Analysis of Food Safety Performance in Meat and Poultry Establishments

Transcription

1 November 2007 Analysis of Food Safety Performance in Meat and Poultry Establishments Revised Final Report Contract No. 53-3A , Task Order 18 Prepared for Flora Tsui James Wilkus U.S. Department of Agriculture Food Safety and Inspection Service 1400 Independence Avenue SW Washington, DC Prepared by Mary K. Muth Mansour Fahimi Shawn A. Karns Yan Li RTI International Health, Social, and Economics Research Research Triangle Park, NC RTI Project Number

2 RTI Project Number Analysis of Food Safety Performance in Meat and Poultry Establishments Revised Final Report Contract No. 53-3A , Task Order 18 November 2007 Prepared for Flora Tsui James Wilkus U.S. Department of Agriculture Food Safety and Inspection Service 1400 Independence Avenue SW Washington, DC Prepared by Mary K. Muth Mansour Fahimi Shawn A. Karns Yan Li RTI International Health, Social, and Economics Research Research Triangle Park, NC RTI International is a trade name of Research Triangle Institute

3 Preface In September 2005, the U.S. Department of Agriculture s Food Safety and Inspection Service (USDA, FSIS) contracted with RTI International to conduct an analysis of food safety performance in meat and poultry slaughter establishments using available FSIS data sets. 1 RTI developed a work plan for the project that incorporated comments and input from FSIS on the needs for the analysis. Based on the tasks outlined in the work plan, we conducted the analyses described in this report. In this preface, we discuss the background for the study and the objective of the analysis and provide an overview of the remainder of the report. 1. BACKGROUND Data from recent industry surveys conducted by FSIS were combined with existing FSIS databases to conduct a unique analysis of the factors affecting food safety performance in meat and poultry slaughter establishments. Recent research based on 2000 data has shown that an establishment s Hazard Analysis and Critical Control Points (HACCP) plan compliance, certain establishment characteristics, and its use of particular slaughter processing practices and technologies can be reasonably reliable indicators of that establishment s performance in pathogen control. Pathogen control is a major determinant of the food safety performance and the level of food safety risk associated with the establishment. This research provides an empirical basis for and general insights into the level of inspection resources that FSIS may allocate to various types of establishments under a risk-based system of inspection. 1 RTI will conduct additional analyses under a modification to the task order that will consider more recent data for young chicken slaughter establishments and help contribute to the development of a risk-based sampling process. 1

4 Analysis of Food Safety Performance in Meat and Poultry Establishments FSIS conducted surveys of meat and poultry slaughter and processing establishments in 2004 and of processing-only establishments in These more recent data may provide better estimates of current establishment food safety performance. The databases from these surveys and agency databases (e.g., Performance-Based Inspection System [PBIS], Enhanced Facilities Database [EFD], and Microbiological and Residue Contamination Information System [MARCIS]) can be used to provide improved estimates of establishments food safety performance and an empirical basis for measures the agency may consider in the transition to risk-based systems. The agency places a high priority on developing information from the 2004 and 2005 surveys and agency databases that can be used to inform regular and frequent agency decisions concerning inspection and testing activities. Survey data provide one source of external information available to the agency on which to make these decisions. FSIS anticipates that guidance for these decisions may be developed based on the analysis of these data. Several data analysis methods were implemented to determine key predictors of food safety performance from a large number of available variables. As outlined in the project work plan, RTI constructed analysis data sets for the following four categories: young chicken carcasses pork carcasses beef carcasses ground beef produced in cattle slaughter establishments RTI used a number of statistical methods to identify which relationships are most significant and which factors are strong predictors of food safety performance as indicated by Salmonella test results. Detailed analyses were conducted for young chicken carcasses and market hog carcasses; the results of these detailed analyses are the primary focus of this report. We also conducted limited analyses for beef carcasses and ground beef produced in cattle slaughter establishments because products tend to have very few positive Salmonella test results; thus, detailed analyses are less informative; the results of these limited analyses are provided in Appendix C. The analysis methods included 2

5 Preface development of a classification tree, including identification of the relative importance of the potential predictors of food safety performance; factor analysis to identify underlying themes or factors (i.e., combinations of predictors) associated with food safety performance; and logistic regression analysis to determine the relative contributions of the predictors to food safety performance. These analytical methods provide actionable results on which informed policy decisions can be made and provide information for allocating inspection program resources, including personnel. The approach developed for this task could also be applied to analyze food safety performance in processing-only establishments. FSIS also fielded a survey of processing-only establishments, and results of the survey could be used to construct a similar data set. Performance measures for processing-only establishments could include both Listeria test results and Salmonella test results. 2. OBJECTIVE The objective of this task order is to conduct statistical analyses of food safety performance in meat and poultry slaughter establishments. Detailed analyses were conducted for young chicken and pork slaughter establishments, and limited analyses were conducted for beef carcasses and ground beef produced in cattle slaughter establishments. 3. ORGANIZATION OF THE REPORT The body of this report contains the manuscript prepared under this task order. It includes an introduction, description of the data sets, overview of the methods, results of the analysis, and a discussion of the results. The manuscript is formatted for submission for publication. By submitting the manuscript for publication, it will be peer reviewed and the analysis can be improved based on the feedback received. In addition to the manuscript, Appendix A includes the data dictionary for meat establishments, and Appendix B includes the data dictionary for poultry establishments. Exploratory analyses conducted for beef carcasses and for ground beef 3

6 Analysis of Food Safety Performance in Meat and Poultry Establishments produced in cattle slaughter establishments are included in Appendix C. Finally, the presentation of the results of the analysis at FSIS is provided in Appendix D. 4

7 Running Head: Meat and Poultry Pathogen Control Performance Analysis of Pathogen Control Performance in U.S. Young Chicken Slaughter and Pork Slaughter Establishments Mary K. Muth, * Mansour Fahimi, Shawn A. Karns RTI International 3040 Cornwallis Road P.O. Box Research Triangle Park, NC October Key words: Salmonella, young chicken slaughter, pork slaughter, establishment performance, risk-based inspection * Mary K. Muth, RTI International, 3040 Cornwallis Road, P.O. Box 12194, Research Triangle Park, NC 27709; Voice: (919) ; Fax: (919) ; muth@rti.org 1

8 Analysis of Pathogen Control Performance in U.S. Young Chicken Slaughter and Pork Slaughter Establishments ABSTRACT In the 1996 U.S. Department of Agriculture, Food Safety and Inspection Service s (FSIS s) Pathogen Reduction; Hazard Analysis and Critical Control Point (PR/HACCP) Systems, Final Rule, Salmonella was selected for microbiological testing and monitoring. Using data from an FSIS-sponsored survey of meat and poultry slaughter establishments, inspection results, and other establishment characteristics, potential variables affecting pathogen control, as measured by Salmonella test results, were investigated. The analysis data sets included 153 federally inspected young chicken slaughter establishments, of which 111 failed half the Salmonella performance standard at least once from 2003 through 2005, and 121 federally inspected pork slaughter establishments, of which 28 failed half the Salmonella performance standard. Logistic regression results for young chicken slaughter establishments seem to indicate they were more likely to fail if they had higher inspection noncompliance rates (p =.10) and older production space (p =.10) and were less likely to fail if they used a higher percentage of raw poultry inputs purchased from outside sources (p =.10). Results for pork slaughter establishments indicate they were more likely to fail if they had a higher rate of voluntary microbiological testing (p =.08) and were less likely to fail if they were larger (p =.08) and used a higher percentage of raw pork inputs purchased from outside sources (p =.02). In general, indicators of plant characteristics, food safety practices, and management philosophy are associated with different levels of pathogen control performance that vary by species slaughtered. 2

