden of Illness’ studies: the global burden of nontyphoidal salmonella gastroenteritis. Clin Infect Dis. 2010, 50 (6): 8829. 10.1086/650733.View ArticlePubMedGoogle Scholar
 Newell DG, Koopmans M, Verhoef L, Duizer E, AidaraKane A, Sprong H, Opsteegh M, Langelaar M, Threfall J, Scheutz F, van der Giessen J, Kruse H: Foodborne diseases  the challenges of 20 years ago still persist while new ones continue to emerge. Int J Food Microbiol. 2010, 139 (Suppl 1): S315.View ArticlePubMedGoogle Scholar
 Cardoen S, Van Huffel X, Berkvens D, Quoilin S, Ducoffre G, Saegerman C, Speybroeck N, Imberechts H, Herman L, Ducatelle R, Dierick K: Evidencebased semiquantitative methodology for prioritization of foodborne zoonoses. Foodborne Pathog Dis. 2009, 6 (9): 108396. 10.1089/fpd.2009.0291.View ArticlePubMedGoogle Scholar
 Verge P, Cloeckaert A, Barrow P: Emergence of Salmonella epidemics: the problems related to Salmonella enterica serotype Enteritidis and multiple antibiotic resistance in other major serotypes. Vet Res. 2005, 26: 267288.Google Scholar
 Weinberger M, Keller N: Recent trends in the epidemiology of nontyphoid Salmonella and antimicrobial resistance: the Israeli experience and worldwide review. Curr Opin Infect Dis. 2005, 18 (6): 51321. 10.1097/01.qco.0000186851.33844.b2.View ArticlePubMedGoogle Scholar
 Public Health Agency of Canada (PHAC): National Enteric Surveillance Program. Annual Summary of Laboratory Surveillance Data for Enteric Pathogens in Canada. 2009Google Scholar
 Ruzante JM, Majowicz SE, Fazil A, Davidson VJ: Hospitalization and deaths for select enteric illnesses and associated sequelae in Canada, 2001–2004. Epidemiol Infect. 2011, 139 (6): 93745. 10.1017/S0950268810001883.View ArticlePubMedGoogle Scholar
 Nesbitt A, Ravel A, Murray R, McCormick R, Savelli C, Finley R, Parmley J, Agunos A, Majowicz SE, Gilmour M: The Canadian integrated program for antimicrobial resistance surveillance public health partnership; the Canadian public health laboratory network: integrated surveillance and potential sources of salmonella enteritidis in human cases in Canada from 2003 to 2009. Epidemiol Infect. 2011, 14: 116.Google Scholar
 Voetsch AC, Poole C, Hedberg CW, Hoekstra RM, Ryder RW, Weber DJ, Angulo FJ: Analysis of the FoodNet case–control study of sporadic Salmonella serotype Enteritidis infections using persons infected with other Salmonella serotypes as the comparison group. Epidemiol Infect. 2009, 137 (3): 40816. 10.1017/S0950268808000897.View ArticlePubMedGoogle Scholar
 Kimura AC, Reddy V, Marcus R, Cieslak PR, MohleBoetani JC, Kassenborg HD, Segler SD, Hardnett FP, Barrett T, Swerdlow DL: Emerging infections program FoodNet working group: chicken consumption is a newly identified risk factor for sporadic salmonella enterica serotype enteritidis infections in the United States: a case–control study in FoodNet sites. Clin Infect Dis. 2004, 38 (Suppl 3): S24452.View ArticlePubMedGoogle Scholar
 Kuehn BM: Salmonella cases traced to egg producers: findings trigger recall of more than 500 million eggs. JAMA. 2010, 304 (12): 131610.1001/jama.2010.1330.View ArticlePubMedGoogle Scholar
