Relationship Between Census Tract–Level Poverty and Domestically Acquired Salmonella Incidence: Analysis of Foodborne Diseases Active Surveillance Network Data, 2010–2016

Abstract

Salmonellosis is an important public health problem in the United States, accounting for an estimated >1 million infections, >19 000 hospitalizations, and nearly 400 deaths per year [1]. Reducing the incidence of salmonellosis and other foodborne pathogens has been a focus of the Centers for Disease Control and Prevention (CDC), US Department of Agriculture and Food and Drug Administration for decades. To monitor progress in reducing salmonellosis and selected other foodborne pathogens, the Foodborne Disease Active Surveillance Network (FoodNet) was established by the CDC in 1996 [2].

Surveillance conducted by FoodNet and most disease reporting does not routinely include collection of data that include the affected individual’s socioeconomic status (SES) (eg, household income and education level), although analyzing surveillance data through an SES lens can identify previously undocumented and potentially unrecognized health disparities. Recognizing and documenting disparities by SES can help guide overall prevention efforts and set objectives for reducing disparities in the quest for health equity.

In the absence of collecting individual SES data, area-based SES measures that are based on residential address information have been increasingly used to reflect census tract poverty levels [3–11]. The CDC’s Emerging Infectious Disease Program (EIP), including FoodNet, has recently undertaken more systematic efforts to geocode surveillance data and analyze it using area-based measures [12]. Such analyses combining data from all FoodNet sites have included campylobacteriosis [13], Shiga toxin–producing Escherichia coli disease [11], and shigellosis [14].

Two studies have been done in the United States examining salmonellosis surveillance data by area-based SES measures at the local governmental level [15, 16], and 1 study examined national reportable disease data [17]. All 3 studies found a weak association of higher incidence with higher SES. None fully distinguished between domestically and internationally acquired salmonellosis, a potential limitation because international travel in the United States tends to be associated with higher SES [18], and at least international travel–associated non-O157 Shiga toxin–producing E. coli disease in the United States has been more common among those of higher SES [11, 19]. Only 1 study looked at the relationship between Salmonella serotype-specific incidence and SES, finding that the incidence of different serotypes varied by SES [15].

Because each serotype has its own ecologic niche and potential vehicles by which humans may be exposed, it is important to determine whether they differ epidemiologically by SES. The objectives of the current analysis were to examine the overall relationship between the incidence of domestically acquired, laboratory-confirmed salmonellosis and SES as found in FoodNet, as well as the relationship between the incidence of leading, domestically acquired Salmonella serotypes and SES.

METHODS

FoodNet is the principal foodborne disease surveillance component of the CDC’s EIP, a collaboration among CDC, the US Department of Agriculture’s Food Safety and Inspection Services, the Food and Drug Administration, and 10 state health departments. FoodNet includes the states of Connecticut, Georgia, Maryland, Minnesota, New Mexico, Oregon, and Tennessee, along with selected counties in California, Colorado, and New York. FoodNet staff conduct active population-based surveillance for laboratory-confirmed cases of Salmonella and collect and determine the serotype of all available isolates. Enhanced surveillance methods have been described elsewhere [2]. Data collected on each laboratory-confirmed case of Salmonella infection include demographic information (age, sex, race or ethnicity, street address of residence), whether the infection was part of an outbreak, and whether the patient had traveled internationally in the 7 days before onset of illness.

For the current analysis, each FoodNet site geocoded the residential address of all Salmonella cases for the years 2010–2016 inclusive. Geocoded addresses were assigned to census tracts. Census tract poverty level, defined as the percentage of households in the census tract living below the federal poverty level, was determined from the 2011–2015 American Community Survey 5-year estimates [20]. Census tracts were categorized by their percentage of households living below the poverty level (<5%, 5% to <10%, 10% to <20%, or ≥20%), as recommended by the Public Health Disparities Geocoding Project [3] and as used for other multisite EIP data analysis projects [10, 11, 14]. Census tract–specific denominators were determined from the 2010 US Census [21].