9 The 1996 U.S. Department of Agriculture, Food Safety and Inspection Service (FSIS) Pathogen Reduction; Hazard Analysis and Critical Control Point (PR/HACCP) Systems, Final Rule sets Salmonella performance standards for establishments slaughtering selected classes of food animals or those producing selected classes of raw ground products (Federal Register, 2006). FSIS selected Salmonella for microbiological testing because it is a common bacterial cause of foodborne illness, it occurs often enough to be detected and monitored, methods are available for recovering Salmonella from a variety of meat and poultry products, and intervention strategies aimed at reducing Salmonella should be effective at reducing other foodborne pathogens (17). Testing began under the PR/HACCP rule in January 1998 for large establishments (500 or more employees), in January 1999 for small establishments (from 10 to 499 employees), and in January 2000 for very small establishments (9 or fewer employees or less than $2.5 in annual sales) (17). The performance standards for Salmonella testing vary by type of product, as indicated in Table 1. The sample set size varies from 51 to 82 samples, and the maximum positives varied from 1 to 29 depending on product type (18). Establishments that slaughter young chickens fail the performance standards with 13 positives out of 51 samples, and establishments that slaughter hogs fail the performance standard with 7 positives out of 55 samples. Recently, FSIS developed scheduling criteria such that establishments that fail half the performance standard (i.e., are at 51% of the performance standard or higher) during a sample set are tested more frequently than establishments that are at 50% of the performance standard or less (19). Thus, young chicken slaughter establishments that have 7 or more positives and hog slaughter establishments that have 4 or more positives are subsequently tested more frequently. 3

10 62 63 Table 1. Salmonella Sample Set Size and Performance Standard for Each Meat and Poultry Product Type Product Type Sample Set Size Maximum Positives Steer and heifer carcasses 82 1 Cow and bull carcasses 58 2 Ground beef 53 5 Market hog carcasses 55 6 Chicken carcasses Ground chicken Ground turkey Salmonella test results provide a measure of pathogen control in meat and poultry produced at slaughter and processing establishments and thus can be used to verify that industry systems are effective in controlling the contamination of raw meat and poultry products with disease-causing bacteria. Whether an establishment typically passes the Salmonella performance standard may depend on a variety of factors, including the characteristics of the establishment, use of food safety technologies, and management practices in the establishment. Recent surveys of food safety technologies and practices were conducted for FSIS to determine the prevalence of various food safety technologies and practices used in the U.S. poultry and meat slaughter industries, to characterize the use and types of microbiological testing, and to determine the prevalence of different types of employee food safety training (5, 6). The results of the survey 4

11 combined with other data sources were used to determine which factors are typically associated with better pathogen control performance as measured by Salmonella test results. The use of different types of food safety technologies and characteristics of slaughter establishments has been shown to affect pathogen levels on carcasses. In particular, the use of different types of carcass rinse solutions, carcass decontamination methods, and skin removal methods has been shown to reduce microbial carcass counts (3, 4, 8, 10, 12, 20). Line speeds within the establishment also affect pathogen levels on carcasses as shown for E. coli populations on beef carcasses (2). The use of online versus off-line reprocessing of poultry carcasses to meet zero fecal tolerance standards affects Campylobacter and Salmonella levels on poultry carcasses (11). Furthermore, the use of different carcass chilling methods (immersion versus air) affects Salmonella and Campylobacter incidence on broiler carcasses (15). In addition to using specific food safety technologies, other factors shown to affect pathogen levels on carcasses include location and size of the establishment (1). The purpose of this study was to identify young chicken and hog slaughter establishment characteristics, including the use of specific food safety technologies, associated with better pathogen control performance as measured by Salmonella test results. The analysis was conducted using data from the FSIS surveys of meat and poultry slaughter establishments, data on establishment characteristics from FSIS s databases and a private vendor, and data on inspection procedure results from FSIS. This study s results provide information useful for informing the development of a risk-based inspection system and for identifying establishments most likely to be in need of technical assistance for controlling pathogens. 5

12 MATERIALS AND METHODS Data Sets. The data sets used for the analyses were constructed from the following sources: FSIS s Enhanced Facilities Database, which contains production and economic data for fiscal year 2004 on establishment characteristics for all meat and poultry establishments under federal inspection from FSIS s Performance-Based Inspection System (PBIS) database, FSIS s Animal Disposition Reporting System (ADRS), and a private data vendor, infousa (13); results of FSIS s meat and poultry slaughter establishment survey, which was conducted using nationally representative sampling procedures and includes data on food safety technologies and practices for all meat and poultry establishments that completed the survey in late 2004 and early 2005 (5, 6); FSIS s Salmonella test results for all establishments that were tested in FY 2003, FY 2004, and FY 2005; and FSIS s noncompliance report data from the PBIS database for facility, sanitation, and HACCP inspection procedures conducted in Data from each of the databases were matched based on FSIS establishment number to construct a complete record for each establishment. The variables in the data set include Salmonella test results for each product produced in the establishment; self-reported establishment responses to survey questions on food safety practices during slaughter and further processing, voluntary microbiological testing practices, employee food safety training, and other characteristics of the establishment; economic variables such as number of employees, annual sales, number of other establishments owned by the company, and slaughter volumes by species; and results of 6

13 inspection procedures conducted by FSIS inspectors. All of the data pertain primarily to performance and activities in 2004 with the exception of the Salmonella test results, which pertain to FY 2003, FY 2004, and FY We constructed the outcome measures (dependent variables) for the analyses based on the results of Salmonella testing for the first sample set for each establishment. Results from the second and third sample sets, if they were conducted, were not included because most establishments pass the first set and thus only have data for the first set and because subsequent sample sets are not representative of typical operating practices of the establishment since establishments make adjustments after failing the first set. We included test results for 2003 through 2005 to allow for a greater number of test sets to be represented in the data. However, by including multiple years of test results, we assumed that establishment characteristics over 3 years time are relatively constant. To correspond to FSIS s new scheduling criteria for Salmonella testing, we classified establishments as failures for the purposes of the analysis if they failed half the performance standard at least once during the 2003 through 2005 time period. We constructed the set of predictors (independent variables) in the analysis data sets using the remaining variables contained in the data sets. All variables that could be potential predictors of pathogen control performance were retained. Data preparation steps were conducted to logically impute missing values from the survey responses and eliminate redundant variables from multiple data sources. In addition, regulatory compliance variables were constructed from inspection procedure data from PBIS. Specifically, we defined facility, sanitation, and HACCP compliance variables by dividing the number of inspection procedures 7

14 that resulted in a noncompliance report by the total number of inspection procedures performed under each category of procedures. As indicated in Table 2, the resulting data set for young chicken slaughter establishments included 153 establishments, of which 111 failed half the Salmonella performance standard at least once from 2003 through The resulting data set for pork slaughter establishments included 121 establishments, of which 28 failed half the Salmonella performance standard. The young chicken slaughter data set represents a random sample of approximately 70% of the population of federally inspected establishments that slaughter young chickens; these establishments slaughtered 82% of the volume of young chickens slaughtered in The pork slaughter data set represents a random sample of approximately 20% of the population of federally inspected establishments that slaughter hogs; these establishments slaughtered 81% of the volume of market hogs slaughtered in Establishments not included in the data sets are those that were not surveyed or were not subject to Salmonella testing during the time period of the data. We used all inspection task results, including those identified as a review of a single critical control point (CCPs) or a review of all CCPs and those that were scheduled and unscheduled. We defined facility compliance using inspection code 06D01, which addresses whether establishment facilities (e.g., lighting, ventilation, and plumbing), equipment, and premises meet regulatory requirements; sanitation compliance using inspection codes 01B01, 01B02, 01C01, and 01C02, which address whether establishments conduct procedures before and during operations to prevent direct product contamination or alteration; and HACCP compliance using inspection codes 03B01, 03B02, 03C01, 03C02, 03J01, and 03J02, which address whether establishments are monitoring and recording CCPs to ensure compliance with critical limits and are performing verification activities. The establishments that slaughter market hogs also slaughter a small percentage of culled sows. On average, about 5% of their slaughter volumes comprise culled sows. 8

15 Table 2. Total Numbers of Establishments in the Analysis Data Sets and Number of Establishments that Failed Half the Performance Standard for Salmonella Number of Slaughter Number of Number that Failed Establishments in Establishments in Half the Performance Analysis Category Each Category Analysis Data Set a Standard Young chicken slaughter establishments Pork slaughter establishments a Establishments were excluded from the analysis data set if they were not subject to Salmonella testing in 2003 through 2005 or were not surveyed. Analysis Methods. Before conducting the analysis, we took two steps to further prepare the data sets. First, some variables, particularly those related to sales and slaughter volume, exhibited a great deal of variability across establishments. Thus, continuous variables with a large degree of variability were stabilized using a natural log transformation. Second, a few measures that depend on (i.e., increase with) the size of the establishment were normalized with respect to their corresponding establishment size. For example, the number of production employees who have completed HACCP training should be compared relative to the size of the establishment. Thus, these types of variables were expressed as a proportion to the total number of employees. The data sets contained a large number of potential predictor variables (66 variables for young chicken slaughter establishments and 89 variables for pork slaughter establishments). Thus, the classification and regression tree (CART) procedure was used to identify potential 9

