Traditional Versus Hazard Analysis and Critical Control Point Based Inspection: Results from a Poultry Slaughter Project

Transcription

1 826 Journal of Food Protection, Vol. 64, No. 6, 2001, Pages Copyright q, International Association for Food Protection Traditional Versus Hazard Analysis a Critical Control Point Based Inspection: Results from a Poultry Slaughter Project SHERYL C. CATES, 1 * DONALD W. ANDERSON, 1 SHAWN A. KARNS, 1 AND PATRICIA A. BROWN 2 1 Research Triangle Institute, 3040 Cornwallis Road, Research Triangle Park, North Carolina 27709; a 2 Biovet, Inc., 3055 Old Highway 8, Suite 100, St. Anthony, Minnesota 55418, USA MS : Received 25 October 2000/Accepted 14 January 2001 ABSTRACT Federal meat a poultry inspection has changed little since the Federal Meat Inspection Act was passed in 1906, followed by the Poultry Products Inspection Act of 1957 a related amements. These acts maate sensory or organoleptic (sight, smell, a touch) inspection of all carcasses. For several decades, the U.S. Department of Agriculture s Food Safety a Inspection Service (FSIS) has been urged by various organizations to move to a scienti c, risk-based inspection system. In partial response to these calls, the FSIS has developed new slaughter inspection models that are currently being tested with volunteer plants in the hazard analysis a critical control point (HACCP) based inspection models project. To evaluate whether plants operating uer the new inspection models perform at least as well as they did uer the current or traditional system, microbial a organoleptic data are being collected before a after the implementation of the new inspection models. In this article, we describe the baseline a models data collection procedures a present the results of the baseline a models data collection for eight plants that slaughter young chickens. The results from the rst eight volunteer plants suggest that inspection uer the new models is equivalent a in some ways superior to that of traditional inspection. This pilot project suggests that new slaughter inspection systems, which rely on HACCP principles with FSIS oversight a veri cation services, can maintain or even improve food safety a other consumer protection coitions relative to traditional has-on inspection methods. In the mid-1980s, the National Academy of Science (NAS) released two reports calling for a risk-based approach to meat a poultry inspection (5, 6). The NAS fou that current inspection methods do not adequately target a reduce microbial pathogens such as Salmonella on raw meat a poultry. In 1998, the NAS released a report titled, Ensuring Safe Food from Production to Consumption (7). Congress requested this study to determine the scienti c a organizational needs of an effective food safety system. In its study, the NAS described the inspection of all carcasses as an impediment to improving the safety of meat a poultry. The General Accounting Of ce (GAO), in its testimony before the House of Representatives (March 1993) (13), similarly stated that the current inspection system relies primarily on sensory or organoleptic (sight, smell, a touch) inspections. Organoleptic inspection is not capable of detecting microbial pathogens, which pose the greatest public health risk. The GAO also charged that the Food Safety a Inspection Service (FSIS) allocates considerable resources to activities that are not safety related. The GAO recommeed that the FSIS move to a modern, scienti c, risk-based inspection system. In summary, the NAS, the GAO, a other experts have recommeed that the FSIS reduce its reliance on organoleptic inspection a move to prevention-oriented sys- * Author for correspoence. Tel: ; Fax: ; scc@rti.org. tems based on public health risk. This would enable redeployment of their inspection resources to better protect the consuming public from foodborne illness. The publication of the Final Rule on Pathogen Reduction a Hazard Analysis a Critical Control Point (HACCP) Systems (61 FR 38806) in 1996 was the rst step by the FSIS to address the gaps identi ed in the inspection system (9). Approximately 1 year later, in a June 10, 1997 Federal Register (62 FR 31553) notice a at a public meeting held June 24 to 25, 1997, the FSIS discussed the need to reconsider the current procedures for slaughter inspection uer HACCP (10). According to the FSIS, uer the current inspection system inspectors couct certain process control activities as part of the carcass sorting process that are not inspection activities a thus should be the responsibility of the plant, uer FSIS oversight. To address what the FSIS a the regulated iustry should do when operating uer HACCP, the FSIS developed new slaughter inspection models, which are being tested with volunteer plants in the HACCP-based inspection models project (HIMP) (11). The project will also test new FSIS food safety a other consumer protection (OCP) activities in transportation a distribution channels. HIMP is designed to test whether new government slaughter inspection procedures, applied with revised plant HACCP controls a new plant process controls, can improve food safety a increase consumer protection. Only meat a poultry plants that slaughter exclusively young, healthy, uniform animalsmarket hogs, fed cattle, or