 Doorduyn Y, Van Den Brandhof WE, Van Duynhoven YT, Wannet WJ, Van Pelt W: Risk factors for Salmonella Enteritidis and Typhimurium (DT104 and nonDT104) infections in The Netherlands: predominant roles for raw eggs in Enteritidis and sandboxes in Typhimurium infections. Epidemiol Infect. 2006, 134 (3): 61726. 10.1017/S0950268805005406.View ArticlePubMedGoogle Scholar
 Medus C, Smith KE, Bender JB, Leano F, Hedberg CW: Salmonella infections in food workers identified through routine Public Health Surveillance in Minnesota: impact on outbreak recognition. J Food Prot. 2010, 73 (11): 20538.PubMedGoogle Scholar
 Hedican E, Hooker C, Jenkins T, Medus C, Jawahir S, Leano F, Smith K: Restaurant Salmonella Enteritidis outbreak associated with an asymptomatic infected food worker. J Food Prot. 2009, 72 (11): 23326.PubMedGoogle Scholar
 Beatty ME, Shevick G, ShupeRicksecker K, Bannister E, Tulu A, Lancaster K, Alexander N, Zellner DE, Lyszkowicz E, Braden CR: Large Salmonella Enteritidis outbreak with prolonged transmission attributed to an infected food handler, Texas, 2002. Epidemiol Infect. 2009, 137 (3): 41727. 10.1017/S0950268808001362.View ArticlePubMedGoogle Scholar
 Younus M, Wilkins MJ, Davies HD, Rahbar MH, Funk J, Nguyen C, Siddiqi AE, Cho S, Saeed AM: The role of exposures to animals and other risk factors in sporadic, nontyphoidal Salmonella infections in Michigan children. Zoonoses Public Health. 2010, 57 (7–8): 1706.View ArticleGoogle Scholar
 Marcus R, Varma JK, Medus C, Boothe EJ, Anderson BJ, Crume T, Fullerton KE, Moore MR, White PL, Lyszkowicz E, Voetsch AC, Angulo FJ: Emerging Infections Program FoodNet Working Group: Reassessment of risk factors for sporadic Salmonella serotype Enteritidis infections: a case–control study in five FoodNet Sites, 2002–2003. Epidemiol Infect. 2007, 135 (1): 8492. 10.1017/S0950268806006558.View ArticlePubMedGoogle Scholar
 Johnson LR, Gould LH, Dunn JR, Berkelman R: Mahon For The Foodnet Travel Working Group BE: Salmonella Infections Associated with International Travel: A Foodborne Diseases Active Surveillance Network (FoodNet) Study. Foodborne Pathog Dis. 2011, 8 (9): 10317. 10.1089/fpd.2011.0854.View ArticlePubMedGoogle Scholar
 Tighe MK, Savage R, Vrbova L, Toolan M, Whitfield Y, Varga C, Lee B, Allen V, Maki A, Walton R, Johnson C, Dhar B, Ahmed R, Crowcroft NS, Middleton D: The epidemiology of travelrelated Salmonella Enteritidis in Ontario, Canada, 2010–2011. BMC Public Health. 2012, 12 (1): 31010.1186/1471245812310.View ArticlePubMedPubMed CentralGoogle Scholar
 Domingues AR, Pires SM, Halasa T, Hald T: Source attribution of human salmonellosis using a metaanalysis of case–control studies of sporadic infections. Epidemiol Infect. 2012, 140 (6): 95969. 10.1017/S0950268811002172.View ArticlePubMedGoogle Scholar
 Majowicz SE, Horrocks J, Bocking K: Demographic determinants of acute gastrointestinal illness in Canada: a population study. BMC Public Health. 2007, 18 (7): 162View ArticleGoogle Scholar
 Green CG, Krause DO, Wylie JL: Spatial analysis of Campylobacter infection in the Canadian province of Manitoba. Int J Health Geogr. 2006, 5: 210.1186/1476072X52.View ArticlePubMedPubMed CentralGoogle Scholar