16 predictors of the Salmonella outcome measure (Salford Systems, Inc.). Specifically, for each data set, a classification tree was constructed to show the partitioning of establishments based on variables that had an effect on Salmonella control performance. Using this exploratory investigation, the relative importance of the highest ranking predictors was established based on their cumulative contribution in classifying establishments. The relative importance of the variables was used to identify a smaller set of key predictors used in the subsequent analyses. Multicollinearity is a common problem in statistical modeling and arises when predictor variables are correlated. Because most statistical techniques, particularly regression-based procedures, rely on the assumption that all predictor variables are independent, inclusion of correlated data can result in highly unstable results. To address this problem in the analyses, we used factor analysis to develop predictors of Salmonella control performance that are orthogonal (independent) to one another. Furthermore, factor analysis enabled us to identify sets of underlying factors, or themes, among the predictor variables. Factor analysis can be conducted using continuous variables but does not incorporate binary variables; however, binary variables were incorporated in the final analysis step. Using the resulting themes and key binary variables identified in the CART analyses, we estimated logistic regressions to determine the magnitude and direction of the effects of the key variables on Salmonella control performance. The logistic regressions were first applied using only the factors obtained in the factor analysis for young chicken slaughter and pork slaughter establishments. Then, the logistic regressions were reapplied, including not only the factors but also the key binary predictor variables that were identified in the CART analysis to determine whether these variables might provide additional explanatory power not already captured by the factors. 10

17 For young chicken slaughter establishments, we also included a binary variable indicating whether the establishment operates under a HACCP-based inspection models project (HIMP) because FSIS was interested in knowing whether HIMP establishments performed better or worse on Salmonella performance when controlling for other variables that might affect establishment performance. Under HIMP inspection, establishment employees carry out the sorting activities previously conducted by FSIS inspectors to determine which carcasses and parts are unacceptable because they are diseased or unwholesome (7). A total of 17 establishments operating under HIMP inspection were included in the young chicken slaughter data set. In summary, the analyses for each data set were conducted in the following steps: Step 1. Transformed variables and prepared data sets for the analysis Step 2. Identified important predictor variables of Salmonella control performance using CART classification trees Step 3. Identified key themes, or clusters of variables, for Salmonella control performance using factor analysis Step 4. Estimated the magnitude and direction of effects of the key themes and binary variables on Salmonella control performance using logistic regression RESULTS The results of the series of analyses for young chicken slaughter and pork slaughter establishments are described below. Young Chicken Slaughter Establishments. Figure 1 shows the top nodes of the CART classification tree for young chicken slaughter establishments. The strongest single predictor was the percentage of production space that is 20 years old or older. Establishments with less than 11

18 Figure 1. Classification Tree for the Salmonella Outcome Variable for Young Chicken Slaughter Establishments ND = not defined (cannot be calculated for 100% failure rates) % of their production space that is 20 years old or older were more likely to pass half the Salmonella performance standard (odds ratio = 3.1). Of establishments with younger production space on the left branch of the tree, establishments with a sanitation noncompliance rate below 5% were more likely to pass (odds ratio = 3.5). Of establishments along the far left branch of the tree, age of the establishment production space again emerged as an important predictor. Establishments with less than 28% of their production space that is 20 years old or older were less likely to pass (odds ratio = 0.2), perhaps indicating an effect of less experienced establishment management. Establishments that have a sanitation noncompliance rate above 5% and do not use chemical sanitizers on hand tools all failed. On the right branch of the tree, which includes establishments with more than 56% of their production space 20 years old or older, establishments were less likely to pass if they are owned by a company that owns three or fewer other establishments (odds ratio = 0.1). Of The breakpoint of 56% was determined by the CART procedure through a search algorithm that identifies the value of the variable that splits the parent node into the most homogenous child nodes. 12

19 establishments owned by a company that owns three or fewer other establishments, establishments were subsequently more likely to pass if their sanitation inspection noncompliance rate was less than 11% (odds ratio = 5.3). On the far right branches of the tree, establishments owned by a company with four or more establishments and with sales exceeding $178 million (natural log of 19) all failed half the Salmonella performance standard. Thus, although on average establishments owned by larger companies performed better, the largest establishments owned by larger companies performed worse. Table 3 provides the results of the factor analysis, including all continuous predictor variables with a nonzero relative importance level as indicated by the CART analysis. The set of underlying themes, which comprise subgroups of the predictor variables, emerged in the factor analysis based on the factor loadings for each set of variables. These themes included measures of establishment size (number of establishments owned by the company, slaughter volume, number of employees, and dollar sales volume), employee training and quality assurance, inspection noncompliance rates (facility and sanitation inspection procedures), percentage of poultry processed using raw poultry received or purchased from another establishment (owned by the same or a different company), and age of the production space in the establishment. These five factors were then used as regressors in the logistic regression analysis. Table 4 presents two sets of logistic regression results for young chicken slaughter establishments. In the first set on the left, only the factors identified in the factor analysis were included in the regression. In the second set on the right, the same set of factors was included along with binary variables that had high relative importance levels in the CART analysis results

20 Table 3. Factor Analysis Results for the Young Chicken Slaughter Establishment Salmonella Outcome Measure Variable Factor 1. Factor 2. Factor 3. Factor 4. Factor 5. Establish- Training & Noncompli- Use of Raw Age of ment Size QA ance Rates Inputs Production Space Inspected establishments owned Young chickens slaughtered Employees at location 0.75 Sales at location Employees in QA department Employees trained HACCP Facility noncompliance rate Sanitation noncompliance rate Raw poultry imported 0.91 Raw poultry purchased % production space 20 years old or older Note: Factor loadings less than 0.4 are not shown. 14

21 Table 4. Results of Logistic Regression for the Young Chicken Slaughter Salmonella Outcome Measure Estimates with Factors Only Estimates with Factors and Key Binary Variables Variable Estimate P Value Estimate P Value (Standard Error) (Standard Error) Intercept (0.1919) < (0.7951) Factor 1: Establishment size (0.1821) (0.2270) Factor 2: Training & QA (0.1711) (0.1770) Factor 3: Noncompliance rates (0.1994) (0.2062) Factor 4: Use of raw inputs (0.2184) (0.2300) Factor 5: Age of production space (0.1936) (0.2049) Chemical sanitizers on hand tools (binary) (0.4788) NELS inspection (binary) (0.4992) HIMP inspection (binary) (0.6333) 15

22 Tracks products forward (binary) (0.6374) Likelihood ratio χ Note: NELS is the New Line Speed inspection system and HIMP is the HACCP-based Models Project inspection system. Based on the results of the logistic regression including only the factors, higher inspection noncompliance rates increased the likelihood that an establishment failed half the Salmonella performance standard (p =.10), a higher percentage of raw poultry inputs purchased from outside sources decreased the likelihood (p =.10), and use of older production space increased the likelihood (p =.10). The effects of establishment size and employee training were not significant, and their estimated coefficients indicated a much smaller magnitude of effect compared with the three significant factors. When the binary variables were also included in the regression, the p-values for some of the factors increased compared to the previous specification. Under this specification, only the variable indicating whether the establishment tracks products forward was significant at the 10% level. These results indicate that establishments that track their products forward are less likely to fail half the Salmonella performance standard (p =.06). Whether the establishment operated under HIMP did not have a statistically significant effect when controlling for other variables affecting Salmonella performance (p =.80). The coefficient for whether the establishment uses chemical sanitizers on hand tools had a large magnitude and indicated that use of chemical sanitizers increased the likelihood of failure, but the effect was not significant (p =.18). In addition, the coefficient for whether the establishment operates under New Line Speed (NELS) inspection had a large magnitude and indicated that establishments under NELS inspection were 16