Data Analysis

Age-adjusted (2000 US standard population [22]) incidence rates per 100 000 person-years overall and for each of the 4 poverty categories were calculated for all Salmonella cases combined and for each of the 10 leading serotypes. Age standardization was done using 5 age categories: 0–4, 5–17, 18–49, 50–64, and ≥65 years. These categories were based on overall age group-specific incidence rates, combining age-specific rates that were similar into the same age groups. These age-adjusted analyses were done first for all Salmonella cases and then only those that were likely domestically acquired. Outbreak cases were included among the domestically acquired cases.

Presumed domestically acquired cases included 2 groups: cases in interviewed patients with no international travel in the 7 days before onset of illness (70% of all cases) and a subset of cases with no interviews (23% of all patients were not able to be interviewed). This subset was determined by applying the fraction of all interviewed patients in an age and poverty-level-specific group (eg, children <5 years old in the ≥20% poverty group) who had no international travel to the number of patients in that group who were not interviewed. Analysis of interview data showed that the probability of international travel was associated with lower census tract poverty level (progressively ranging from 16.1% of cases in the <5% poverty to 5.1% in the ≥20% poverty group), and the probability of not being interviewed was associated with higher census tract poverty level (progressively ranging from 13.0% of cases in the <5% poverty group to 35.0% in the ≥20% poverty group).

We then examined whether there was a gradient relationship between domestic Salmonella incidence and poverty level (χ ² for trend) across all 4 poverty categories for the overall age-adjusted incidence for all Salmonella cases, then by age group, race/ethnicity, and FoodNet site. We also calculated incidence rate ratios (IRRs) and their 95% confidence intervals (CIs), comparing age-adjusted incidence in the highest (≥20%) with that in the lowest (<5%) poverty category where a statistically significant trend was found. We then examined whether there was a gradient relationship between age-adjusted domestic Salmonella incidence and census tract poverty level for each of the 10 leading serotypes. Because we found the strongest relationships in children <5 years old, we repeated all analyses limited to the cases in this age group.

Finally, we removed outbreak-associated cases to determine whether the findings held for the remaining domestically acquired cases. Statistical analyses were performed using SAS software (version 9.3; SAS Institute). IRRs and 95% CIs were calculated using the Statcalc function in Epi Info 7 software (Centers for Disease Control and Prevention). This study was an analysis of surveillance data and was thus exempt from human subjects review.

RESULTS

A total of 52 821 cases (96%) could be geocoded to the census tract level. The demographic characteristics of these cases, including travel and census tract poverty status, are shown in Table 1. Overall, 48 111 (91.1%) were estimated to be due to domestic exposure, for an incidence of 14.75 per 100 000 person-years. The incidence was highest in children <5 years old (58.25 per 100 000), 3.5 times higher than in the age group with the next-highest incidence, ≥65-year-olds. Although there was little variation by race/ethnicity (range, 11.52 in non-Hispanic Asians to 12.93 in Hispanics), there was a gradient relationship of increasing incidence with increasing census tract poverty (range, 12.15 in the <5% group to 17.21 in the ≥20% group).

The associations of age group-specific and age-adjusted salmonellosis incidence with census tract poverty are shown in Table 2. Statistically significant associations between salmonellosis incidence and increasing census tract poverty were observed in 4 of the 5 age groups, with the strongest association in children <5 years old (relative risk for highest [≥20%] vs lowest [<5%] census tract poverty level, 2.07).

No association was found for those 18–49 years old. Statistically significant associations were also observed for each of the 4 major racial and ethnic groups and by geographic location in 5 of the 10 FoodNet sites (California, Georgia, Maryland, Minnesota, and New Mexico).

Because children <5 years old had the highest incidence and strongest age-specific relationship of incidence to census tract poverty, we further examined whether the relationship held in each race/ethnic group and FoodNet site (Table 3). A relationship was found in each major race/ethnic group in only 5 of the 10 FoodNet sites (California, Colorado, Georgia, Maryland, and Minnesota).

Serotype-Specific Findings

The 10 most common serotypes over the 7 years accounted for 66% of all domestically acquired cases, with 4 accounting for half (50%) of such cases: Salmonella Enteritidis (16%), Salmonella Newport (12%), Salmonella Typhimurium (12%), and Salmonella Javiana (10%) (Table 4). Of note, the percentage of all cases that were domestically acquired varied widely by serotype. Only 82% of S. Enteritidis and 89% of S. Newport and S. Typhimurium cases were domestically acquired, compared with 98% each of Salmonella Muenchen, Salmonella Infantis, Salmonella Heidelberg, Salmonella Saintpaul, and Salmonella Montevideo cases.