 Simonsen J, Frisch M, Ethelberg S: Socioeconomic risk factors for bacterial gastrointestinal infections. Epidemiology. 2008, 19 (2): 28290. 10.1097/EDE.0b013e3181633c19.View ArticlePubMedGoogle Scholar
 Broner S, Torner N, Dominguez A, Martínez A, Godoy P: The Working Group for the Study of Outbreaks of Acute Gastroenteritis in Catalonia: Sociodemographic inequalities and outbreaks of foodborne diseases: An ecologic study. Food Control. 2010, 21: 947951. 10.1016/j.foodcont.2009.12.002.View ArticleGoogle Scholar
 Younus M, Hartwick E, Siddiqi AA, Wilkins M, Davies HD, Rahbar M, Funk J, Saeed M: The role of neighborhood level socioeconomic characteristics in Salmonella infections in Michigan (1997–2007): assessment using geographic information system. Int J Health Geogr. 2007, 6: 5610.1186/1476072X656.View ArticlePubMedPubMed CentralGoogle Scholar
 Leonard EK, Pearl DL, Finley RL, Janecko N, Peregrine AS, ReidSmith RJ, Weese JS: Evaluation of petrelated management factors and the risk of Salmonella spp. carriage in pet dogs from volunteer households in Ontario (2005–2006). Zoonoses Public Health. 2011, 58 (2): 1409. 10.1111/j.18632378.2009.01320.x.View ArticlePubMedGoogle Scholar
 Reaser JK, Clark EE, Meyers NM: All creatures great and minute: a public policy primer for companion animal zoonoses. Zoonoses Public Health. 2008, 55 (8–10): 385401.PubMedGoogle Scholar
 Chang M, Groseclose SL, Zaidi AA, Braden CR: An ecological analysis of sociodemographic factors associated with the incidence of salmonellosis, shigellosis, and E. coli O157:H7 infections in US counties. Epidemiol Infect. 2009, 137 (6): 81020. 10.1017/S0950268808001477.View ArticlePubMedGoogle Scholar
 Varga C, Pearl DL, McEwen SA, Sargeant JM, Pollari F, Guerin MT: Incidence, distribution, seasonality, and demographic risk factors of Salmonella Enteritidis human infections in Ontario, Canada, 2007–2009. BMC Infect Dis. 2013, 13: 21210.1186/1471233413212.View ArticlePubMedPubMed CentralGoogle Scholar
 Ministry of Health and LongTerm Care: Infectious Diseases Protocol. Toronto: Queen’s Printer for Ontario. 2009, http://www.health.gov.on.ca/english/providers/program/pubhealth/oph_standards/ophs/infdispro.html,Google Scholar
 Health Protection and Promotion Act, R.S.O. 1990, http://www.elaws.gov.on.ca/html/statutes/english/elaws_statutes_90h07_e.htm, , c. H.7,
 Popoff MY, LeMinor L: Antigenic Formulas of the Salmonella Serovars. 1997, Pasteur Institute, Paris, France: World Health Organization Collaborating Centre for Reference and Research on Salmonella, 6Google Scholar
 Statistics Canada: 2006 Census Dictionary: Census Year. 2006, Ottawa, http://www12.statcan.gc.ca/censusrecensement/2006/ref/dict/pdf/92566eng.pdf, . Catalogue no. 92566X,Google Scholar
 Jenks GF: The data model concept in statistical mapping. International Yearbook of Cartography, Volume 7. Edited by: Zurich FK. 1967, Zurich, Switzerland: Orell Fussli, 186190.Google Scholar
 Kulldorff M and Information Management Services, Inc. SaTScanTM v9.1.1: Software for the spatial and spacetime scan statistics. http://www.satscan.org,
 Kulldorff M: A spatial scan test. Comm Statistics Theory Methods. 1997, 26 (6): 14871496.View ArticleGoogle Scholar