23 less likely to fail compared with establishments under other types of inspection, but the effect was not significant (p =.14). Pork Slaughter Establishments. Figure 2 shows the top nodes of the CART classification tree for pork slaughter establishments. The strongest single predictor was whether the establishment produces ready-to-eat (RTE) products in addition to conducting slaughter activities. This initial result suggests that management philosophy of the establishment, as reflected in the complexity of the products produced, could affect Salmonella control performance. Establishments that do not produce RTE products were less likely to pass half the Salmonella performance standard (odds ratio = 0.2). Of the establishments that do not produce RTE products, establishments with a facility inspection noncompliance rate of less than 9% were more likely to pass (odds ratio = 8.5). Of the establishments on the far left branch of the tree that do not produce RTE products but have a low facility noncompliance rate, no establishments failed if they also produce processed not ready-to-eat (NRTE) products. ** For establishments that do not produce RTE products and have a facility noncompliance rate of 9% or more, none of the establishments that use bioluminescent testing on carcasses failed. This result may indicate these establishments increased voluntary testing in response to inspection results in the past. On the right branch of the tree, which indicates establishments that produce RTE products, there was essentially no effect of the facility noncompliance rate on the likelihood of passing (odds ratio < 0.1). ** Note that some establishments produce only products for further processing and thus produce neither RTE or NRTE products as defined in the survey. 17

24 Figure 2. Classification Tree for the Salmonella Outcome Variable for Pork Slaughter Establishments ND = not defined (cannot be calculated for 0% failure rates) Table 5 provides the results of the factor analysis including all continuous predictor variables with a nonzero relative importance level as indicated by the CART analysis. The set of underlying themes, which comprise subgroups of the predictor variables, emerged in the factor analysis based on the factor loadings for each set of variables. These themes included measures of establishment size (number of employees, dollar sales volume, square footage, HACCP size category, and number of QC/QA employees), inspection noncompliance rates (facility, HACCP, and sanitation inspection procedures), frequency of microbiological testing before and after fabrication of carcasses, and percentage of raw pork processed using raw pork inputs received or purchased from another establishment (owned by the same or a different company). These four factors were then used as regressors in the logistic regression analysis. Table 6 presents two sets of logistic regression results for pork slaughter establishments. In the first set on the left, only the factors identified in the factor analysis were included in the regression. In the second set on the right, the same set of factors was included along with binary 18

25 316 Table 5. Results of Factors Analysis for the Pork Slaughter Salmonella Outcome Measure Variable Factor 1. Factor 2. Factor 3. Factor 4. Establishment Noncompli- Microtesting Use of Raw Size ance Rates Frequency Inputs Employees at location 0.91 Sales at location 0.90 Square footage of production space 0.89 HACCP size 0.88 Employees in QC/QA dept Facility noncompliance rate 0.80 HACCP noncompliance rate 0.74 Sanitation noncompliance rate Microbiological testing prior to fabrication Microbiological testing after fabrication Raw meat purchased 0.84 Raw meat imported Note: Factor loadings less than 0.4 are not shown. variables that had high relative importance levels in the CART analysis results. These additional binary variables included whether the establishment dehairs carcasses, produces RTE product, and produces NRTE processed product. Based on the results of the logistic regression including only the factors, larger establishments were less likely to fail half the Salmonella performance standard (p =.04), 19

26 Table 6. Results of Logistic Regression for the Pork Slaughter Salmonella Outcome Measure Estimates with Factors Only Estimates with Factors and Key Binary Variables Variable Estimate P Value Estimate P Value (Standard Error) (Standard Error) Intercept (0.2665) < (0.9517) < Factor 1: Establishment size (0.3172) (0.3498) Factor 2: Noncompliance rates (0.2377) (0.2866) Factor 3: Microbiological testing frequency (0.2248) (0.2398) Factor 4: Use of raw inputs (0.2839) Does not dehair carcasses (binary) Produces RTE product (binary) Produces NRTE processed product (binary) (0.3448) (0.7152) (0.7299) (0.6169) Likelihood ratio χ <

27 establishments with a higher frequency of microbiological testing were more likely to fail (p =.08), and establishments having a higher use of raw inputs were less likely to fail (p =.02). The result for frequency of microbiological testing may indicate that establishments conduct more testing if they have had problems with Salmonella control in the past. The effect of inspection noncompliance rates was not significant (p =.31), and the magnitude of the coefficient was smaller than for the significant factors. In terms of the relative magnitude of significant variables, the frequency of use of raw pork inputs from outside sources and establishment size had similar magnitudes, and the frequency of microbiological testing had a smaller but still relatively large effect. When the binary variables were also included in the regression, the p-values for some of the factors increased compared to the previous specification. Under this specification, establishment size (p =.01) and microbiological testing frequency (p =.10) were significant at the 10% level in addition to all three of the included binary variables. The results for the binary variables indicate that establishments that do not dehair carcasses were more likely to fail (p =.01). In contrast to the CART analysis results, establishments that produce NRTE processed products were more likely to fail (p =.03) as were establishments that produce RTE product (p =.11). The results for whether the establishment produces RTE or NRTE processed product indicate that controlling for other variables affecting establishment performance is important. That is, when conducting multivariate analysis that accounts for the effects of establishment size, frequency of microbiological testing, and frequency of use of raw inputs from outside the establishment, the effect of whether the establishment produces further processed products switched signs. 21

28 DISCUSSION The results of the analyses presented above indicate that characteristics of meat and poultry establishments and results of inspection procedures may be useful indicators of pathogen control performance under a more risk-based inspection system. However, the variables that are potential indicators are likely to vary across species in slaughter establishments. Among the large number of potential predictors of pathogen control performance, only one set of variables the proportions of raw inputs purchased from outside sources appeared to have a statistically significant association with better Salmonella control in both young chicken slaughter and pork slaughter establishments. Although the use of raw inputs is not directly associated with Salmonella test results on raw carcasses, a higher use of raw inputs may be a reflection of the use of more sophisticated production processes or management philosophy that results in better pathogen control. For young chicken slaughter establishments, another significant variable reflecting differences in management practices and philosophy was whether the establishment tracks products forward in distribution. Young chicken slaughter establishments that track products forward had better Salmonella control performance. For pork slaughter establishments, whether the establishment produces RTE or processed NRTE product was associated with a higher likelihood of failure, thus possibly reflecting differences in management philosophy regarding pathogen control on carcasses. Some of the variables that had a statistically significant effect on Salmonella control performance are indirect or direct indicators of the types of production technologies used in the establishment. For young chicken slaughter establishments, older production space, which likely also indicates older production technologies, was associated with a higher likelihood of failure. For pork slaughter establishments, larger establishments, which are likely to use more 22

29 sophisticated production technologies, were less likely to fail. Furthermore, pork slaughter establishments that do not dehair carcasses were substantially more likely to fail. However, use of voluntary microbiological testing in pork slaughter establishments was associated with a higher likelihood of failure; this result may indicate that establishments implemented testing to monitor performance in response to Salmonella test results in the past. Finally, inspection results appear to be associated with Salmonella control performance in young chicken slaughter establishments, but type of inspection system does not. Establishments under HIMP inspection appeared to be no more or less likely to fail; the result for this variable is similar to Cates et al. (7). However, young chicken slaughter establishments with a higher rate of noncompliance with inspection procedures were more likely to fail. The analyses presented in this paper are limited to some extent by the use of self-reported survey data and the fact that food safety technologies and practices used by establishments might have changed over the 3-year time period of the Salmonella test results data. In the future, more objective information on the use of food safety technologies and practices (e.g., if observed by a third party) might provide additional confidence in the results. Furthermore, although the pork slaughter data set represents a large percentage of the total industry slaughter volume, many establishments are not represented in the data set. Inclusion of a higher percentage of pork slaughter establishments might provide additional insights not captured by the establishments represented in the data. A larger data set for both poultry and pork slaughter establishments would also likely increase the statistical significance level of important variables in the models. Lastly, as establishments implement additional food safety technologies and practices and FSIS further refines its testing procedures and databases, a similar analysis approach could be 23

30 implemented with new data sets to provide the most current information for risk-based inspection initiatives. 24

