Analysis of Food Safety Performance in Meat and … 2007 Analysis of Food Safety Performance in Meat and Poultry Establishments Revised Final Report Contract No. 53-3A94-03-12, Task

November 2007

Analysis of Food Safety Performance in Meat and Poultry

Establishments

Revised Final Report Contract No. 53-3A94-03-12, 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 20250

Prepared by

Mary K. Muth Mansour Fahimi Shawn A. Karns

Yan Li RTI International

Health, Social, and Economics Research Research Triangle Park, NC 27709

RTI Project Number 0208893.018

RTI Project Number 0208893.018

Analysis of Food Safety Performance in Meat and Poultry

Establishments

Revised Final Report Contract No. 53-3A94-03-12, 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 20205

Prepared by

Mary K. Muth Mansour Fahimi Shawn A. Karns

Yan Li RTI International

Health, Social, and Economics Research Research Triangle Park, NC 27709

RTI International is a trade name of Research Triangle Institute

1

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 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.

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.

Analysis of Food Safety Performance in Meat and Poultry Establishments

2

FSIS conducted surveys of meat and poultry slaughter and processing establishments in 2004 and of processing-only establishments in 2005. 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.

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

Several data analysis methods were implemented to determine key predictors of food safety performance from a large number of available variables.

Preface

3

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


4

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.

1

Running Head: Meat and Poultry Pathogen Control Performance 1

2 Analysis of Pathogen Control Performance in 3

U.S. Young Chicken Slaughter and Pork Slaughter Establishments 4

5

Mary K. Muth,* Mansour Fahimi, Shawn A. Karns 6

RTI International 7

3040 Cornwallis Road 8

P.O. Box 12194 9

Research Triangle Park, NC 27709 10

11

October 2007 12

13

Key words: Salmonella, young chicken slaughter, pork slaughter, establishment performance, 14

risk-based inspection 15

16

17

* Mary K. Muth, RTI International, 3040 Cornwallis Road, P.O. Box 12194, Research Triangle Park, NC 27709;

Voice: (919) 541-7289; Fax: (919) 541-6683; E-mail: [email protected]

2

Analysis of Pathogen Control Performance in 18

U.S. Young Chicken Slaughter and Pork Slaughter Establishments 19

ABSTRACT 20

In the 1996 U.S. Department of Agriculture, Food Safety and Inspection Service’s 21

(FSIS’s) “Pathogen Reduction; Hazard Analysis and Critical Control Point (PR/HACCP) 22

Systems, Final Rule,” Salmonella was selected for microbiological testing and monitoring. Using 23

data from an FSIS-sponsored survey of meat and poultry slaughter establishments, inspection 24

results, and other establishment characteristics, potential variables affecting pathogen control, as 25

measured by Salmonella test results, were investigated. The analysis data sets included 153 26

federally inspected young chicken slaughter establishments, of which 111 failed half the 27

Salmonella performance standard at least once from 2003 through 2005, and 121 federally 28

inspected pork slaughter establishments, of which 28 failed half the Salmonella performance 29

standard. Logistic regression results for young chicken slaughter establishments seem to indicate 30

they were more likely to fail if they had higher inspection noncompliance rates (p = .10) and 31

older production space (p = .10) and were less likely to fail if they used a higher percentage of 32

raw poultry inputs purchased from outside sources (p = .10). Results for pork slaughter 33

establishments indicate they were more likely to fail if they had a higher rate of voluntary 34

microbiological testing (p = .08) and were less likely to fail if they were larger (p = .08) and used 35

a higher percentage of raw pork inputs purchased from outside sources (p = .02). In general, 36

indicators of plant characteristics, food safety practices, and management philosophy are 37

associated with different levels of pathogen control performance that vary by species 38

slaughtered. 39

3

The 1996 U.S. Department of Agriculture, Food Safety and Inspection Service (FSIS) 40

“Pathogen Reduction; Hazard Analysis and Critical Control Point (PR/HACCP) Systems, Final 41

Rule” sets Salmonella performance standards for establishments slaughtering selected classes of 42

food animals or those producing selected classes of raw ground products (Federal Register, 43

2006). FSIS selected Salmonella for microbiological testing because it is a common bacterial 44

cause of foodborne illness, it occurs often enough to be detected and monitored, methods are 45

available for recovering Salmonella from a variety of meat and poultry products, and 46

intervention strategies aimed at reducing Salmonella should be effective at reducing other 47

foodborne pathogens (17). Testing began under the PR/HACCP rule in January 1998 for large 48

establishments (500 or more employees), in January 1999 for small establishments (from 10 to 49

499 employees), and in January 2000 for very small establishments (9 or fewer employees or less 50

than $2.5 in annual sales) (17). 51

The performance standards for Salmonella testing vary by type of product, as indicated in 52

Table 1. The sample set size varies from 51 to 82 samples, and the maximum positives varied 53

from 1 to 29 depending on product type (18). Establishments that slaughter young chickens fail 54

the performance standards with 13 positives out of 51 samples, and establishments that slaughter 55

hogs fail the performance standard with 7 positives out of 55 samples. Recently, FSIS developed 56

scheduling criteria such that establishments that fail half the performance standard (i.e., are at 57

51% of the performance standard or higher) during a sample set are tested more frequently than 58

establishments that are at 50% of the performance standard or less (19). Thus, young chicken 59

slaughter establishments that have 7 or more positives and hog slaughter establishments that 60

have 4 or more positives are subsequently tested more frequently. 61

4

Table 1. Salmonella Sample Set Size and Performance Standard for Each Meat and Poultry 62

Product Type 63

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 51 12

Ground chicken 53 26

Ground turkey 53 29

64

Salmonella test results provide a measure of pathogen control in meat and poultry 65

produced at slaughter and processing establishments and thus can be used to verify that industry 66

systems are effective in controlling the contamination of raw meat and poultry products with 67

disease-causing bacteria. Whether an establishment typically passes the Salmonella performance 68

standard may depend on a variety of factors, including the characteristics of the establishment, 69

use of food safety technologies, and management practices in the establishment. Recent surveys 70

of food safety technologies and practices were conducted for FSIS to determine the prevalence of 71

various food safety technologies and practices used in the U.S. poultry and meat slaughter 72

industries, to characterize the use and types of microbiological testing, and to determine the 73

prevalence of different types of employee food safety training (5, 6). The results of the survey 74

5

combined with other data sources were used to determine which factors are typically associated 75

with better pathogen control performance as measured by Salmonella test results. 76