 Kulldorff M, Nagarwalla N: Spatial disease clusters: detection and inference. Stat Med. 1995, 14: 799810. 10.1002/sim.4780140809.View ArticlePubMedGoogle Scholar
 Dwass M: Modified randomization tests for nonparametric hypotheses. Ann Math Statist. 1957, 28: 181187. 10.1214/aoms/1177707045.View ArticleGoogle Scholar
 Pfeiffer DU, Robinson TP, Stevenson M, Stevens KB, Rogers DJ, Clements ACA: Spatial Analysis in Epidemiology. 2008, New York: Oxford University PressView ArticleGoogle Scholar
 Lawson A: Statistical Methods in Spatial Epidemiology. 2006, John Wiley and Sons Ltd: Chichester, EnglandView ArticleGoogle Scholar
 Dohoo I, Martin W, Stryhn H: Veterinary Epidemiology Research. 2009, Charlottetown, Prince Edward Island, Canada: AVC Inc.Google Scholar
 Cameron AC, Trivedi PK: Regression Analysis of Count Data. 1998, New York: Cambridge PressView ArticleGoogle Scholar
 Hilbe J: Negative Binomial Regression. 2007, Cambridge, England: Cambridge University PressView ArticleGoogle Scholar
 Griffith D: Spatial Autocorrelation: A Primer. 1987, Washington, D.C.: Association of American GeographersGoogle Scholar
 Lasser KE, Himmelstein DU, Woolhandler S: Access to care, health status, and health disparities in the United States and Canada: results of a crossnational populationbased survey. Am J Public Health. 2006, 96: 13001307. 10.2105/AJPH.2004.059402.View ArticlePubMedPubMed CentralGoogle Scholar
 King N, Lake R, Campbell D: Source attribution of nontyphoid salmonellosis in new zealand using outbreak surveillance data. J Food Prot. 2011, 74 (3): 43845. 10.4315/0362028X.JFP10323.View ArticlePubMedGoogle Scholar
 Jones TF, Ingram LA, Cieslak PR, Vugia DJ, TobinD’Angelo M, Hurd S, Medus C, Cronquist A, Angulo FJ: Salmonellosis outcomes differ substantially by serotype. J Infect Dis. 2008, 198 (1): 10914. 10.1086/588823.View ArticlePubMedGoogle Scholar
 Koro ME, Anandan S, Quinlan JJ: Microbial quality of food available to populations of differing socioeconomic status. Am J Prev Med. 2010, 38 (5): 47881. 10.1016/j.amepre.2010.01.017.View ArticlePubMedGoogle Scholar
 Signs RJ, Darcey VL, Carney TA, Evans AA, Quinlan JJ: Retail food safety risks for populations of different races, ethnicities, and income levels. J Food Prot. 2011, 74 (10): 171723. 10.4315/0362028X.JFP11059.View ArticlePubMedGoogle Scholar
 Darcey VL, Quinlan JJ: Use of geographic information systems technology to track critical health code violations in retail facilities available to populations of different socioeconomic status and demographics. J Food Prot. 2011, 74 (9): 152430. 10.4315/0362028X.JFP11101.View ArticlePubMedGoogle Scholar
 Younus M, Wilkins MJ, Arshad MM, Rahbar MH, Saeed AM: Demographic risk factors and incidence of Salmonella enteritidis infection in Michigan. Foodborne Pathog Dis. 2006, 3 (3): 26673. 10.1089/fpd.2006.3.266.View ArticlePubMedGoogle Scholar
 Arshad MM, Wilkins MJ, Downes FP, Rahbar MH, Erskine RJ, Boulton ML, Saeed AM: A registrybased study on the association between human salmonellosis and routinely collected parameters in Michigan, 1995–2001. Foodborne Pathog Dis. 2007, 4 (1): 1625. 10.1089/fpd.2006.48.View ArticlePubMedGoogle Scholar