31 ACKNOWLEDGEMENTS This study was conducted under U.S. Department of Agriculture, Food Safety and Inspection Service, Contract No. 53-3A , Task Order 18. The authors gratefully acknowledge comments and suggestions provided by FSIS staff. All remaining errors and views expressed are those of the authors and not of the U.S. Department of Agriculture. 25

32 REFERENCES 1. Altekruse, S. F., N. Bauer, A. Chanlongbutra, R. DeSagun, A. Naugle, W. Schlosser, R. Umholtz, and P. White Salmonella enteritidis in broiler chickens, United States, Emerg. Infect. Dis. 12(12): Bacon, R. T., K. E. Belk, J. N. Sofos, R. P. Clayton, J. O. Reagan, and G. C. Smith Microbial populations on animal hides and beef carcasses at different stages of slaughter in plants employing multiple-sequential interventions for decontamination. J. Food Prot. 63(8): Barboza de Martinez, Y., K. Ferrer, and E. Marquez Salas Combined effects of lactic acid and nisin solution in reducing levels of microbiological contamination in red meat carcasses. J. Food Prot. 65(11): Berends, B. R., F. Van Knapen, J. M. A. Snijders, and D. A. A. Mossel Identification and quantification of risk factors regarding Salmonella spp. on pork carcasses. Int. J. Food Microbiol. 36: Cates, S. C., C. L. Viator, S. A. Karns, and M. K. Muth. Forthcoming. Food safety practices of meat slaughter plants: findings from a national survey. Food Prot. Trends. 6. Cates, S. C., C. L. Viator, S. A. Karns, and M. K. Muth. Forthcoming. Food safety practices of poultry slaughter: findings from a national survey. Food Prot. Trends. 7. Cates, S. C., D. W. Anderson, S. A. Karns, and P. A. Brown Traditional versus hazard analysis and critical control point-based inspection: results from a poultry slaughter project. J. Food Prot. 64(6):

33 Duffy, E. A., K. E. Belk, J. N. Sofos, S. B. LeValley, M. L. Kain, J. D. Tatum, G. C. Smith, and C. V. Kimberling Microbial contamination occurring on lamb carcasses processed in the U.S. J. Food Prot. 64(4): Federal Register. Pathogen Reduction: Hazard Analysis and Critical Control Point (HACCP) Systems; Final Rule. 61(144) (1996): Gill, C. O., and C. Landers Microbiological effects of carcass decontaminating treatments at four beef packing plants. Meat Science. 65: Kemp, G. K., M. L. Aldrich, M. L. Guerra, and K. R. Schneider Continuous online processing of fecal- and ingesta-contaminated poultry carcasses using an acidified sodium chlorite antimicrobial intervention. J. Food Prot. 64(6): Rivera-Betancourt, M. et al Prevalence of Escherichia coli O157:H7, Listeria monocytogenes, and Salmonella in two geographically distant commercial beef processing plants in the United States. J. Food Prot. 67(2): RTI International Enhanced facilities database. Prepared for the U.S. Department of Agriculture, Food Safety and Inspection Service. RTI International, Research Triangle Park, NC. 14. Salford Systems, Inc. CART. Available at: San Diego, CA. 15. Sanchez, M. X., W. M. Fluckey, M. M. Brashears, and S. R. McKee Microbial profile and antibiotic susceptibility of Campylobacter spp and Salmonella spp in broilers processed in air-chilled and immersion chilled environments. J. Food Prot. 65(6):

34 U.S. Department of Agriculture, Food Safety and Inspection Service [Docket No N] Salmonella verification sample result reporting: agency policy and use in public health protection: notice and response to comments. Federal Register. 71(38): U.S. Department of Agriculture, Food Safety and Inspection Service Issue paper: strategy for Salmonella testing. Available at: issue02.htm. 18. U.S. Department of Agriculture, Food Safety and Inspection Service Enforcement instructions for the Salmonella performance standard. FSIS Directive 10, Available at: U.S. Department of Agriculture, Food Safety and Inspection Service FSIS scheduling criteria for Salmonella sets in raw classes of product. Available at: Whyte, P., J. D. Collins, K. McGill, C. Monahan, and H. O Mahony Quantitative investigation of the effects of chemical decontamination procedures on the microbiological status of broiler carcasses during processing. J. Food Prot. 64(2):

35 Appendix A: Meat Slaughter Data Dictionary This appendix provides the complete list of available variables for meat slaughter establishments (plants), including the variable name, label, format, and explanations for coding. The variables include Salmonella test results for pork carcasses, beef carcasses, and ground beef produced in slaughter establishments; industry survey results; establishment characteristics from the EFD; and noncompliance report data from PBIS. The analysis data sets for pork slaughter and beef slaughter establishments were derived from this list of variables. A-1

36 A-2 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary Variable Label Format Salmonella Data HogOutcome Market hogs Pass/Fail indicator (1/2 perf standard). CattleOutcome GrdBeefOutcome Survey Questions Steers & Heifers, Cows & Bulls Pass/Fail indicator (1/2 perf standard). Ground Beef Pass/Fail indicator (1/2 perf standard). Format Description Dummy 1,0 0=Pass, 1=Fail Dummy 1,0 0=Pass, 1=Fail Dummy 1,0 0=Pass, 1=Fail Q1_1_1 Plant does not dehair carcasses Q1_1_2 Q1_2_2 Q1_2_3 Q1_2_4 Q1_2_5 Q1_2_6 q1_3 Q1_4 Plant uses scald and rinse to dehair carcasses Plant uses skinning knife to dehide carcasses Plant uses air knife to dehide carcasses Plant uses mechanical side puller to dehide carcasses Plant uses mechanical down puller to dehide carcasses Plant uses mechanical up puller to dehide carcasses Independent, third-party auditors conduct audits of slaughter operations Percentage of live animals slaughtered at plant during the past year were imported Ordinal 1 5 1) None 2) 1% to 9% 3) 10% to 24% 4) 25% to 49% 5) 50% or more *Missing coded as 1 Notes (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

37 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary (continued) A-3 Variable Label Format Format Description Q1_5n Total amount of raw product, not Continuous 0 1,980,000,000 ground, primal cuts produced during the past year Q1_6r Q1_7r Q1_8a1 Q1_8b1 Q1_8c1 Q1_8d1 Q1_8e1 Q1_8f1 Routine frequency for sanitizing hands or gloves that contact raw product in the slaughter area of the plant Routine frequency for sanitizing hands or gloves that contact raw product in the fabrication area of the plant Technologies slaughter Companyowned lab for microbiological testing Technologies slaughter Bioluminescent testing system Technologies slaughter Conveyor belts made from materials designed to prevent bacterial growth Technologies slaughter Steam pasteurization systems Technologies slaughter Steam vacuum units Technologies slaughter Organic acid rinse Notes Ordinal 1 5 1) No specific routine frequency 2) One or more times per shift, but less than once per hour 3) Once per hour 4) More than once per hour 5) Always before handling the next unit of product *Missing coded as 1 Ordinal 1 5 1) No specific routine frequency 2) One or more times per shift, but less than once per hour 3) Once per hour 4) More than once per hour 5) Always before handling the next unit of product *Missing coded as 1 (continued) Appendix A Meat Slaughter Data Dictionary

38 A-4 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary (continued) Variable Label Format Format Description Notes Q1_8g1 Technologies slaughter Positive air pressure from clean side to dirty side Q1_8i1 Q1_9a1 Q1_9c1 Q1_9d1 Q1_9e1 Q1_9j1 Technologies slaughter Tempered carcass rinse/wash Requires and documents that its animal growers use stipulated production practices to control pathogens Rotates sanitizing chemicals it uses in the slaughter area on an annual basis or more frequently Uses chemical sanitizers for food contact hand tools used in the slaughter area during operations Uses sterilizer pots for heat sterilization of hand tools used in the slaughter area during operations Identifies and tracks its products, by production lot, backward to specific animal growers Section 2 of Survey: 67 plants do not further process. Did not answer Section 2. Q2_2_1 Q2_2_2 Q2_2_3 Q2_6 Plant produces ready-to-eat (RTE) products Plant produces not-ready-to-eat (NRTE) products Plant produces products that are inputs to further processing Percentage of raw meat processed during the past year that was received or purchased from another plant Ordinal 1 5 1) None 2) 1% to 9% 3) 10% to 24% 4) 25% to 49% 5) 50% or more *Missing coded as 1 (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