For each leading serotype and all nonleading serotypes combined, higher incidence was strongly associated with higher levels of census tract poverty (Table 4). When analysis was restricted to patients <5 years old, the findings were similar, with the exception of those with S. Infantis infection (χ ² for trend = 0.07) (Table 5). Even for S. Infantis, however, the IRR of the highest to the lowest poverty group was >1.0 (IRR, 2.37; 95% CI, 1.38–4.31). Interestingly, when the analysis included cases acquired through international travel, the relationship between S. Enteritidis and census tract poverty was reversed, with higher incidence associated with higher SES (IRR, 0.87; P < .01[χ ² for trend]).

Outbreak-Related Domestically Acquired Cases

Overall, 2989 (6.2%) of domestically acquired cases were part of recognized outbreaks. This percentage varied somewhat by serotype, with 20.4% of S. Heidelberg and 12.0% of S. Infantis cases exceeding the average percentage of cases attributed to outbreaks. The percentage of domestically acquired outbreak-associated cases tended to decrease with increasing poverty level, both overall (6.9% for census tract poverty level <5% to 4.8% for poverty level ≥20%), for children <5 years old and for each serotype. Thus, when these cases were excluded, the magnitude of the association between incidence and increasing census tract poverty became even greater.

Discussion

Domestically acquired, laboratory-confirmed salmonellosis in the United States during the 7-year surveillance period (2010–2016) was statistically associated with increasing census tract poverty. This association was strongest for children <5 years of age, was present among all major racial and ethnic groups, and was observed for the 10 leading Salmonella serotypes. These findings were not seen among persons 18–49 years of age nor among all FoodNet sites. Furthermore, infections acquired from international travel were more likely to occur in higher-SES census tracts. These findings have implications for further study and prevention.

This is the first analysis of surveillance data in the United States to show that salmonellosis is associated with higher levels of poverty, albeit inconsistently across age groups and geographic regions. A major difference between our study and others conducted in the United States [15–17] is that we were able to separate out international travel, which tends to be associated with higher SES, and examine domestic acquisition separately. The relationship between salmonellosis and SES has been examined in several other countries including Denmark, England, and Italy [23–26]. The studies in Denmark and England found an overall association of increased incidence with higher SES that appeared to be driven by S. Enteritidis [24, 25]. In Denmark, the reverse association was found with S. Typhimurium, whereas in England no association between SES and S. Typhimurium was found. Of interest, cases acquired through international travel were not excluded, age groups examined were much broader than in our study, and the studies covered a different time period (1993 [25] and 1993–2004 [24]). In the English study, travel abroad in the week before symptom onset increased risk of salmonellosis by >40-fold, and the authors of the Danish study concluded that the differences between SES groups could be at least partially explained by differences in diet and travel activity.

We found that S. Enteritidis cases had the highest proportion with international travel (18%) and the lowest domestic IRR of highest to lowest census tract poverty groups (1.08), and that when international travel cases were included, the relationship between incidence and poverty reversed. Thus, it is possible that, had international travel-associated cases been excluded in these studies, a different result would have been found. The Italian study was a case-control study in children 0–14 years of age, limited to a single geographic area of Italy, and the SES measure was defined as the father being unemployed or an unskilled blue collar worker [26]. Among risk factors found in multivariate analysis, low SES increased risk, as we found, whereas breastfeeding decreased it. In the United States, breastfeeding is associated with higher SES [27], and this may have contributed to the especially high risk in children <5 years old living in higher-poverty census tracts.

Two age groups stood out in our analysis: <5-year-olds, who consistently had the highest SES-related risk, and 18–49-year-olds, who had no SES-related risk. Previous analyses of surveillance data by SES have not reported on whether the SES-related risk varied by age group. A case-control study of sporadic salmonellosis in infants <1 year old in the United States conducted in FoodNet sites identified a number of environmental risk factors contributing to their higher risk of salmonellosis, including having reptiles in the household, having ridden in a shopping cart next to meat or poultry, attending daycare with a child with diarrhea, and having consumed concentrated liquid infant formula in the 5 days before illness [28].

