Salmonella is a common cause of foodborne illness in the United States and often is linked to chicken products. Salmonella contamination has been associated with meat processing facility characteristics, such as the number of employees (i.e., hazard analysis critical control point [HACCP]–based definition of size). The risk factors for Salmonella contamination in U.S. poultry have not been evaluated since implementation of the New Poultry Inspection System (NPIS) in 2014. The goal of this study was to determine whether risk factors for Salmonella contamination changed after implementation of the NPIS. Presence or absence of Salmonella in whole chicken carcasses was modeled using microbiological testing data collected from 203 poultry processing establishments by the U.S. Department of Agriculture Food Safety and Inspection Service between May 2015 and December 2019. A model was fit using generalized estimating equations for weekly presence or absence of Salmonella, and production volume, geographic location, and season were included as potential covariates among other establishment demographics. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated from the marginal model. Of the 40,497 analyzable samples, 1,725 (4.26%) were positive for Salmonella. Odds of contamination was lower among establishments slaughtering ≥10,000,000 birds per year (OR = 0.466; 95% CI, [0.307, 0.710]) and establishments producing ready-to-eat finished products (OR = 0.498; 95% CI, [0.298, 0.833]) and higher among establishments historically (previous 84 days) noncompliant with HACCP regulations (OR = 1.249; 95% CI, [1.071, 1.456]). Contamination also significantly varied by season and geographic region, with higher odds of contamination during summer and outside the MidEast Central region. These results support continuation of targeted food safety policies and initiatives promoting pathogen reduction by establishments with smaller volumes and those noncompliant with HACCP regulations.
Low production volume and summer season are risk factors for Salmonella contamination.
Salmonella contamination varied by region, even after adjusting for seasonality.
Facilities producing ready-to-eat product have lower odds of contaminated raw product.
HACCP noncompliance (9 CFR §417.4) was associated with Salmonella contamination.
Salmonella is a common cause of foodborne illness in the United States, accounting for approximately 1.02 million illnesses, 20,000 hospitalizations, and 400 deaths annually (18). Three of the most common Salmonella serotypes associated with human illness (Enteritidis, Newport, and Infantis) are frequently recovered from poultry. Approximately 196,000 illnesses are linked to chicken consumption annually, resulting in annual economic losses of $1.1 billion to $2.8 billion (2, 19, 20). These cost estimates do not include government or industry costs such as those associated with investigations, recalls, and legal action (17, 26). Given the magnitude and scope of this Salmonella issue, targeted efforts regarding chicken and other poultry products could significantly reduce the burden of salmonellosis.
Identification of risk factors for Salmonella contamination of food is an important step for reducing disease burden, and mitigating these risk factors will likely have the greatest impact. Risk factor research often focuses on preharvest conditions (e.g., rearing conditions (8) and feed withdrawal times (10)), but such mitigation strategies can be difficult to implement industry-wide due to the complexity of pathogen control in animal agriculture (1). Mandatory control of pathogens already exists in processing facilities, and industry-wide mitigation strategies may be easier to implement if they target specific risk factors. Further, food safety practices are driven in large part by processing facility characteristics, such as the number of employees or the amount of product produced (4, 12).
In previous studies, the U.S. Department of Agriculture (USDA) Food Safety and Inspection Service (FSIS) has evaluated the relationship between establishment characteristics and Salmonella contamination (6, 11). However, these studies were conducted before implementation of major changes to regulatory oversight of poultry processors (Table 1). In 2014, the FSIS began implementing the New Poultry Inspection System (NPIS), which was designed to focus FSIS efforts on inspection activities more directly related to public health outcomes such as process control systems for Salmonella (28). Major changes to sampling protocols for performance standard verification have also been implemented. In previous risk factor studies, data from set-based sampling (i.e., 51 samples collected in an establishment over consecutive days) were used. However, in 2015 the FSIS (29) adopted a continuous sampling approach (i.e., approximately one sample collected per establishment per week on a continuous basis). In 2016, the FSIS (30) also began collecting carcass rinse samples with neutralizing buffered peptone water (nBPW) to neutralize the antimicrobial chemicals that are widely used in the poultry industry but negatively impact recovery of Salmonella from samples (7).
Given these recent changes in sampling and regulatory oversight of poultry production, a reevaluation of risk factors for Salmonella contamination is warranted. The purpose of this study was to provide such an update by identifying risk factors for Salmonella contamination of whole chicken carcasses, such as establishment characteristics and inspection compliance history, through an evaluation of Salmonella-positive test results among FSIS Salmonella performance standard verification data.