39 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary (continued) A-5 Variable Label Format Q2_7 Percentage of raw meat processed during the past year that was imported as raw meat Q2_12An Q2_12Bn Q2_12Cn Q2_12Dn Q2_12En Q2_12Fn Q2_12Gn Estimate total pounds during the past year. [Raw, ground meat (03B)] Estimate total pounds during the past year. [Thermally processed, commercially sterile (03D)] Estimate total pounds during the past year. [Not heat treated, shelf stable (03E)] Estimate total pounds during the past year. [Heat treated, shelf stable (03F)] Estimate total pounds during the past year. [Fully cooked, not shelf stable (03G)] Estimate total pounds during the past year. [Heat treated, but not fully cooked, not shelf stable (03H)] Estimate total pounds during the past year. [Secondary inhibitors, not shelf stable (03I)] Format Description Ordinal 1 5 1) None 2) 1% to 9% 3) 10% to 24% 4) 25% to 49% 5) 50% or more *Missing coded as 1 Continuous 0 257,745,000 Continuous 0 40,291,574 Continuous 0 2,848,254 Continuous 0 11,459,424 Continuous 0 166,000,000 Continuous 0 84,000,000 Continuous 0 500,000 Section 3 of Survey: 40 plants do not conduct additional micro testing. Did not answer Section 3. Q3_4G Q3_6G Organisms by frequency of microbial testing on carcasses prior to fabrication [Salmonella species] Organisms by frequency of microbial testing on raw meat after fabrication [Salmonella species] Notes Ordinal plants that do not test coded as 1 Ordinal plants that do not test coded as 1 (continued) Appendix A Meat Slaughter Data Dictionary

40 A-6 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary (continued) Variable Label Format Format Description Notes Q4_1 Plant provides formal food safety course for newly hired production employees Q4_3 Q5_1 Q5_2 q5_3_1r q5_3_2r q5_3_3r Q5_4 Approximate number of production employees currently working have completed formal HACCP training Calendar year the plant was built or renovated Approximate total square footage of the production space for plant Approximate percentage of the square footage of the production space of this plant that is [Under 5 years old] Approximate percentage of the square footage of the production space of this plant that is [5 years to just under 20 years old] Approximate percentage of the square footage of the production space of this plant that is [20 years old or more] How many slaughter and fabrication shifts does this plant operate daily? Ordinal 1 5 1) None 2) 1 to 3 employees 3) 4 to 9 employees 4) 10 to 20 employees 5) More than 20 employees *Missing coded as 1 Continuous Used grant date from EFD for 17 plants Continuous 200 1,053,175 Used production area from EFD for 23 plants Continuous Missing coded as 0 Continuous Missing coded as 0 Continuous Missing coded as 0 Ordinal 1 4 1) This plant does not operate on a daily basis 2) One 3) Two 4) Three (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

41 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary (continued) A-7 Variable Label Format Q5_5 How many further processing shifts does this plant operate daily? Q5_6 Q5_8 Q5_9 Q5_10 Q5_11 How many clean up shifts does this plant operate daily? Plant has a person on staff whose primary responsibility is to manage food safety activities at the plant Percentage of food safety manager s time devoted to managing food safety activities Plant has a quality control/quality assurance department Approximately how many employees at this plant work in the plants quality control/quality assurance department? Format Description Notes Ordinal 1 5 1) None 2) Further processing shift does not operate on a daily basis 3) One 4) Two 5) Three Ordinal 1 5 1) None 2) Clean up shift does not operate on a daily basis 3) One 4) Two 5) Three Ordinal skipped out of question (No FS manager) 1) 1% to 24% 2) 25% to 49% 3) 50% to 74% 4) 75% to 99% 5) 100% *86 Missing coded as 0 Continuous skipped out of question (No QA dept) *Missing coded as 0 (continued) Appendix A Meat Slaughter Data Dictionary

42 A-8 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary (continued) Variable Label Format Q5_12 Number of USDA- or state-inspected plants owned by the company that owns this plant EFD Data Plant Inspection Activities Unique Plant ID DO NOT USE IN ANALYSIS Format Description Ordinal 1 4 1) 1 2) 2 to 5 3) 6 to 20 4) 21 or more *Missing coded as 1 Text CustomSlaughter Activity Custom slaughter Dummy 1,0 Retail Activity Retail Dummy 1,0 OnLine Activity On line slaughter Dummy 1,0 Business Information POPCODE Midpoint of population within the establishment city Continuous 12, ,000 MULTIPLANT Indicates company ownership Dummy 1,0 Notes EMPACTUAL Actual employment at location Continuous Used Q5_7 for 10 missing ESTSALES Actual sales of location Continuous 291,000 5,796,000,000 HACCP & Inspection Information District District number Number 5 90 Used midpoint of ranges of Q5_13 for 14 missing Circuit Circuit number Number digit circuits indicate T/A TALAIKEN Indicates a Talmadge-Aiken plant Dummy 1,0 Meat Slaughter Volumes St_Hf Number of steers and heifers slaughtered Continuous 0 1,725,607 Cw_Bl Veal Number of cows, bulls and stags slaughtered Number of veal calves and heavy calves slaughtered Continuous 0 356,054 Continuous 0 88,446 (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

43 Table A-1. Food Safety Performance Analysis: Meat Slaughter Data Dictionary (continued) A-9 Variable Label Format Format Description Shp_Gts Number of mature sheep, lambs, Continuous 0 135,833 goats slaughtered Mkt_Hg Old_Hg Oth_Mt Inspection Systems Number of barrows and gilts slaughtered Number of sows, boars and stags slaughtered Number of equines and other animals slaughtered Continuous 0 7,640,614 Continuous 0 411,334 Continuous 0 11,358 FDADual FDA Dual Jurisdiction Dummy 1,0 HIMP HACCP-Based Inspection Models Project Dummy 1,0 NELS New Line Speed Inspection System Dummy 1,0 Plant Information HACCPSIZE HACCP Size Ordinal 1 3 1:very small 2:small 3:large PBIS Inspection Procedures (including 01 and 02, scheduled and unscheduled) Sanitation Facility HACCP Ecoli Inspection procedures failure rate of sanitation procedures Inspection procedures failure rate of facility procedures Inspection procedures failure rate of HACCP procedures Inspection procedures failure rate of E. coli procedures Notes Continuous B01, 01B02, 01C01, and 01C02 procedures Continuous D01 procedure Continuous B01, 03B02, 03C01, 03C02, 03J01, and 03J02 procedures Continuous A01 and 05A02 procedures Appendix A Meat Slaughter Data Dictionary

44 Appendix B: Young Chicken Slaughter Data Dictionary This appendix provides the complete list of available variables for young chicken slaughter establishments (plants), including the variable name, label, format, and explanations for coding. The variables include Salmonella test results for young chicken carcasses, industry survey results, establishment characteristics from the EFD, and noncompliance report data from PBIS. The analysis data set for young chicken carcasses was derived from this list of variables. B-1

45 B-2 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary Variable Label Format Format Description Salmonella Data BroilerOutcome Broiler-pass/fail indicator. half of perf Dummy 1,0 0=Pass, 1=Fail std = 12/51 Survey Questions q1_1 Q1_2r Q1_3r Q1_4 Q1_5n Independent, third-party auditors conduct audits of slaughter operations Routine frequency for sanitizing hands or gloves that contact raw poultry in the slaughter area of the plant Routine frequency for sanitizing hands or gloves that contact raw poultry in the deboning area of the plant Percentage of live birds slaughtered at plant during the past year were imported Total amount of raw product, not ground, primal cuts produced during the past year Notes Ordinal 1 5 1) No specific routine frequency 2) One or more times per shift, but less than once per hour 3) Once per hour 4) More than once per hour 5) Always before handling the next unit of product *Missing coded as 1 Ordinal 1 5 1) No specific routine frequency 2) One or more times per shift, but less than once per hour 3) Once per hour 4) More than once per hour 5) Always before handling the next unit of product *Missing coded as 1 Ordinal 1 5 1) None 2) 1% to 9% 3) 10% to 24% 4) 25% to 49% 5) 50% or more *Missing coded as 1 Continuous 0 3,000,000,000 (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