The use of different types of food safety technologies and characteristics of slaughter 77

establishments has been shown to affect pathogen levels on carcasses. In particular, the use of 78

different types of carcass rinse solutions, carcass decontamination methods, and skin removal 79

methods has been shown to reduce microbial carcass counts (3, 4, 8, 10, 12, 20). Line speeds 80

within the establishment also affect pathogen levels on carcasses as shown for E. coli 81

populations on beef carcasses (2). The use of online versus off-line reprocessing of poultry 82

carcasses to meet zero fecal tolerance standards affects Campylobacter and Salmonella levels on 83

poultry carcasses (11). Furthermore, the use of different carcass chilling methods (immersion 84

versus air) affects Salmonella and Campylobacter incidence on broiler carcasses (15). In addition 85

to using specific food safety technologies, other factors shown to affect pathogen levels on 86

carcasses include location and size of the establishment (1). 87

The purpose of this study was to identify young chicken and hog slaughter establishment 88

characteristics, including the use of specific food safety technologies, associated with better 89

pathogen control performance as measured by Salmonella test results. The analysis was 90

conducted using data from the FSIS surveys of meat and poultry slaughter establishments, data 91

on establishment characteristics from FSIS’s databases and a private vendor, and data on 92

inspection procedure results from FSIS. This study’s results provide information useful for 93

informing the development of a risk-based inspection system and for identifying establishments 94

most likely to be in need of technical assistance for controlling pathogens. 95

6

MATERIALS AND METHODS 96

Data Sets. The data sets used for the analyses were constructed from the following 97

sources: 98

• FSIS’s Enhanced Facilities Database, which contains production and economic data 99

for fiscal year 2004 on establishment characteristics for all meat and poultry 100

establishments under federal inspection from FSIS’s Performance-Based Inspection 101

System (PBIS) database, FSIS’s Animal Disposition Reporting System (ADRS), and 102

a private data vendor, infoUSA (13); 103

• results of FSIS’s meat and poultry slaughter establishment survey, which was 104

conducted using nationally representative sampling procedures and includes data on 105

food safety technologies and practices for all meat and poultry establishments that 106

completed the survey in late 2004 and early 2005 (5, 6); 107

• FSIS’s Salmonella test results for all establishments that were tested in FY 2003, FY 108

2004, and FY 2005; and 109

• FSIS’s noncompliance report data from the PBIS database for facility, sanitation, and 110

HACCP inspection procedures conducted in 2004. 111

Data from each of the databases were matched based on FSIS establishment number to construct 112

a complete record for each establishment. The variables in the data set include Salmonella test 113

results for each product produced in the establishment; self-reported establishment responses to 114

survey questions on food safety practices during slaughter and further processing, voluntary 115

microbiological testing practices, employee food safety training, and other characteristics of the 116

establishment; economic variables such as number of employees, annual sales, number of other 117

establishments owned by the company, and slaughter volumes by species; and results of 118

7

inspection procedures conducted by FSIS inspectors. All of the data pertain primarily to 119

performance and activities in 2004 with the exception of the Salmonella test results, which 120

pertain to FY 2003, FY 2004, and FY 2005. 121

We constructed the outcome measures (dependent variables) for the analyses based on 122

the results of Salmonella testing for the first sample set for each establishment. Results from the 123

second and third sample sets, if they were conducted, were not included because most 124

establishments pass the first set and thus only have data for the first set and because subsequent 125

sample sets are not representative of typical operating practices of the establishment since 126

establishments make adjustments after failing the first set. We included test results for 2003 127

through 2005 to allow for a greater number of test sets to be represented in the data. However, by 128

including multiple years of test results, we assumed that establishment characteristics over 3 129

years’ time are relatively constant. To correspond to FSIS’s new scheduling criteria for 130

Salmonella testing, we classified establishments as “failures” for the purposes of the analysis if 131

they failed half the performance standard at least once during the 2003 through 2005 time period. 132

We constructed the set of predictors (independent variables) in the analysis data sets 133

using the remaining variables contained in the data sets. All variables that could be potential 134

predictors of pathogen control performance were retained. Data preparation steps were 135

conducted to logically impute missing values from the survey responses and eliminate redundant 136

variables from multiple data sources. In addition, regulatory compliance variables were 137

constructed from inspection procedure data from PBIS. Specifically, we defined facility, 138

sanitation, and HACCP compliance variables by dividing the number of inspection procedures 139

8

that resulted in a noncompliance report by the total number of inspection procedures performed 140

under each category of procedures.† 141

As indicated in Table 2, the resulting data set for young chicken slaughter establishments 142

included 153 establishments, of which 111 failed half the Salmonella performance standard at 143

least once from 2003 through 2005. The resulting data set for pork slaughter establishments 144

included 121 establishments, of which 28 failed half the Salmonella performance standard. The 145

young chicken slaughter data set represents a random sample of approximately 70% of the 146

population of federally inspected establishments that slaughter young chickens; these 147

establishments slaughtered 82% of the volume of young chickens slaughtered in 2004. The pork 148

slaughter data set represents a random sample of approximately 20% of the population of 149

federally inspected establishments that slaughter hogs; these establishments slaughtered 81% of 150

the volume of market hogs slaughtered in 2004.‡ Establishments not included in the data sets are 151

those that were not surveyed or were not subject to Salmonella testing during the time period of 152

the data. 153

† 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.

9

Table 2. Total Numbers of Establishments in the Analysis Data Sets and Number of 154

Establishments that Failed Half the Performance Standard for Salmonella 155

Analysis Category

Number of Slaughter

Establishments in

Each Category

Number of

Establishments in

Analysis Data Seta

Number that Failed

Half the Performance

Standard

Young chicken slaughter

establishments

221 153 111

Pork slaughter

establishments

603 121 28

a Establishments were excluded from the analysis data set if they were not subject to Salmonella testing in 2003 156

through 2005 or were not surveyed. 157

Analysis Methods. Before conducting the analysis, we took two steps to further prepare 158

the data sets. First, some variables, particularly those related to sales and slaughter volume, 159

exhibited a great deal of variability across establishments. Thus, continuous variables with a 160

large degree of variability were stabilized using a natural log transformation. Second, a few 161

measures that depend on (i.e., increase with) the size of the establishment were normalized with 162

respect to their corresponding establishment size. For example, the number of production 163

employees who have completed HACCP training should be compared relative to the size of the 164

establishment. Thus, these types of variables were expressed as a proportion to the total number 165

of employees. 166

The data sets contained a large number of potential predictor variables (66 variables for 167

young chicken slaughter establishments and 89 variables for pork slaughter establishments). 168

Thus, the classification and regression tree (CART) procedure was used to identify potential 169

10

predictors of the Salmonella outcome measure (Salford Systems, Inc.). Specifically, for each data 170

set, a classification tree was constructed to show the partitioning of establishments based on 171

variables that had an effect on Salmonella control performance. Using this exploratory 172

investigation, the relative importance of the highest ranking predictors was established based on 173

their cumulative contribution in classifying establishments. The relative importance of the 174

variables was used to identify a smaller set of key predictors used in the subsequent analyses. 175

Multicollinearity is a common problem in statistical modeling and arises when predictor 176

variables are correlated. Because most statistical techniques, particularly regression-based 177

procedures, rely on the assumption that all predictor variables are independent, inclusion of 178

correlated data can result in highly unstable results. To address this problem in the analyses, we 179

used factor analysis to develop predictors of Salmonella control performance that are orthogonal 180