MATERIALS AND METHODS
Risk factors for Salmonella contamination of whole chicken carcasses were evaluated based on Salmonella performance standard verification data from the Raw Chicken Carcass Sampling Program obtained through the FSIS Web site; data on establishment characteristics obtained from the FSIS Meat, Poultry and Egg Product Inspection (MPI) Directory; and FSIS inspection and noncompliance data for individual establishments obtained through a Freedom of Information Act request to FSIS.
The primary outcome of interest was presence or absence of Salmonella in individual samples collected by FSIS inspectors. Based on previous studies and preliminary analyses, the primary risk factors considered were year, season, region (as defined by FSIS Office of Field Operations District), whether the sample was collected in nBPW, annual slaughter volume, product class(es) slaughtered by establishment (e.g., turkey and other poultry), monthly processing volume, type of finished product (e.g., raw intact and raw nonintact), and noncompliance with individual regulations in Title 9 of the Code of Federal Regulations (21).
To address data sparsity issues, several variables were dichotomized (Table 2). Slaughter volume was dichotomized to indicate whether an establishment slaughtered ≥10,000,000 birds per year (i.e., the largest of five slaughter volume categories assigned by FSIS within the MPI Directory). The FSIS Salmonella verification testing program (25, 28, 29) includes a routine sampling approach in which approximately five and two sample units (carcass rinses) are collected per month from higher volume establishments (i.e., ≥5,000,000 lb [2,267,962 kg] of product per month) and lower volume establishments (i.e., <5,000,000 lb of product per month), respectively. To align with these sample allocation practices, monthly processing volume (calculated by multiplying mean daily processing volume by the number of production days per month) was dichotomized to indicate a high or low processing volume. Dichotomous variables were also created to indicate hazard analysis critical control point (HACCP) product categories produced by each establishment.
Establishment compliance was determined for 146 individual regulations by combining data on scheduled and performed inspection tasks with noncompliance records issued to establishments. To align with the time frame used in determination of public health regulations (PHRs) by the FSIS (32), a variable indicating compliance, noncompliance, or nonobservance in the 84 days preceding sample collection was created for each regulation (Supplemental Table S1).
Because seasonality has been associated with Salmonella contamination, season was assigned for each observation using week definitions from the Morbidity and Mortality Weekly Report: winter, weeks 48 through 52 and 1 through 8 (roughly December through February); spring, weeks 9 through 21 (roughly March through May); summer, weeks 22 through 35 (roughly June through August); and autumn, weeks 36 through 47 (roughly September through November). To limit the impact of potential uncontrolled confounders, establishments for which demographic data were missing were excluded from analyses. A complete list and description of variables in the analytical data set are provided in Table S1.
Statistical analyses and model development
Descriptive statistics were estimated for establishment characteristics, microbiological results, and inspection data. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated for potential risk factors in the analyses. Appropriateness of the multivariable model was evaluated using methods described by Preisser and Qaqish (15). All analyses were conducted in SAS 9.4 (SAS Institute, Cary, NC), and code for the multivariable risk factor model is provided in Supplemental Material S2.
A generalized estimating equation (GEE) approach was used to develop a multivariable risk factor model for Salmonella contamination of whole chicken carcasses (16, 37). Although the GEE approach to statistical modeling requires no assumptions about the distribution of data beyond specifying an appropriate link (e.g., logit for binary data), a pairwise chi-square analysis of establishment demographic factors (e.g., slaughter volume and processing volume) was conducted to test for independence of predictor variables and to inform the model building process. Variables were tested for association with a Salmonella-positive test result using a working independence correlation structure. Establishment characteristic variables were included in the multivariable model selection process when marginally associated (P < 0.20) with a Salmonella-positive result. Only 84-day historical compliance variables with <10% of observations classified as not observed and a marginally significant (P < 0.20) univariable association with a Salmonella-positive result were considered in the multivariable process.