46 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary (continued) Variable Label Format Format Description Notes B-3 Q1_6a1 Q1_6b1 Q1_6c1 Q1_6d1 Q1_6e1 Q1_6g1 Q1_7a1 Q1_7c1 Q1_7d1 Q1_7e1 Q1_7j1 Technologies slaughter-companyowned lab for microbiological testing Technologies slaughterbioluminescent testing system Technologies slaughter-conveyor belts made from materials designed to prevent bacterial growth Technologies slaughter-insideoutside bird washers Technologies slaughter-organic acid rinse Technologies slaughter-automatic bird transfer (from kill line to evisceration line) Requires and documents that its animal growers use stipulated production practices to control pathogens Rotates sanitizing chemicals it uses in the slaughter area on an annual basis or more frequently Uses chemical sanitizers for food contact hand tools used in the slaughter area during operations Uses sterilizer pots for heat sterilization of hand tools used in the slaughter area during operations Identifies and tracks its products, by production lot, backward to specific animal growers (continued) Appendix B Young Chicken Slaughter Data Dictionary

47 B-4 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary (continued) Variable Label Format Format Description Notes Q1_7k1 Identifies and tracks its products, by production lot, forward to specific buyers (not consumers) of its products Q1_7l1 Conducts fat pad sampling on a regular schedule Section 2 of Survey: 58 plants do not further process. Did not answer Section 2. Q2_2_1 Q2_2_2 Q2_2_3 Q2_6 Q2_7 Q2_11An Q2_11Bn Plant produces ready-to-eat (RTE) products Plant produces not-ready-to-eat (NRTE) products Plant produces products that are inputs to further processing Percentage of raw poultry processed during the past year that was received or purchased from another plant Percentage of raw poultry processed during the past year that was imported as raw poultry Estimate total pounds during the past year. [Raw, ground meat (03B)] Estimate total pounds during the past year. [Thermally processed, commercially sterile (03D)] Ordinal 1 5 1) None 2) 1% to 9% 3) 10% to 24% 4) 25% to 49% 5) 50% or more *Missing coded as 1 Ordinal 1 5 1) None 2) 1% to 9% 3) 10% to 24% 4) 25% to 49% 5) 50% or more *Missing coded as 1 Continuous 0 288,000,000 Continuous 0 (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

48 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary (continued) B-5 Variable Label Format Format Description Notes Q2_11Cn Estimate total pounds during the past year. [Not heat treated, shelf stable (03E)] Continuous 0 23,400,000 2 plants with volumes Q2_11Dn Q2_11En Q2_11Fn Q2_11Gn Estimate total pounds during the past year. [Heat treated, shelf stable (03F)] Estimate total pounds during the past year. [Fully cooked, not shelf stable (03G)] Estimate total pounds during the past year. [Heat treated, but not fully cooked, not shelf stable (03H)] Estimate total pounds during the past year. [Secondary inhibitors, not shelf stable (03I)] Continuous 0 80,000,000 3 plants with volumes Continuous 0 238,000,000 Continuous 0 145,000,000 Continuous 0 93,436,000 3 plants with volumes Section 3 of Survey: 8 plants do not conduct additional micro testing. Did not answer Section 3. Q3_3D Organisms by frequency of microbial testing on carcasses prior to fabrication [Salmonella species] Ordinal 1 9 1) Never 2) Less than once per month 3) Once per month 4) More than once per month 5) Once per week 6) More than once per week 7) Once per day 8) Once per shift 9) More than once per shift *Plants that do not test coded as 1 (continued) Appendix B Young Chicken Slaughter Data Dictionary

49 B-6 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary (continued) Variable Label Format Q3_5D Organisms by frequency of microbial testing on raw meat after fabrication [Salmonella species] Q4_1 Q4_3 Q5_1 Q5_2 q5_3_1r q5_3_2r Plant provides formal food safety course for newly hired production employees Approximate number of production employees currently working have completed formal HACCP training Calendar year the plant was built or renovated Approximate total square footage of the production space for plant Approximate percentage of the square footage of the production space of this plant that is [Under 5 years old] Approximate percentage of the square footage of the production space of this plant that is [5 years to just under 20 years old] Format Description Notes Ordinal 1 9 1) Never 2) Less than once per month 3) Once per month 4) More than once per month 5) Once per week 6) More than once per week 7) Once per day 8) Once per shift 9) More than once per shift *Plants that do not test coded as 1 Ordinal 1 5 1) None 2) 1 to 3 employees 3) 4 to 9 employees 4) 10 to 20 employees 5) More than 20 employees *Missing coded as 1 Continuous Used grant date from EFD for 1 plants Continuous ,000 Used production area from EFD for 7 plants Continuous Missing coded as 0 Continuous Missing coded as 0 (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

50 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary (continued) B-7 Variable Label Format Format Description Notes q5_3_3r Approximate percentage of the square footage of the production space of this plant that is [20 years old or more] Continuous Missing coded as 0 Q5_4 Q5_5 Q5_6 Q5_7 Q5_9 How many slaughter and fabrication shifts does this plant operate daily? How many deboning shifts does this plant operate daily? How many further processing shifts does this plant operate daily? How many clean up shifts does this plant operate daily? Plant has a person on staff whose primary responsibility is to manage food safety activities at the plant Ordinal 1 4 1) This plant does not operate on a daily basis 2) One 3) Two 4) Three Missing coded as 1 Ordinal 1 5 1) None 2) Deboning shift is not operated on a daily basis 3) One 4) Two 5) Three Missing coded as 1 Ordinal 1 5 1) None 2) Further processing shift does not operate on a daily basis 3) One 4) Two 5) Three Ordinal 1 5 1) None 2) Clean up shift does not operate on a daily basis 3) One 4) Two 5) Three Missing coded as 1 (continued) Appendix B Young Chicken Slaughter Data Dictionary

51 B-8 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary (continued) Variable Label Format Q5_10 Percentage of food safety manager s time devoted to managing food safety activities Q5_11 Q5_12 Q5_13 EFD Data Plant Business Information POPCODE Plant has a quality control/quality assurance department Approximately how many employees at this plant work in the plants quality control/quality assurance department? Number of USDA- or state-inspected plants owned by the company that owns this plant Unique plant ID DO NOT USE IN ANALYSIS Midpoint of population within the establishment city Format Description Notes Ordinal skipped out of question (No FS manager) 1) 1% to 24% 2) 25% to 49% 3) 50% to 74% 4) 75% to 99% 5) 100% *34 Missing coded as 0 Continuous skipped out of question (No QA dept) *Missing coded as 0 Ordinal 1 4 1) 1 2) 2 to 5 3) 6 to 20 4) 21 or more *Missing coded as 1 Text Continuous 12, ,000 MULTIPLANT Indicates company ownership Dummy 1,0 EMPACTUAL Actual employment at location Continuous Used Q5_7 for 17 missing ESTSALES Actual sales of location Continuous 1,116,000 1,518,000,000 Used midpoint of ranges of Q5_13 for 21 missing (continued) Analysis of Food Safety Performance in Meat and Poultry Establishments

52 Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter Data Dictionary (continued) B-9 Variable Label Format Format Description HACCP & Inspection Information District District number Number 5 90 Circuit Circuit number Number Young Chicken Slaughter Volumes Yng_Chick Number of young chickens Continuous 0 135,225,459 slaughtered Oth_Chick Yng_Turk Oth_Turk Oth_Pltry Inspection Systems Number of capons, light and heavy fowl slaughtered Number of young turkeys slaughtered Number of roaster, breeder, old breeder turkeys slaughtered Number of ducks, geese, rabbits, slaughtered Continuous 0 14,203,280 Continuous 0 15,926,998 Continuous 0 649,516 Continuous 0 2,857,272 FDADual FDA Dual Jurisdiction Dummy 1,0 HIMP HACCP-Based Inspection Models Project Dummy 1,0 NELS New Line Speed Inspection System Dummy 1,0 Plant Information HACCPSIZE HACCP Size Ordinal 1 3 1:very small 2:small 3:large PBIS Inspection Procedures (including 01 and 02, scheduled and unscheduled) Sanitation Facility HACCP Ecoli Inspection procedures failure rate of sanitation procedures Inspection procedures failure rate of facility procedures Inspection procedures failure rate of HACCP procedures Inspection procedures failure rate of E. coli procedures Notes Continuous B01, 01B02, 01C01, and 01C02 procedures Continuous D01 procedure Continuous B01, 03B02, 03C01, 03C02, 03J01, and 03J02 procedures Continuous A01 and 05A02 procedures Appendix B Young Chicken Slaughter Data Dictionary