Whether these risk factors and/or ingestion of risky foods are more common among infants and children <5 years old living in high-poverty areas is unknown and a question open to future study. However, in this case-control study, as in the Italian study [26], breastfeeding, associated with higher SES, was protective. That SES was not associated with salmonellosis in persons 18–49 years old but was associated with it in other age groups suggests differential types of exposure by SES by age. Future studies of the relationship between salmonellosis or its risk factors and SES need to be designed to account for age in a nonlinear fashion.

Although combined findings from all FoodNet sites demonstrated a clear association between domestic salmonellosis and higher census tract poverty level that was consistent across racial/ethnic groups and leading serotypes, the findings were not consistent by FoodNet site: in half the sites, this relationship was not found. Possible hypotheses are that some types of exposure to Salmonella have different relative frequencies in different states, especially exposures that are SES related. In addition to domestic exposures that are likely to be more frequent among persons of lower SES, exposures clearly exist that are related to higher SES. In these same 10 FoodNet sites during overlapping time periods, higher incidences of Campylobacter and of Shiga toxin-producing E. coli were associated with higher SES [6, 11, 13], although for Campylobacter, the reverse association was found in children <10 years old when age-specific associations were examined [6], and when examined by FoodNet site, the overall Campylobacter association with higher SES was inconsistent.

The current study has several strengths, as well as important limitations. Strengths include the use of 7 years of data from the national sentinel foodborne disease surveillance system, the use of census tract–level poverty rather than a larger, more heterogeneous geographic area to define SES, analysis of individual serotypes in addition to analysis of all serotypes combined, and the exclusion of international travel–acquired cases.

The study also has a number of important limitations. First, this analysis is of laboratory-confirmed Salmonella cases only; hence, findings may not be applicable to all Salmonella infections in the United States. However, to the extent that there is a potential bias for those of higher SES to seek medical care for salmonellosis and to be tested, the rates in those of higher SES would be biased toward being higher than rates in those of lower SES. This would make our findings a low estimate of the association of higher incidence with lower SES. Second, census tract poverty is an area-based measure; it does not directly measure individual-level poverty. Thus, the findings can be interpreted from 2 perspectives: using the area-based measure as a surrogate for individual poverty and/or thinking of it as potentially measuring neighborhood factors that influence individual exposure risk [3]. The potential for there being an ecologic fallacy (eg, the excess relative risk in poorer census tracts occurs among those of high individual SES) cannot be ruled out, especially to explain possible differences between FoodNet sites.

Third, the assumption used to interpret missing travel history information could be wrong. Importantly, missing international travel information was much more common among persons from higher-poverty (lower-SES) census tracts (35% in the lowest-SES vs 13% in the highest-SES census tracts). Because there would have been a systematic bias toward much lower rates in persons living in low-SES census tracts compared with those in high-SES census tracts, resulting from excluding cases with missing information, we attempted to minimize this bias by imputing travel history. To the extent that those with missing information might have differed from those with information available, our assumptions might have been inaccurate. Although we were able to control for age group and SES stratum in imputing travel, there may have been other factors that we were unable to control for. Finally, data from 10 FoodNet sites were combined: there was no attempt to weight the relative contribution from each site to the United States as a whole.

In summary, the findings from this FoodNet analysis of population-based data on domestically acquired, laboratory-confirmed salmonellosis show that, in general, children and older adults living in higher-poverty neighborhoods are at higher risk of acquiring Salmonella infection overall and with each of the 10 most common serotypes. Because these findings varied by FoodNet site and were not found in adults 18–49 years old, risk factors for salmonellosis should be identified by age group and location in order to focus future prevention efforts. In the meantime, current salmonellosis prevention efforts should include all SES groups, with emphasis on young children living in higher-poverty areas.

Notes

Acknowledgements. We thank the many FoodNet surveillance staff at each site and at the US Centers for Disease Control and Prevention (CDC) for their work in collecting, geocoding, and collating the data used in this study.

Disclaimer. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the CDC.

Financial support. This work was supported by the CDC (grant 5 NU50CK000488-03-00).

Potential conflicts of interest: All authors: No potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Presented in part: Council of State and Territorial Epidemiologists Meeting, Raleigh, North Carolina, 2–6 June 2019. Abstract 10758.

References