The multivariable model was developed using a multistaged approach in a manner that minimized the quasi-likelihood under the independence model criterion statistic first proposed by Pan (14) for assessing GEE model fit (5). Ideally, all risk factors univariably associated with a Salmonella-positive result would be jointly modeled with backward selection of the final multivariable model. Nonconvergence of this model paired with a high number of not observed values among historical compliance variables necessitated a modified approach. To avoid potential model misspecification from not observed inspection data, an initial multivariable model containing only establishment characteristics and administrative information (e.g., year or season of sample collection) was developed through backward selection of variables marginally associated with a Salmonella-positive result. Historical compliance variables that met the inclusion criteria were then sequentially added through forward selection and kept in the multivariable model when statistically significant (P < 0.05) according to the Wald test.
Samples were collected from 225 establishments through the whole chicken carcass Salmonella verification testing program from 2015 through 2019, but 22 establishments were excluded from this analysis because of missing demographic factors. Of the remaining 203 establishments, 72.4% slaughtered ≥10,000,000 birds per year (high volume) and 27.6% slaughtered <10,000,000 birds per year (low volume) (Table 2). A total of 40,497 carcass rinse samples were collected and tested from the 203 establishments during the study period, and 1,725 (4.26%) were positive for Salmonella, although this rate differed by year and season (Fig. 1). Of the 1,722 Salmonella isolates that were serotyped, 872 (50.6%) were Salmonella Kentucky, 244 (14.2%) were Salmonella Enteritidis, 168 (9.8%) were Salmonella Infantis, 167 (9.7%) were Salmonella Typhimurium, and 271 (15.7%) were identified as other serotypes. These proportions of samples positive for Salmonella should not be interpreted as an industry-wide prevalence because verification sampling programs are not designed for prevalence estimation. A total of 7,487,102 inspection tasks were scheduled among the 203 establishments during the study period, resulting in a total of 193,664 noncompliance records.
Eight of the 13 establishment demographic factors considered in the univariable GEE model were significantly associated (P < 0.05) and 1 factor was marginally associated (P < 0.20) with a Salmonella-positive test result (Table 2). Year and sample collection method (i.e., in nBPW or not in nBPW) were significantly associated (P < 0.05) and season was marginally associated (P < 0.20) with a Salmonella-positive result (Fig. 2). Of the 146 historical compliance variables evaluated by univariable analysis, 30 were significantly associated (P < 0.05) and 25 were marginally associated (P < 0.20) with a Salmonella-positive result (Table S3). Only 22 of the 146 variables considered had a not observed value for <10% of observations; 13 of these were at least marginally associated with a Salmonella-positive result in the univariable analysis and were considered in the multivariable model (Table 3). Although exclusion of 124 regulation variables from further analyses was not ideal, doing so limits the potential for model misspecification, which could negatively impact generalizability of the results. Historical compliance variables considered in the multivariable model pertained to a wide range of regulations, including general labeling, product weight labeling, poultry product inspection, sanitation, and HACCP.
Backward selection of establishment demographic factors followed by forward selection of historical compliance variables resulted in a multivariable model with eight factors significantly (P < 0.05) or marginally (P < 0.10) associated with a Salmonella-positive result (Table 4). As expected, odds of contamination were higher for samples collected in summer than for those collected in winter. After adjusting for seasonal and establishment-specific characteristics (e.g., annual slaughter volume), region was also significantly associated with contamination. Higher odds of contamination were identified among establishments located in the South West (District 05), South Atlantic (District 85), and the East North Central (District 50) regions of the United States compared with establishments in the MidEast Central (District 90) region. Only two inspection variables remained significant in the multivariable model: 9 CFR §417.4 (HACCP system validation, verification, and reassessment) and 9 CFR §317.2 (label definition and required features) (21). Further investigation revealed that compliance with 9 CFR §317.2 provided minimal improvement in model fit but led to exclusion of 2,265 observations (5.6% of all observations) due to not observed values. To avoid biasing other model estimates, this variable was excluded from the final multivariable model. Of all factors included in the model, the use of nBPW during sampling was by far the strongest predictor of a Salmonella-positive result.
As governmental regulatory agencies and food manufacturers work to continue improving the safety of meat and poultry products, our understanding of how different practices or factors impact the safety of such products also needs improvement. In response to recent major changes in U.S. regulatory oversight of poultry production, this study was conducted to reevaluate risk factors for Salmonella contamination of whole chicken carcasses. The results of this study (i) confirmed that establishment volume and season remain risk factors for Salmonella contamination; (ii) revealed new potential risk factors for Salmonella contamination; and (iii) revealed how accounting for correlation between observations can impact results.