53 Appendix C: Classification Tree for Beef Carcasses and Ground Beef Produced in Cattle Slaughter Establishments This appendix provides the results of the classification tree analysis for beef carcasses and ground beef produced in cattle slaughter establishments. Because few establishments fail the Salmonella performance standards for beef carcasses and ground beef, only the initial analyses were conducted for establishments that produce these products. However, the results of this initial analysis provide useful insights into the characteristics of establishments likely to fail. C.1 RESULTS FOR ESTABLISHMENTS THAT PRODUCE BEEF CARCASSES The classification tree for cattle slaughter establishments based on their Salmonella performance on beef carcasses is provided in Figure C-1. A total of 14 establishments, out of 133 that completed the survey and were subject to Salmonella testing, failed half the Salmonella performance standard during the 2003 through 2005 time period. The results of the classification tree analysis indicate the following: C-1

54 C-2 Figure C-1. Classification of Cattle Slaughter Establishments with Respect to the Beef Carcass Outcome Measure N = 133 QCW_BL_L <= QCW_BL_L > N = 51 N = 82 ESTSALES_L <= ESTSALES_L > N = 53 N = 29 ESTSALES_L <= ESTSALES_L > N = 21 N = 32 QCW_BL_L <= QCW_BL_L > HACCP <= HACCP > N = 1 N = 20 N = 7 N = 25 COTH_MT = (1) COTH_MT = (0) N = 2 N = 23 CQ1_9D1 = (0) CQ1_9D1 = (1) N = 8 N = 15 SANITATION <= SANITATION > N = 1 N = 14 FACILITY <= FACILITY > N = 1 N = 13 HACCP_TRAINED <= HACCP_TRAINED > N = 1 N = 12 Analysis of Food Safety Performance in Meat and Poultry Establishments

55 Appendix C Classification Tree for Beef Carcasses and Ground Beef Produced in Cattle Slaughter Establishments The cattle slaughter establishments that failed are those with the number of cows, bulls, and stages slaughtered exceeding 712 (QCW_BL-l = 5.15 in log scale) and with an estimated sales volume less than $54,843,816 (ESTSALES_L = in log scale). Virtually all of the establishments (13 out of 14) are further classified as those whose estimated sales volume is greater than $5,075,827 (ESTSALES = in log scale), have a HACCP noncompliance report rate greater than 0.3% (HACCP > 0.003), and do not slaughter equines and other meat species (COTH_MT = 0). The great majority of the above 13 establishments (11 establishments) are further classified as those that use chemical sanitizers for food hand tools (CQ1_9D1 = 1), have a sanitation noncompliance report rate exceeding 0.7% (SANITATION > 0.007), have a facility noncompliance report rate exceeding 5.7% (FACILITY > 0.057), and have more than 2% of their production employees completed formal HACCP training (HACCP_TRAINED > 0.021). The classification depicts the most homogeneous classification of establishments based on an evolving set of establishment characteristics defined by the predictor variables that emerge as the strongest at each node of the classification tree. Considering the cumulative classification power of each predictor variable, however, a measure of relative importance for each variable can be calculated that reflects the given predictor s overall classification power. Table C-1 indicates the cumulative power of the most important variables in descending order of importance for beef carcasses. Table C-1. Relative Importance of the Predictor Variables for Beef Carcasses Variable Description Relative Importance QCW_BL_L Log of number of cows, bulls, and stags slaughtered 22.0% annually ESTSALES_L Log of midpoint of establishment sales category 21.7% Q4_3 Number of production employees who have completed 14.5% formal HACCP training EMPACTUAL Number of employees at the establishment 13.6% ST_HF_L Log of number of steers and heifers slaughtered 12.5% annually Q1_2_4 Establishment uses mechanical side puller to dehide 7.7% carcasses Q5_12 Number of USDA- or state-inspected establishments owned by the company that owns the establishment 5.0% C-3

56 Analysis of Food Safety Performance in Meat and Poultry Establishments C.2 RESULTS FOR CATTLE SLAUGHTER ESTABLISHMENTS THAT PRODUCE GROUND BEEF The classification tree for cattle slaughter establishments based on their Salmonella performance for ground beef is provided in Figure C-2. A total of 9 establishments, out of 80 that completed the survey and were subject to Salmonella testing, failed half the Salmonella performance standard during the 2003 through 2005 time period. The results of the classification tree analysis indicate that establishments with the following characteristics are more likely to fail: a HACCP noncompliance report rate exceeding 1% (HACCP > 0.01), less than 18% of their production employees completed formal HACCP training (HACCP_TRAINED 0.18), production space of less than 603,198 square feet (Q5_2_L = in log scale), annual number of steers and heifers slaughtered less than 1,155,449 (ST_HF_L in log scale), less than 2.5% of raw meet processed during the last year imported as raw meat (Q2_7 2.5), and no requirement for documents that animal growers use stipulated production practices to control pathogens (CQ1_9A1 = 0). Similar to Table C-1, Table C-2 indicates the cumulative power of the most important variables in descending order of importance for ground beef produced by establishments that slaughter cattle. C-4

57 Appendix C Classification Tree for Beef Carcasses and Ground Beef Produced in Cattle Slaughter Establishments Figure C-2. Classification of Cattle Slaughter Establishments with Respect to the Ground Beef Outcome Measure N = 80 HACCP <= 0.01 Class = N = 38 HACCP > N = 42 HACCP_TRAINED <= N = 32 HACCP_TRAINED > 0.18 Class = N = 10 CQ1_1_1 = (1) N = 25 CQ1_1_1 = (0) Class = N = 7 HACCP_TRAINED <= 0.00 Class = N = 5 HACCP_TRAINED > N = 20 CECOLI = (0) N = 17 CECOLI = (1) Class = N = 3 Q5_2_L <= N = 15 Q5_2_L > Class = N = 2 ST_HF_L <= N = 13 ST_HF_L > Class = N = 2 Q2_7 <= N = 12 Q2_7 > 2.50 Class = N = 1 CQ1_9A1 = (0) Class = N = 11 CQ1_9A1 = (1) Class = N = 1 C-5

58 Analysis of Food Safety Performance in Meat and Poultry Establishments Table C-2. Relative Importance of the Predictor Variables for Beef Carcasses Variable Description Relative Importance HACCP Q4_3 Q1_7R Percentage of HACCP procedures that resulted in noncompliance reports Number of production employees who have completed formal HACCP training Frequency of sanitizing hands or gloves that contact raw product in the fabrication area of the establishment 29.3% 19.7% 18.8% EMPACTUAL Number of employees at the establishment 15.5% Q1_8F1 Use of organic acid rinse during slaughter 10.2% SANITATION Percentage of sanitation procedures that resulted in noncompliance reports 6.5% C-6

59 Appendix D: Presentation on Results of the Analysis This appendix includes the presentation of the results for broiler slaughter and pork slaughter establishments (plants) that RTI presented at FSIS in Washington, DC, on March 12, The analysis was later revised and finalized for this report. D-1

60 Analysis of Food Safety Performance in Meat and Poultry Establishments D-2

61 Appendix D Presentation on Results of the Analysis D-3

62 Analysis of Food Safety Performance in Meat and Poultry Establishments D-4

63 Appendix D Presentation on Results of the Analysis D-5

64 Analysis of Food Safety Performance in Meat and Poultry Establishments D-6

65 Appendix D Presentation on Results of the Analysis D-7

66 Analysis of Food Safety Performance in Meat and Poultry Establishments D-8

67 Appendix D Presentation on Results of the Analysis D-9

68 Analysis of Food Safety Performance in Meat and Poultry Establishments D-10

69 Appendix D Presentation on Results of the Analysis D-11

70 Analysis of Food Safety Performance in Meat and Poultry Establishments D-12