(independent) to one another. Furthermore, factor analysis enabled us to identify sets of 181

underlying factors, or themes, among the predictor variables. Factor analysis can be conducted 182

using continuous variables but does not incorporate binary variables; however, binary variables 183

were incorporated in the final analysis step. 184

Using the resulting themes and key binary variables identified in the CART analyses, we 185

estimated logistic regressions to determine the magnitude and direction of the effects of the key 186

variables on Salmonella control performance. The logistic regressions were first applied using 187

only the factors obtained in the factor analysis for young chicken slaughter and pork slaughter 188

establishments. Then, the logistic regressions were reapplied, including not only the factors but 189

also the key binary predictor variables that were identified in the CART analysis to determine 190

whether these variables might provide additional explanatory power not already captured by the 191

factors. 192

11

For young chicken slaughter establishments, we also included a binary variable indicating 193

whether the establishment operates under a HACCP-based inspection models project (HIMP) 194

because FSIS was interested in knowing whether HIMP establishments performed better or 195

worse on Salmonella performance when controlling for other variables that might affect 196

establishment performance. Under HIMP inspection, establishment employees carry out the 197

sorting activities previously conducted by FSIS inspectors to determine which carcasses and 198

parts are unacceptable because they are diseased or unwholesome (7). A total of 17 199

establishments operating under HIMP inspection were included in the young chicken slaughter 200

data set. 201

In summary, the analyses for each data set were conducted in the following steps: 202

• Step 1. Transformed variables and prepared data sets for the analysis 203

• Step 2. Identified important predictor variables of Salmonella control performance 204

using CART classification trees 205

• Step 3. Identified key themes, or clusters of variables, for Salmonella control 206

performance using factor analysis 207

• Step 4. Estimated the magnitude and direction of effects of the key themes and binary 208

variables on Salmonella control performance using logistic regression 209

RESULTS 210

The results of the series of analyses for young chicken slaughter and pork slaughter 211

establishments are described below. 212

Young Chicken Slaughter Establishments. Figure 1 shows the top nodes of the CART 213

classification tree for young chicken slaughter establishments. The strongest single predictor was 214

the percentage of production space that is 20 years old or older. Establishments with less than 215

12

Figure 1. Classification Tree for the Salmonella Outcome Variable for Young 216

Chicken Slaughter Establishments 217

218 ND = not defined (cannot be calculated for 100% failure rates). 219

56% of their production space that is 20 years old or older were more likely to pass half the 220

Salmonella performance standard (odds ratio = 3.1).§ Of establishments with younger production 221

space on the left branch of the tree, establishments with a sanitation noncompliance rate below 222

5% were more likely to pass (odds ratio = 3.5). Of establishments along the far left branch of the 223

tree, age of the establishment production space again emerged as an important predictor. 224

Establishments with less than 28% of their production space that is 20 years old or older were 225

less likely to pass (odds ratio = 0.2), perhaps indicating an effect of less experienced 226

establishment management. Establishments that have a sanitation noncompliance rate above 5% 227

and do not use chemical sanitizers on hand tools all failed. 228

On the right branch of the tree, which includes establishments with more than 56% of 229

their production space 20 years old or older, establishments were less likely to pass if they are 230

owned by a company that owns three or fewer other establishments (odds ratio = 0.1). Of 231 § 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.

13

establishments owned by a company that owns three or fewer other establishments, 232

establishments were subsequently more likely to pass if their sanitation inspection 233

noncompliance rate was less than 11% (odds ratio = 5.3). On the far right branches of the tree, 234

establishments owned by a company with four or more establishments and with sales exceeding 235

$178 million (natural log of 19) all failed half the Salmonella performance standard. Thus, 236

although on average establishments owned by larger companies performed better, the largest 237

establishments owned by larger companies performed worse. 238

Table 3 provides the results of the factor analysis, including all continuous predictor 239

variables with a nonzero relative importance level as indicated by the CART analysis. The set of 240

underlying themes, which comprise subgroups of the predictor variables, emerged in the factor 241

analysis based on the factor loadings for each set of variables. These themes included measures 242

of establishment size (number of establishments owned by the company, slaughter volume, 243

number of employees, and dollar sales volume), employee training and quality assurance, 244

inspection noncompliance rates (facility and sanitation inspection procedures), percentage of 245

poultry processed using raw poultry received or purchased from another establishment (owned 246

by the same or a different company), and age of the production space in the establishment. These 247

five factors were then used as regressors in the logistic regression analysis. 248

Table 4 presents two sets of logistic regression results for young chicken slaughter 249

establishments. In the first set on the left, only the factors identified in the factor analysis were 250

included in the regression. In the second set on the right, the same set of factors was included 251

along with binary variables that had high relative importance levels in the CART analysis results. 252

253

14

Table 3. Factor Analysis Results for the Young Chicken Slaughter Establishment 254

Salmonella Outcome Measure 255

Variable Factor 1.

Establish-

ment Size

Factor 2.

Training &

QA

Factor 3.

Noncompli-

ance Rates

Factor 4.

Use of Raw

Inputs

Factor 5.

Age of

Production

Space

Inspected

establishments owned 0.87 — — — —

Young chickens

slaughtered 0.78 — — — —

Employees at location 0.75 — — — —

Sales at location 0.70 −0.52 — — —

Employees in QA

department — 0.93 — — —

Employees trained

HACCP — 0.92 — — —

Facility

noncompliance rate — — 0.87 — —

Sanitation

noncompliance rate — — 0.85 — —

Raw poultry imported — — — 0.91 —

Raw poultry purchased 0.45 — — 0.61 —

% production space 20

years old or older — — — — 0.97

Note: Factor loadings less than 0.4 are not shown. 256

15

Table 4. Results of Logistic Regression for the Young Chicken Slaughter Salmonella 257

Outcome Measure 258

Estimates with Factors Only Estimates with Factors and Key

Binary Variables

Variable Estimate

(Standard Error)

P Value Estimate

(Standard Error)

P Value

Intercept 1.0332

(0.1919)

<0.0001 1.4762

(0.7951)

0.0622

Factor 1: Establishment

size

−0.0911

(0.1821)

0.6168 −0.2515

(0.2270)

0.2680

Factor 2: Training & QA −0.0773

(0.1711)

0.6516 −0.1030

(0.1770)

0.5607

Factor 3: Noncompliance

rates

0.3288

(0.1994)

0.0993 0.2938

(0.2062)

0.1543

Factor 4: Use of raw

inputs

−0.3531

(0.2184)

0.1059 −0.3568

(0.2300)

0.1208

Factor 5: Age of

production space

0.3212

(0.1936)

0.0971 0.2906

(0.2049)

0.1560

Chemical sanitizers on

hand tools (binary)

0.6495

(0.4788)

0.1750

NELS inspection (binary) −0.7365

(0.4992)

0.1401

HIMP inspection (binary) 0.1635

(0.6333)

0.7963

16

Tracks products forward

(binary)

−1.2070

(0.6374)