 Ekdahl K, de Jong B, Wollin R, Andersson Y: Travelassociated nontyphoidal salmonellosis: geographical and seasonal differences and serotype distribution. Clin Microbiol Infect. 2005, 11 (2): 138144. 10.1111/j.14690691.2004.01045.x.View ArticlePubMedGoogle Scholar
 Cellucci T, Seabrook JA, Chagla Y, Bannister SL, Salvadori MI: A 10year retrospective review of Salmonella infections at the Children’s Hospital in London, Ontario. Can J Infect Dis Med Microbiol. 2010, 21 (2): 7882.PubMedPubMed CentralGoogle Scholar
 Patrick ME, Mahon BE, Zansky SM, Hurd S, Scallan E: Riding in shopping carts and exposure to raw meat and poultry products: prevalence of, and factors associated with, this risk factor for salmonella and campylobacter infection in children younger than 3 years. J Food Prot. 2010, 73 (6): 1097100.PubMedGoogle Scholar
 Ethelberg S, Olsen KE, GernerSmidt P, Mølbak K: Household outbreaks among cultureconfirmed cases of bacterial gastrointestinal disease. Am J Epidemiol. 2004, 159 (4): 40612. 10.1093/aje/kwh049.View ArticlePubMedGoogle Scholar
 Sauteur PM, Relly C, Hug M, Wittenbrink MM, Berger C: Risk factors for invasive reptileassociated salmonellosis in children. Vector Borne Zoonotic Dis. 2013, 13 (6): 41921. 10.1089/vbz.2012.1133.View ArticleGoogle Scholar
 Aiken AM, Lane C, Adak GK: Risk of Salmonella infection with exposure to reptiles in England, 2004–2007. Euro Surveill. 2010, 15 (22): 19581PubMedGoogle Scholar
 Statistics Canada: Canada’s Ethnocultural Mosaic, 2006 Census: Findings. 2006, Available at: https://www12.statcan.gc.ca/censusrecensement/2006/assa/97562/indexeng.cfm Google Scholar
 Thomas MK, Murray R, Flockhart L, Pintar K, Pollari F, Fazil A, Nesbitt A, Marshall B: Estimates of the Burden of Foodborne Illness in Canada for 30 Specified Pathogens and Unspecified Agents, Circa 2006. Foodborne Pathog Dis. 2013, 10 (7): 63948. 10.1089/fpd.2012.1389.View ArticlePubMedPubMed CentralGoogle Scholar
 Scallan E, Jones TF, Cronquist A, Thomas S, Frenzen P, Hoefer D, Medus C, Angulo FJ, FoodNet Working Group: Factors associated with seeking medical care and submitting a stool sample in estimating the burden of foodborne illness. Foodborne Pathog Dis. 2006, 3 (4): 4328. 10.1089/fpd.2006.3.432.View ArticlePubMedGoogle Scholar
 MacDougall L, Majowicz S, Doré K, Flint J, Thomas K, Kovacs S, Sockett P: Underreporting of infectious gastrointestinal illness in British Columbia, Canada: who is counted in provincial communicable disease statistics?. Epidemiol Infect. 2008, 136 (2): 24856.View ArticlePubMedGoogle Scholar
 Morgenstern H: Ecologic studies in epidemiology: concepts, principles, and methods. Annu Rev Public Health. 1995, 16: 6181. 10.1146/annurev.pu.16.050195.000425.View ArticlePubMedGoogle Scholar
 Weisent J, Rohrbach B, Dunn JR, Odoi A: Detection of high risk campylobacteriosis clusters at three geographic levels. Geospat Health. 2011, 6 (1): 6576.View ArticlePubMedGoogle Scholar
 Armhein C: Searching for the elusive aggregation effect: evidence from statistical simulations. Environ Plan A. 1995, 27 (1): 105119. 10.1068/a270105.View ArticleGoogle Scholar
 The prepublication history for this paper can be accessed here:http://www.biomedcentral.com/14712458/13/1078/prepub
Prepublication history
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.