Confirmation of prior risk factors
Following the major changes to regulatory oversight of poultry, establishment size remains a risk factor for Salmonella contamination. Eblen et al. (6) found that large establishments (as defined by the number of employees) had lower odds of failing to meet the Salmonella performance standard than did small establishments. Our study, in which establishment size was defined by slaughter and processing volume, revealed that high-volume establishments had significantly lower odds of a Salmonella-positive test result than did low-volume establishments. Although both studies found lower odds of contamination among larger establishments, the difference in how establishment size was defined in the two studies is significant. HACCP-based size could be indicative of processing capacity (e.g., greater volume requires more employees to process), but defining establishment size by either slaughter volume or processing volume provides a more accurate estimate of how much product is routing through an establishment. This approach would be especially important for poultry processing in which scalding, picking, and chilling of carcasses can lead to serial cross-contamination (3). Thus, the number of carcasses slaughtered and processed could truly impact contamination. Because of data limitations, size was defined in the present study by dichotomous indicators for slaughter and processing volume; these two variables were highly correlated (results not shown), and only one (slaughter volume) was kept in the model by the backward selection process.
Regardless of how establishment size is defined, the higher odds of contamination among establishments with lower slaughter volume is logical. For example, facilities processing a larger volume are more likely to implement processing technologies that improve safety (e.g., inside-outside bird washer), microbiological testing, and food safety trainings for employees (4, 12). In contrast, lower volume facilities rely more on sanitation practices and modification to daily operations to maintain food safety (12). Extending this work, an index for food safety technology among meat and poultry plants was developed, and smaller volume facilities had lower technology indices than did larger volume facilities and facilities owned by multiplant firms (13). This higher technology index was also observed among meat and poultry plants that were subject to food safety auditing (i.e., third-party food safety evaluations).
Season also remained a risk factor; the odds of Salmonella contamination were higher in the summer than in winter. Poultry contamination typically follows a seasonal pattern (35). Further temporality was seen with decreasing odds of contamination across years, even after adjusting for other variables. The FSIS started using nBPW for collecting samples in 2016 because of the negative impact of antimicrobial chemicals used in poultry processing on Salmonella recovery from carcass rinses (7, 36). Our results were consistent with these findings; the odds of testing positive for Salmonella were significantly higher for samples collected in nBPW. However, the introduction of nBPW did not account fully for the temporal trends in contamination. After adjusting for nBPW use, significantly higher odds of contamination were found for 2016 and 2017 samples than for samples from 2019. Although results from this study cannot be used for causal interpretation, we hypothesized that the apparent increase in Salmonella contamination following introduction of nBPW encouraged establishments to implement additional control measures that lead to contamination reductions over time. The FSIS continues to collect samples under the Salmonella verification testing program, and follow-up evaluations of data collected after 2018 could provide more insight into temporal trends and identify additional risk factors that may have been masked by the change to nBPW.
Identification and reevaluation of risk factors
Region and type of finished product were identified as new risk factors for Salmonella contamination. In contrast to previous studies, region (as defined by FSIS district) was significantly associated with a Salmonella-positive result in the present study. The South West (District 05), South Atlantic (District 85), and East North Central (District 50) regions had significantly higher odds of contamination than did the MidEast Central (District 90) region. The MidEast Central region was selected as the reference group because it has the highest concentration of poultry processing establishments. These observed differences may have been due to differences in establishment characteristics or preharvest seasonal conditions, but because the differences remained after adjusting for seasonal conditions and establishment characteristics, region may represent an unmeasured risk factor. For example, region could be a surrogate measure for differences in management approaches between corporations and/or district offices. Given sustained interest in food safety culture and the impact of human behavior on food safety, collection of such data within regulatory contexts could facilitate inclusion of human factors and management approaches in future work on Salmonella contamination of meat and poultry products.
Finished product type, which was not considered in previous studies, was associated with Salmonella contamination in the present study. Lower odds of contamination were found for establishments that produce both raw and ready-to-eat (RTE) poultry products. Although larger establishments typically process a wider variety of products, slaughter volume and production of RTE products were not significantly correlated in this study (results not shown). Further exploration of how the 11 establishments producing raw and RTE finished products differed from other establishments could identify future areas for targeted improvement in Salmonella reduction.