0.0583

Likelihood ratio χ2 9.4744 0.0916 17.2814 0.0445

Note: NELS is the New Line Speed inspection system and HIMP is the HACCP-based Models Project inspection 259

system. 260

Based on the results of the logistic regression including only the factors, higher inspection 261

noncompliance rates increased the likelihood that an establishment failed half the Salmonella 262

performance standard (p = .10), a higher percentage of raw poultry inputs purchased from 263

outside sources decreased the likelihood (p = .10), and use of older production space increased 264

the likelihood (p = .10). The effects of establishment size and employee training were not 265

significant, and their estimated coefficients indicated a much smaller magnitude of effect 266

compared with the three significant factors. 267

When the binary variables were also included in the regression, the p-values for some of 268

the factors increased compared to the previous specification. Under this specification, only the 269

variable indicating whether the establishment tracks products forward was significant at the 10% 270

level. These results indicate that establishments that track their products forward are less likely to 271

fail half the Salmonella performance standard (p = .06). Whether the establishment operated 272

under HIMP did not have a statistically significant effect when controlling for other variables 273

affecting Salmonella performance (p = .80). The coefficient for whether the establishment uses 274

chemical sanitizers on hand tools had a large magnitude and indicated that use of chemical 275

sanitizers increased the likelihood of failure, but the effect was not significant (p = .18). In 276

addition, the coefficient for whether the establishment operates under New Line Speed (NELS) 277

inspection had a large magnitude and indicated that establishments under NELS inspection were 278

17

less likely to fail compared with establishments under other types of inspection, but the effect 279

was not significant (p = .14). 280

Pork Slaughter Establishments. Figure 2 shows the top nodes of the CART 281

classification tree for pork slaughter establishments. The strongest single predictor was whether 282

the establishment produces ready-to-eat (RTE) products in addition to conducting slaughter 283

activities. This initial result suggests that management philosophy of the establishment, as 284

reflected in the complexity of the products produced, could affect Salmonella control 285

performance. Establishments that do not produce RTE products were less likely to pass half the 286

Salmonella performance standard (odds ratio = 0.2). Of the establishments that do not produce 287

RTE products, establishments with a facility inspection noncompliance rate of less than 9% were 288

more likely to pass (odds ratio = 8.5). Of the establishments on the far left branch of the tree that 289

do not produce RTE products but have a low facility noncompliance rate, no establishments 290

failed if they also produce processed not ready-to-eat (NRTE) products.** For establishments that 291

do not produce RTE products and have a facility noncompliance rate of 9% or more, none of the 292

establishments that use bioluminescent testing on carcasses failed. This result may indicate these 293

establishments increased voluntary testing in response to inspection results in the past. On the 294

right branch of the tree, which indicates establishments that produce RTE products, there was 295

essentially no effect of the facility noncompliance rate on the likelihood of passing (odds ratio < 296

0.1). 297

** Note that some establishments produce only products for further processing and thus produce neither RTE or

NRTE products as defined in the survey.

18

Figure 2. Classification Tree for the Salmonella Outcome Variable for Pork Slaughter 298

Establishments 299

300 ND = not defined (cannot be calculated for 0% failure rates). 301

302

Table 5 provides the results of the factor analysis including all continuous predictor 303

variables with a nonzero relative importance level as indicated by the CART analysis. The set of 304

underlying themes, which comprise subgroups of the predictor variables, emerged in the factor 305

analysis based on the factor loadings for each set of variables. These themes included measures 306

of establishment size (number of employees, dollar sales volume, square footage, HACCP size 307

category, and number of QC/QA employees), inspection noncompliance rates (facility, HACCP, 308

and sanitation inspection procedures), frequency of microbiological testing before and after 309

fabrication of carcasses, and percentage of raw pork processed using raw pork inputs received or 310

purchased from another establishment (owned by the same or a different company). These four 311

factors were then used as regressors in the logistic regression analysis. 312

Table 6 presents two sets of logistic regression results for pork slaughter establishments. 313

In the first set on the left, only the factors identified in the factor analysis were included in the 314

regression. In the second set on the right, the same set of factors was included along with binary 315

19

Table 5. Results of Factors Analysis for the Pork Slaughter Salmonella Outcome Measure 316

Variable Factor 1.

Establishment

Size

Factor 2.

Noncompli-

ance Rates

Factor 3.

Microtesting

Frequency

Factor 4.

Use of Raw

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. 0.88 — — —

Facility noncompliance rate — 0.80 — —

HACCP noncompliance rate — 0.74 — —

Sanitation noncompliance rate 0.52 0.70 — —

Microbiological testing prior to

fabrication — — 0.85 —

Microbiological testing after

fabrication — — 0.84 —

Raw meat purchased — — — 0.84

Raw meat imported — — — 0.74

Note: Factor loadings less than 0.4 are not shown. 317

variables that had high relative importance levels in the CART analysis results. These additional 318

binary variables included whether the establishment dehairs carcasses, produces RTE product, 319

and produces NRTE processed product. 320

Based on the results of the logistic regression including only the factors, larger 321

establishments were less likely to fail half the Salmonella performance standard (p = .04), 322

20

Table 6. Results of Logistic Regression for the Pork Slaughter Salmonella Outcome 323

Measure 324

Estimates with Factors Only Estimates with Factors and Key

Binary Variables

Variable Estimate

(Standard Error)

P Value Estimate

(Standard Error)

P Value

Intercept −1.4531

(0.2665)

<0.0001 −4.3813

(0.9517)

<0.0001

Factor 1: Establishment

size

−0.6382

(0.3172)

0.0442 −0.8624

(0.3498)

0.0137

Factor 2: Noncompliance

rates

0.2404

(0.2377)

0.3120 0.1897

(0.2866)

0.5081

Factor 3: Microbiological

testing frequency

0.3960

(0.2248)

0.0781 0.3959

(0.2398)

0.0988

Factor 4: Use of raw inputs −0.6808

(0.2839)

0.0165 −0.1660

(0.3448)

0.6302

Does not dehair carcasses

(binary)

1.9431

(0.7152)

0.0066

Produces RTE product

(binary)

1.1745

(0.7299)

0.1076

Produces NRTE processed

product (binary)

1.3276

(0.6169)

0.0314

Likelihood ratio χ2 15.8711 0.0032 34.2226 <0.0001

325

21

establishments with a higher frequency of microbiological testing were more likely to fail (p = 326

.08), and establishments having a higher use of raw inputs were less likely to fail (p = .02). The 327

result for frequency of microbiological testing may indicate that establishments conduct more 328

testing if they have had problems with Salmonella control in the past. The effect of inspection 329

noncompliance rates was not significant (p = .31), and the magnitude of the coefficient was 330

smaller than for the significant factors. In terms of the relative magnitude of significant variables, 331

the frequency of use of raw pork inputs from outside sources and establishment size had similar 332

magnitudes, and the frequency of microbiological testing had a smaller but still relatively large 333

effect. 334

When the binary variables were also included in the regression, the p-values for some of 335

the factors increased compared to the previous specification. Under this specification, 336

establishment size (p = .01) and microbiological testing frequency (p = .10) were significant at 337

the 10% level in addition to all three of the included binary variables. The results for the binary 338

variables indicate that establishments that do not dehair carcasses were more likely to fail (p = 339

.01). In contrast to the CART analysis results, establishments that produce NRTE processed 340

products were more likely to fail (p = .03) as were establishments that produce RTE product (p = 341

.11). The results for whether the establishment produces RTE or NRTE processed product 342

indicate that controlling for other variables affecting establishment performance is important. 343