One area that might help explain the inverse association between Salmonella contamination in raw product and regulation for RTE products is the establishment's HACCP system. In this study, the odds of a Salmonella-positive test result among establishments noncompliant with any component of HACCP system validation, verification, and reassessment (9 CFR §417.4) (22) in the 84 days before sample collection were higher than those for establishments that were compliant. Noncompliance with 9 CFR §417.4 could indicate a systematic process control issue for finished products, and additional exploration of the relationship between HACCP noncompliance and Salmonella contamination is warranted.
None of the sanitation regulations univariably associated with a Salmonella-positive result remained significant in the final multivariable model, even though an association between sanitation noncompliance and Salmonella control performance was found by Muth et al. (11). These authors defined compliance as the percentage of establishments compliant for the specified inspection task code. In contrast, the present study used a categorical variable indicating 84-day historical compliance, noncompliance, or not observed status for individual regulations. The approach used by Muth et al. was, until recently, fairly common and used in risk assessments conducted by the FSIS (24, 27). However, our approach aligns more closely with current methods used by the FSIS to determine PHRs for meat and poultry processing on an annual basis (32) and, thus, identifies risk factors that could be more easily targeted through policy initiatives.
GEE approach to risk factor analyses
Regarding interpretation of our results in the context of previous work, odds ratios presented here are not directly comparable to those in previous risk factor studies conducted with U.S. regulatory data. Risk factor analyses conducted by Eblen et al. (6) and Muth et al. (11) included a different outcome measure: a dichotomous variable indicating whether an establishment met a given level of the Salmonella performance standard based on a set of 51 samples. This difference is important because there are multiple ways to address correlation between observations made in the same establishment. In previous studies, a derived-variable approach was used in which correlated observations (e.g., 51 samples collected within the same establishment) were collapsed into a single observation and then analyzed using independent data methods such as logistic regression. In contrast, a GEE approach was used in the present study in which correlation between observations was addressed by postestimation adjustments to model parameter standard errors (9, 34, 37). This approach maintains a larger effective sample size (e.g., 40,497 observations instead of the 203 observations obtained with the derived-variable approach), which allows for a more granular analysis. Muth et al. (11) noted that analysis of regulatory data is often limited by the need for large data sets. As revealed in the present study, this issue can be resolved by using a GEE approach to identify risk factors.
Several limitations to the present study were identified. Twenty-two establishments were excluded from the analytical data set due to missing information for demographic variables. Missing data (e.g., no longer active or multiple establishment names or identifications associated with an address) could have resulted from a lack of Salmonella control within the establishment. Although the primary goal of this study was to identify risk factors following implementation of the NPIS, this study was not designed to assess the direct impact of the NPIS on Salmonella contamination within the poultry industry. The present study criteria for determining eligibility of historical inspection variables were less restrictive than those used in other studies. When determining PHRs, the FSIS requires that a regulation be verified at least 30 times within the historical window for inclusion in analyses; this stipulation is meant to reduce the potential for biased values. Such a restriction was not applied in the present study because the vast majority of regulations would have been excluded. Instead, the researchers addressed not observed values by including only those historical variables with <10% of observations being not observed.
Although meat and poultry safety has improved since implementation of the pathogen reduction HACCP rule in 1996 (22), Salmonella remains a major concern. The proposed Healthy People 2030 goals continue to target a reduction in outbreaks of Salmonella infection (among other foodborne pathogens) linked to poultry (33). Future policies and initiatives targeting the risk factors identified by our GEE approach could reduce Salmonella contamination of whole chicken carcasses and contribute to meeting the Healthy People 2030 goals. Predictive modeling of information collected on a regular basis, such as compliance with regulations as measured during daily inspection activities, could inform directed real-time interventions for controlling Salmonella contamination. Such an approach would likely be limited by not observed values among compliance data; therefore, future work in this area should explore alternative ways to model historical compliance such as use of lagged and latent variables.
The authors thank Isabel Walls, Eric Ebel, Rebecca Fields, Selena Kremer, and Breauna Branch for providing input during the modeling process and reviewing this manuscript. This research was supported by The Ohio State University and an appointment to the FSIS Research Participation Program administered by the Oak Ridge Institute for Science and Education (ORISE) through an interagency agreement between the U.S. Department of Energy (DOE) and the National Institute of Environmental Health Sciences. ORISE is managed by Oak Ridge Associated Universities (ORAU) under DOE contract DE-SC0014664. All opinions expressed in this article are the authors' and do not necessarily reflect the policies and views of USDA, FSIS, DOE, ORAU, or ORISE.