That is, when conducting multivariate analysis that accounts for the effects of establishment size, 344

frequency of microbiological testing, and frequency of use of raw inputs from outside the 345

establishment, the effect of whether the establishment produces further processed products 346

switched signs. 347

22

DISCUSSION 348

The results of the analyses presented above indicate that characteristics of meat and 349

poultry establishments and results of inspection procedures may be useful indicators of pathogen 350

control performance under a more risk-based inspection system. However, the variables that are 351

potential indicators are likely to vary across species in slaughter establishments. Among the large 352

number of potential predictors of pathogen control performance, only one set of variables—the 353

proportions of raw inputs purchased from outside sources—appeared to have a statistically 354

significant association with better Salmonella control in both young chicken slaughter and pork 355

slaughter establishments. Although the use of raw inputs is not directly associated with 356

Salmonella test results on raw carcasses, a higher use of raw inputs may be a reflection of the use 357

of more sophisticated production processes or management philosophy that results in better 358

pathogen control. For young chicken slaughter establishments, another significant variable 359

reflecting differences in management practices and philosophy was whether the establishment 360

tracks products forward in distribution. Young chicken slaughter establishments that track 361

products forward had better Salmonella control performance. For pork slaughter establishments, 362

whether the establishment produces RTE or processed NRTE product was associated with a 363

higher likelihood of failure, thus possibly reflecting differences in management philosophy 364

regarding pathogen control on carcasses. 365

Some of the variables that had a statistically significant effect on Salmonella control 366

performance are indirect or direct indicators of the types of production technologies used in the 367

establishment. For young chicken slaughter establishments, older production space, which likely 368

also indicates older production technologies, was associated with a higher likelihood of failure. 369

For pork slaughter establishments, larger establishments, which are likely to use more 370

23

sophisticated production technologies, were less likely to fail. Furthermore, pork slaughter 371

establishments that do not dehair carcasses were substantially more likely to fail. However, use 372

of voluntary microbiological testing in pork slaughter establishments was associated with a 373

higher likelihood of failure; this result may indicate that establishments implemented testing to 374

monitor performance in response to Salmonella test results in the past. 375

Finally, inspection results appear to be associated with Salmonella control performance in 376

young chicken slaughter establishments, but type of inspection system does not. Establishments 377

under HIMP inspection appeared to be no more or less likely to fail; the result for this variable is 378

similar to Cates et al. (7). However, young chicken slaughter establishments with a higher rate of 379

noncompliance with inspection procedures were more likely to fail. 380

The analyses presented in this paper are limited to some extent by the use of self-reported 381

survey data and the fact that food safety technologies and practices used by establishments might 382

have changed over the 3-year time period of the Salmonella test results data. In the future, more 383

objective information on the use of food safety technologies and practices (e.g., if observed by a 384

third party) might provide additional confidence in the results. Furthermore, although the pork 385

slaughter data set represents a large percentage of the total industry slaughter volume, many 386

establishments are not represented in the data set. Inclusion of a higher percentage of pork 387

slaughter establishments might provide additional insights not captured by the establishments 388

represented in the data. A larger data set for both poultry and pork slaughter establishments 389

would also likely increase the statistical significance level of important variables in the models. 390

Lastly, as establishments implement additional food safety technologies and practices and FSIS 391

further refines its testing procedures and databases, a similar analysis approach could be 392

24

implemented with new data sets to provide the most current information for risk-based 393

inspection initiatives. 394

25

ACKNOWLEDGEMENTS 395

This study was conducted under U.S. Department of Agriculture, Food Safety and 396

Inspection Service, Contract No. 53-3A94-03-12, Task Order 18. The authors gratefully 397

acknowledge comments and suggestions provided by FSIS staff. All remaining errors and views 398

expressed are those of the authors and not of the U.S. Department of Agriculture. 399

26

REFERENCES 400

1. Altekruse, S. F., N. Bauer, A. Chanlongbutra, R. DeSagun, A. Naugle, W. Schlosser, R. 401

Umholtz, and P. White. 2006. Salmonella enteritidis in broiler chickens, United States, 402

2000-2005. Emerg. Infect. Dis. 12(12):1848–1852. 403

2. Bacon, R. T., K. E. Belk, J. N. Sofos, R. P. Clayton, J. O. Reagan, and G. C. Smith. 2000. 404

Microbial populations on animal hides and beef carcasses at different stages of slaughter 405

in plants employing multiple-sequential interventions for decontamination. J. Food Prot. 406

63(8):1080–1086. 407

3. Barboza de Martinez, Y., K. Ferrer, and E. Marquez Salas. 2002. Combined effects of 408

lactic acid and nisin solution in reducing levels of microbiological contamination in red 409

meat carcasses. J. Food Prot. 65(11):1780–1783. 410

4. Berends, B. R., F. Van Knapen, J. M. A. Snijders, and D. A. A. Mossel. 1997. 411

Identification and quantification of risk factors regarding Salmonella spp. on pork 412

carcasses. Int. J. Food Microbiol. 36:199–206. 413

5. Cates, S. C., C. L. Viator, S. A. Karns, and M. K. Muth. Forthcoming. Food safety 414

practices of meat slaughter plants: findings from a national survey. Food Prot. Trends. 415

6. Cates, S. C., C. L. Viator, S. A. Karns, and M. K. Muth. Forthcoming. Food safety 416

practices of poultry slaughter: findings from a national survey. Food Prot. Trends. 417

7. Cates, S. C., D. W. Anderson, S. A. Karns, and P. A. Brown. 2001. Traditional versus 418

hazard analysis and critical control point-based inspection: results from a poultry 419

slaughter project. J. Food Prot. 64(6):826–832. 420

27

8. Duffy, E. A., K. E. Belk, J. N. Sofos, S. B. LeValley, M. L. Kain, J. D. Tatum, G. C. 421

Smith, and C. V. Kimberling. 2001. Microbial contamination occurring on lamb 422

carcasses processed in the U.S. J. Food Prot. 64(4):503–508. 423

9. Federal Register. Pathogen Reduction: Hazard Analysis and Critical Control Point 424

(HACCP) Systems; Final Rule. 61(144) (1996):38806–38815. 425

10. Gill, C. O., and C. Landers. 2003. Microbiological effects of carcass decontaminating 426

treatments at four beef packing plants. Meat Science. 65:1005–1011. 427

11. Kemp, G. K., M. L. Aldrich, M. L. Guerra, and K. R. Schneider. 2001. Continuous online 428

processing of fecal- and ingesta-contaminated poultry carcasses using an acidified 429

sodium chlorite antimicrobial intervention. J. Food Prot. 64(6):807–812. 430

12. Rivera-Betancourt, M. et al. 2004. Prevalence of Escherichia coli O157:H7, Listeria 431

monocytogenes, and Salmonella in two geographically distant commercial beef 432

processing plants in the United States. J. Food Prot. 67(2):295–302. 433

13. RTI International. 2005. Enhanced facilities database. Prepared for the U.S. Department 434

of Agriculture, Food Safety and Inspection Service. RTI International, Research Triangle 435

Park, NC. 436

14. Salford Systems, Inc. CART®. Available at: http://www.salfordsystems.com/cart.php. 437

San Diego, CA. 438

15. Sanchez, M. X., W. M. Fluckey, M. M. Brashears, and S. R. McKee. 2002. Microbial 439

profile and antibiotic susceptibility of Campylobacter spp and Salmonella spp in broilers 440

processed in air-chilled and immersion chilled environments. J. Food Prot. 65(6):948–441

956. 442

28

16. U.S. Department of Agriculture, Food Safety and Inspection Service. 2006. [Docket No. 443

04-026N] Salmonella verification sample result reporting: agency policy and use in 444

public health protection: notice and response to comments. Federal Register. 445

71(38):9772–9777. 446

17. U.S. Department of Agriculture, Food Safety and Inspection Service. 1997. Issue paper: 447

strategy for Salmonella testing. Available at: http://www.fsis.usda.gov/OA/haccp/ 448

issue02.htm. 449

18. U.S. Department of Agriculture, Food Safety and Inspection Service. 1998. Enforcement 450

instructions for the Salmonella performance standard. FSIS Directive 10,011.1. Available 451

at: http://www.fsis.usda.gov/OPPDE/rdad/FSISDirectives/10011-1.pdf. 452

19. U.S. Department of Agriculture, Food Safety and Inspection Service. 2006. FSIS 453

scheduling criteria for Salmonella sets in raw classes of product. Available at: 454

http://www.fsis.usda.gov/PDF/scheduling_criteria_salmonella_sets.pdf. 455

20. Whyte, P., J. D. Collins, K. McGill, C. Monahan, and H. O’Mahony. 2001. Quantitative 456

investigation of the effects of chemical decontamination procedures on the 457

microbiological status of broiler carcasses during processing. J. Food Prot. 64(2):179–458

183. 459

460

461

462

A-1

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.

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

A-2

Table A-1. Food Safety Performance Analysis: Meat Slaughter—Data Dictionary

Variable Label Format Format

Description Notes

Salmonella Data

HogOutcome Market hogs—Pass/Fail indicator (1/2 perf standard).

Dummy 1,0 0=Pass, 1=Fail

CattleOutcome Steers & Heifers, Cows & Bulls—Pass/Fail indicator (1/2 perf standard).


GrdBeefOutcome Ground Beef—Pass/Fail indicator (1/2 perf standard).


Survey Questions

Q1_1_1 Plant does not dehair carcasses Dummy 1,0 Missing coded as 0

Q1_1_2 Plant uses scald and rinse to dehair carcasses

Dummy 1,0 Missing coded as 0

Q1_2_2 Plant uses skinning knife to dehide carcasses


Q1_2_3 Plant uses air knife to dehide carcasses


Q1_2_4 Plant uses mechanical side puller to dehide carcasses


Q1_2_5 Plant uses mechanical down puller to dehide carcasses


Q1_2_6 Plant uses mechanical up puller to dehide carcasses


q1_3 Independent, third-party auditors conduct audits of slaughter operations


Q1_4 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

(continued)

Appen

dix A

— M

eat Slau

ghter—

Data D

ictionary

A-3

Table A-1. Food Safety Performance Analysis: Meat Slaughter—Data Dictionary (continued)


Description Notes

Q1_5n Total amount of raw product, not ground, primal cuts produced during the past year

Continuous 0–1,980,000,000

Q1_6r Routine frequency for sanitizing hands or gloves that contact raw product in the slaughter area of the plant

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

Q1_7r Routine frequency for sanitizing hands or gloves that contact raw product in the fabrication area of the plant




Q1_8a1 Technologies slaughter—Company-owned lab for microbiological testing


Q1_8b1 Technologies slaughter—Bioluminescent testing system


Q1_8c1 Technologies slaughter—Conveyor belts made from materials designed to prevent bacterial growth


Q1_8d1 Technologies slaughter—Steam pasteurization systems


Q1_8e1 Technologies slaughter—Steam vacuum units


Q1_8f1 Technologies slaughter—Organic acid rinse


(continued)

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

A-4



Description Notes

Q1_8g1 Technologies slaughter—Positive air pressure from clean side to dirty side


Q1_8i1 Technologies slaughter—Tempered carcass rinse/wash


Q1_9a1 Requires and documents that its animal growers use stipulated production practices to control pathogens


Q1_9c1 Rotates sanitizing chemicals it uses in the slaughter area on an annual basis or more frequently


Q1_9d1 Uses chemical sanitizers for food contact hand tools used in the slaughter area during operations


Q1_9e1 Uses sterilizer pots for heat sterilization of hand tools used in the slaughter area during operations


Q1_9j1 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 Plant produces ready-to-eat (RTE) products


Q2_2_2 Plant produces not-ready-to-eat (NRTE) products


Q2_2_3 Plant produces products that are inputs to further processing


Q2_6 Percentage of raw meat processed during the past year that was received or purchased from another plant


(continued)

Appen

dix A

— M

eat Slau

ghter—

Data D

ictionary

A-5



Description Notes

Q2_7 Percentage of raw meat processed during the past year that was imported as raw meat


Q2_12An Estimate total pounds during the past year. [Raw, ground meat (03B)]

Continuous 0–257,745,000

Q2_12Bn Estimate total pounds during the past year. [Thermally processed, commercially sterile (03D)]


Q2_12Cn Estimate total pounds during the past year. [Not heat treated, shelf stable (03E)]


Q2_12Dn Estimate total pounds during the past year. [Heat treated, shelf stable (03F)]


Q2_12En Estimate total pounds during the past year. [Fully cooked, not shelf stable (03G)]


Q2_12Fn Estimate total pounds during the past year. [Heat treated, but not fully cooked, not shelf stable (03H)]


Q2_12Gn Estimate total pounds during the past year. [Secondary inhibitors, not shelf stable (03I)]

Continuous 0–500,000

Section 3 of Survey: 40 plants do not conduct additional micro testing. Did not answer Section 3.

Q3_4G Organisms by frequency of microbial testing on carcasses prior to fabrication [Salmonella species]

Ordinal 1–9 41 plants that do not test coded as 1

Q3_6G Organisms by frequency of microbial testing on raw meat after fabrication [Salmonella species]

Ordinal 1–9 67 plants that do not test coded as 1

(continued)

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

A-6



Description Notes

Q4_1 Plant provides formal food safety course for newly hired production employees


Q4_3 Approximate number of production employees currently working have completed formal HACCP training

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

Q5_1 Calendar year the plant was built or renovated

Continuous 1906–2004 Used grant date from EFD for 17 plants

Q5_2 Approximate total square footage of the production space for plant

Continuous 200–1,053,175 Used production area from EFD for 23 plants

q5_3_1r Approximate percentage of the square footage of the production space of this plant that is [Under 5 years old]

Continuous 0–100 Missing coded as 0

q5_3_2r Approximate percentage of the square footage of the production space of this plant that is [5 years to just under 20 years old]


q5_3_3r Approximate percentage of the square footage of the production space of this plant that is [20 years old or more]


Q5_4 How many slaughter and fabrication shifts does this plant operate daily?

Ordinal 1–4 1) This plant does not operate on a daily basis

2) One 3) Two 4) Three

(continued)

Appen

dix A

— M

eat Slau

ghter—

Data D

ictionary

A-7



Description Notes

Q5_5 How many further processing shifts does this plant operate daily?

Ordinal 1–5 1) None 2) Further processing shift does not

operate on a daily basis 3) One 4) Two 5) Three

Q5_6 How many clean up shifts does this plant operate daily?

Ordinal 1–5 1) None 2) Clean up shift does not operate on a

daily basis 3) One 4) Two 5) Three

Q5_8 Plant has a person on staff whose primary responsibility is to manage food safety activities at the plant


Q5_9 Percentage of food safety manager’s time devoted to managing food safety activities

Ordinal 1–5 77 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

Q5_10 Plant has a quality control/quality assurance department


Q5_11 Approximately how many employees at this plant work in the plants quality control/quality assurance department?

Continuous 0–100 106 skipped out of question (No QA dept) *Missing coded as 0

(continued)

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

A-8



Description Notes

Q5_12 Number of USDA- or state-inspected plants owned by the company that owns this plant

Ordinal 1–4 1) 1 2) 2 to 5 3) 6 to 20 4) 21 or more *Missing coded as 1

EFD Data

Plant Unique Plant ID DO NOT USE IN ANALYSIS

Text

Inspection Activities

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 12,500–500,000

MULTIPLANT Indicates company ownership Dummy 1,0

EMPACTUAL Actual employment at location Continuous 1–4500 Used Q5_7 for 10 missing

ESTSALES Actual sales of location Continuous 291,000–5,796,000,000

Used midpoint of ranges of Q5_13 for 14 missing

HACCP & Inspection Information

District District number Number 5–90

Circuit Circuit number Number 15–9024 2-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


Cw_Bl Number of cows, bulls and stags slaughtered


Veal Number of veal calves and heavy calves slaughtered


(continued)

Appen

dix A

— M

eat Slau

ghter—

Data D

ictionary

A-9



Description Notes

Shp_Gts Number of mature sheep, lambs, goats slaughtered


Mkt_Hg Number of barrows and gilts slaughtered


Old_Hg Number of sows, boars and stags slaughtered


Oth_Mt Number of equines and other animals slaughtered


Inspection Systems

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 Inspection procedures—failure rate of sanitation procedures

Continuous 0–0.23 01B01, 01B02, 01C01, and 01C02 procedures

Facility Inspection procedures—failure rate of facility procedures

Continuous 0–0.62 06D01 procedure

HACCP Inspection procedures—failure rate of HACCP procedures

Continuous 0–0.73 03B01, 03B02, 03C01, 03C02, 03J01, and 03J02 procedures

Ecoli Inspection procedures—failure rate of E. coli procedures

Continuous 0–0.50 05A01 and 05A02 procedures

B-1

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.

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

B-2

Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter—Data Dictionary


Description Notes

Salmonella Data

BroilerOutcome Broiler-pass/fail indicator. half of perf std = 12/51


Survey Questions

q1_1 Independent, third-party auditors conduct audits of slaughter operations


Q1_2r Routine frequency for sanitizing hands or gloves that contact raw poultry in the slaughter area of the plant




Q1_3r Routine frequency for sanitizing hands or gloves that contact raw poultry in the deboning area of the plant




Q1_4 Percentage of live birds slaughtered at plant during the past year were imported


Q1_5n Total amount of raw product, not ground, primal cuts produced during the past year

Continuous 0–3,000,000,000

(continued)

Appen

dix B

— Y

oung C

hicken

Slau

ghter—

Data D

ictionary

B-3

Table B-1. Food Safety Performance Analysis: Young Chicken Slaughter—Data Dictionary (continued)


Description Notes

Q1_6a1 Technologies slaughter-company-owned lab for microbiological testing


Q1_6b1 Technologies slaughter-bioluminescent testing system


Q1_6c1 Technologies slaughter-conveyor belts made from materials designed to prevent bacterial growth


Q1_6d1 Technologies slaughter-inside-outside bird washers


Q1_6e1 Technologies slaughter-organic acid rinse


Q1_6g1 Technologies slaughter-automatic bird transfer (from kill line to evisceration line)


Q1_7a1 Requires and documents that its animal growers use stipulated production practices to control pathogens


Q1_7c1 Rotates sanitizing chemicals it uses in the slaughter area on an annual basis or more frequently


Q1_7d1 Uses chemical sanitizers for food contact hand tools used in the slaughter area during operations


Q1_7e1 Uses sterilizer pots for heat sterilization of hand tools used in the slaughter area during operations


Q1_7j1 Identifies and tracks its products, by production lot, backward to specific animal growers


(continued)

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

B-4



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 Plant produces ready-to-eat (RTE) products


Q2_2_2 Plant produces not-ready-to-eat (NRTE) products


Q2_2_3 Plant produces products that are inputs to further processing


Q2_6 Percentage of raw poultry processed during the past year that was received or purchased from another plant


Q2_7 Percentage of raw poultry processed during the past year that was imported as raw poultry


Q2_11An Estimate total pounds during the past year. [Raw, ground meat (03B)]


Q2_11Bn Estimate total pounds during the past year. [Thermally processed, commercially sterile (03D)]

Continuous 0

(continued)

Appen

dix B

— Y

oung C

hicken

Slau

ghter—

Data D

ictionary

B-5



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 Estimate total pounds during the past year. [Heat treated, shelf stable (03F)]


Q2_11En Estimate total pounds during the past year. [Fully cooked, not shelf stable (03G)]


Q2_11Fn Estimate total pounds during the past year. [Heat treated, but not fully cooked, not shelf stable (03H)]


Q2_11Gn Estimate total pounds during the past year. [Secondary inhibitors, not shelf stable (03I)]


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)

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

B-6



Description Notes

Q3_5D Organisms by frequency of microbial testing on raw meat after 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

Q4_1 Plant provides formal food safety course for newly hired production employees


Q4_3 Approximate number of production employees currently working have completed formal HACCP training

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

Q5_1 Calendar year the plant was built or renovated

Continuous 1947–2004 Used grant date from EFD for 1 plants

Q5_2 Approximate total square footage of the production space for plant

Continuous 700–600,000 Used production area from EFD for 7 plants

q5_3_1r Approximate percentage of the square footage of the production space of this plant that is [Under 5 years old]


q5_3_2r Approximate percentage of the square footage of the production space of this plant that is [5 years to just under 20 years old]


(continued)

Appen

dix B

— Y

oung C

hicken

Slau

ghter—

Data D

ictionary

B-7



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]


Q5_4 How many slaughter and fabrication shifts does this plant operate daily?

Ordinal 1–4 1) This plant does not operate on a daily basis

2) One 3) Two 4) Three Missing coded as 1

Q5_5 How many deboning shifts does this plant operate daily?

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

Q5_6 How many further processing shifts does this plant operate daily?

Ordinal 1–5 1) None 2) Further processing shift does not

operate on a daily basis 3) One 4) Two 5) Three

Q5_7 How many clean up shifts does this plant operate daily?

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

Q5_9 Plant has a person on staff whose primary responsibility is to manage food safety activities at the plant


(continued)

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

B-8



Description Notes

Q5_10 Percentage of food safety manager’s time devoted to managing food safety activities

Ordinal 1–5 33 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

Q5_11 Plant has a quality control/quality assurance department


Q5_12 Approximately how many employees at this plant work in the plants quality control/quality assurance department?

Continuous 0–113 2 skipped out of question (No QA dept) *Missing coded as 0

Q5_13 Number of USDA- or state-inspected plants owned by the company that owns this plant

Ordinal 1–4 1) 1 2) 2 to 5 3) 6 to 20 4) 21 or more *Missing coded as 1

EFD Data

Plant Unique plant ID DO NOT USE IN ANALYSIS

Text

Business Information

POPCODE Midpoint of population within the establishment city

Continuous 12,500–175,000

MULTIPLANT Indicates company ownership Dummy 1,0

EMPACTUAL Actual employment at location Continuous 6–3000 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)

Appen

dix B

— Y

oung C

hicken

Slau

ghter—

Data D

ictionary

B-9



Description Notes

HACCP & Inspection Information

District District number Number 5–90

Circuit Circuit number Number 15–9024

Young Chicken Slaughter Volumes

Yng_Chick Number of young chickens slaughtered


Oth_Chick Number of capons, light and heavy fowl slaughtered


Yng_Turk Number of young turkeys slaughtered


Oth_Turk Number of roaster, breeder, old breeder turkeys slaughtered


Oth_Pltry Number of ducks, geese, rabbits, slaughtered


Inspection Systems

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 Inspection procedures—failure rate of sanitation procedures

Continuous 0–0.29 01B01, 01B02, 01C01, and 01C02 procedures

Facility Inspection procedures—failure rate of facility procedures

Continuous 0–0.63 06D01 procedure

HACCP Inspection procedures—failure rate of HACCP procedures

Continuous 0–0.14 03B01, 03B02, 03C01, 03C02, 03J01, and 03J02 procedures

Ecoli Inspection procedures—failure rate of E. coli procedures

Continuous 0–0.15 05A01 and 05A02 procedures

C-1

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:

Analysis o

f Food S

afety Perform

ance in

Meat an

d Po

ultry E

stablish

men

ts

C-2

Figure C-1. Classification of Cattle Slaughter Establishments with Respect to the Beef Carcass Outcome Measure

QCW_BL_L <= 5.153

Class Cases %0 51 100.01 0 0.0

N = 51

QCW_BL_L <= 5.212

Class Cases %0 0 0.01 1 100.0

N = 1

QCW_BL_L > 5.212

Class Cases %0 20 100.01 0 0.0

N = 20

ESTSALES_L <= 15.439

Class Cases %0 20 95.21 1 4.8

N = 21

HACCP <= 0.003

Class Cases %0 7 100.01 0 0.0

N = 7

COTH_MT = (1)

Class Cases %0 2 100.01 0 0.0

N = 2

CQ1_9D1 = (0)

Class Cases %0 6 75.01 2 25.0

N = 8

SANITATION <= 0.007

Class Cases %0 1 100.01 0 0.0

N = 1

FACILITY <= 0.057

Class Cases %0 1 100.01 0 0.0

N = 1

HACCP_TRAINED <= 0.021

Class Cases %0 1 100.01 0 0.0

N = 1

HACCP_TRAINED > 0.021

Class Cases %0 1 8.31 11 91.7

N = 12

FACILITY > 0.057

Class Cases %0 2 15.41 11 84.6

N = 13

SANITATION > 0.007

Class Cases %0 3 21.41 11 78.6

N = 14

CQ1_9D1 = (1)

Class Cases %0 4 26.71 11 73.3

N = 15

COTH_MT = (0)

Class Cases %0 10 43.51 13 56.5

N = 23

HACCP > 0.003

Class Cases %0 12 48.01 13 52.0

N = 25

ESTSALES_L > 15.439

Class Cases %0 19 59.41 13 40.6

N = 32

ESTSALES_L <= 17.825

Class Cases %0 39 73.61 14 26.4

N = 53

ESTSALES_L > 17.825

Class Cases %0 29 100.01 0 0.0

N = 29

QCW_BL_L > 5.153

Class Cases %0 68 82.91 14 17.1

N = 82

Class Cases %0 119 89.51 14 10.5

N = 133

Appendix C — Classification Tree for Beef Carcasses and Ground Beef Produced in Cattle Slaughter Establishments

C-3

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 = 17.82 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 = 15.44 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 annually

22.0%

ESTSALES_L Log of midpoint of establishment sales category 21.7%

Q4_3 Number of production employees who have completed formal HACCP training

14.5%

EMPACTUAL Number of employees at the establishment 13.6%

ST_HF_L Log of number of steers and heifers slaughtered annually

12.5%

Q1_2_4 Establishment uses mechanical side puller to dehide carcasses

7.7%

Q5_12 Number of USDA- or state-inspected establishments owned by the company that owns the establishment

5.0%


C-4

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 = 13.31 in log scale),

annual number of steers and heifers slaughtered less than 1,155,449 (ST_HF_L ≤ 13.96 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.

Appendix C — Classification Tree for Beef Carcasses and Ground Beef Produced in Cattle Slaughter Establishments

C-5

Figure C-2. Classification of Cattle Slaughter Establishments with Respect to the Ground Beef Outcome Measure

HACCP <= 0.01

Class = 0Class Cases %

0 38 100.01 0 0.0

N = 38



0 5 100.01 0 0.0

N = 5

CQ1_9A1 = (0)


0 2 18.21 9 81.8

N = 11

CQ1_9A1 = (1)


0 1 100.01 0 0.0

N = 1

Q2_7 <= 2.50

Class Cases %0 3 25.01 9 75.0

N = 12

Q2_7 > 2.50


0 1 100.01 0 0.0

N = 1

ST_HF_L <= 13.96

Class Cases %0 4 30.81 9 69.2

N = 13

ST_HF_L > 13.96


0 2 100.01 0 0.0

N = 2

Q5_2_L <= 13.31

Class Cases %0 6 40.01 9 60.0

N = 15

Q5_2_L > 13.31


0 2 100.01 0 0.0

N = 2

CECOLI = (0)

Class Cases %0 8 47.11 9 52.9

N = 17

CECOLI = (1)


0 3 100.01 0 0.0

N = 3


Class Cases %0 11 55.01 9 45.0

N = 20

CQ1_1_1 = (1)

Class Cases %0 16 64.01 9 36.0

N = 25

CQ1_1_1 = (0)


0 7 100.01 0 0.0

N = 7


Class Cases %0 23 71.91 9 28.1

N = 32



0 10 100.01 0 0.0

N = 10

HACCP > 0.01

Class Cases %0 33 78.61 9 21.4

N = 42

Class Cases %0 71 88.81 9 11.3

N = 80


C-6

Table C-2. Relative Importance of the Predictor Variables for Beef Carcasses

Variable Description Relative Importance

HACCP Percentage of HACCP procedures that resulted in noncompliance reports

29.3%

Q4_3 Number of production employees who have completed formal HACCP training

19.7%

Q1_7R Frequency of sanitizing hands or gloves that contact raw product in the fabrication area of the establishment

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%

D-1

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, 2007. The analysis was later revised and finalized for this report.


D-2

Appendix D — Presentation on Results of the Analysis

D-3


D-4


D-5


D-6


D-7


D-8


D-9


D-10


D-11


D-12

Analysis of Food Safety Performance in Meat and … 2007 Analysis of Food Safety Performance in Meat and Poultry Establishments Revised Final Report Contract No. 53-3A94-03-12, Task

Documents