2 J. Food Prot., Vol. 64, No. 6 MEAT AND POULTRY 827 young poultry (including turkeys)are eligible for the project. These animals comprise nearly 90% of animals slaughtered in inspected establishments (12). Eligible plants may volunteer to participate in the pilot program. Uer the new inspection models, plant personnel carry out the sorting activities previously coucted by inspection personnel. In the sorting process, the plant determines which carcasses a parts are unacceptable a should be removed from the slaughter line because they are diseased or unwholesome. Plants also revise their HACCP plans to ensure that they control a monitor potential food safety problems in slaughter a develop process control plans to address OCP defects such as breast blisters, feathers, a bruises. FSIS inspectors move off-line a couct oversight a veri cation to ensure that current food safety a OCP staards are met. In September 2000, the FSIS redesigned HIMP so that there is one carcass inspector at a xed location on the line. The results reported in this article are for samples collected uer the original model without an FSIS carcass inspector on the line. Microbial a organoleptic data are collected in the volunteer plants before the implementation of the new inspection models to document the accomplishments of the current inspection system (i.e., baseline data collection). After a transition phase, during which both plant personnel a U.S. Department of Agriculture (USDA) inspectors in effect practice their new roles, organoleptic a microbial data are collected again following the same procedures (i.e., models data collection) to provide a before a after picture. During the testing of the new inspection models, FSIS inspectors a supervisors provide oversight a veri cation to ensure that all regulatory staards are met. Volunteer plants must continue to meet the zero tolerance requirements for the species slaughtered (e.g., fecal contamination for young chickens), the microbial performance staards for Salmonella, a the criteria for biotype I generic Escherichia coli (hereafter referred to as generic E. coli) as speci ed in the pathogen reduction a HACCP nal rule. In addition, volunteer plants must meet the food safety a OCP performance staards for nished product that were developed by the FSIS using the baseline data. The FSIS repeatedly stated that it does not want any new inspection system to compromise food safety. To ensure that plants operating uer the new inspection models perform at least as well as the current system, the project study design required that an iepeent contractor collect microbial a organoleptic data before a after the implementation of the new models. This report describes the data collection procedures along with results for HIMP for eight plants that slaughter young chickens. METHODS The FSIS contracted with the Research Triangle Institute (RTI), a nonpro t research organization, to manage the baseline a models data collection, analyze the data, a report ings. The RTI started baseline data collection in August The RTI has completed baseline data collection at 16 plants that slaughter young chickens, ve plants that slaughter market hogs, a ve TABLE 1. Volunteer plants for HIMP Plant No. Plant name Plant location P192 P468 P517 P548 P758 P1249 P6505 P20245 Gold Kist Townse s of Arkansas Marshall Durbin Cagle s Choctaw Foods Rocco Farm Foods Claxton Poultry Company Cagle Keystone Foods Guntersville, Ala. Batesville, Ark. Hattiesburg, Miss. Collinsville, Ala. Carthage, Miss. Edinburg, Va. Claxton, Ga. Albany, Ky. plants that slaughter young turkeys. In February 2000, the RTI started models-phase data collection. As of September 2000, the RTI has completed models-phase data collection at eight chicken plants a two market hog plants. Table 1 lists the eight chicken plants for which baseline a models data collection has been completed. Each of these plants slaughter more than 30 million young chickens per year. Organoleptic data collection procedures. In both the baseline a models phases for the eight plants, an iepeent veterinarian observed a recorded speci c information at three points in the slaughter process: after antemortem segregation,after carcass segregation (coemned samples), a after the nal wash (passed samples). In this article, we limit our analysis to the passed samples examined after the nal wash. Passed samples are carcasses that have passed FSIS slaughter inspection. The FSIS protocol (11) speci ed that visual carcass performance data be collected on 2,000 raomly selected passed carcass samples at each volunteer plant during a 5-week period. A sample size of 2,000 will detect most coitions, with the exception of the rarest of coitions, with acceptable levels of precision. For a sample size of 2,000, if 1% of the sampled carcasses have a defect, then the estimate of percentage of defects would have a margin of error with 95% con dence of approximately 60.46% (11). The FSIS protocol (11) speci ed that an iepeent contractor collect the organoleptic data. Biovet, Inc., a subcontractor to RTI, recruited, trained, a supervised veterinarians for the organoleptic data collection. The Biovet/RTI veterinarians reviewed the FSIS self-study training materials to uersta what coitions may be present in young chickens a the gross pathology associated with these coitions (e.g., in ammatory process a airsacculitis). Before baseline data collection, the Biovet/RTI veterinarian, the FSIS veterinarian from eld operations (not af liated with the plant), the plant inspector-in-charge, a plant personnel atteed a 2-day training session led by an FSIS veterinarian. The rst day of training included a correlation session in which atteees correlated on the criteria for calling food safety a OCP coitions. The purpose of the correlation was to ensure both in-plant a across-plant consistency in the data. The seco day of training included practice data collection. As part of the training process, the FSIS veterinarian collected data from the same sample set examined by the Biovet/RTI veterinarian for the rst 2 weeks of baseline data collection. In both the baseline a models phases for the eight plants, the Biovet/RTI veterinarian observed 80 carcasses a day 5 days a week for a 5-week period, for a total of 2,000 passed samples at each young chicken plant. As speci ed in the FSIS protocol (11), carcasses were raomly selected (e.g., count forward ve birds a pull the next bird) for examination after the nal wash but before the chiller. The Biovet/RTI veterinarian spent 1 to 2 min

3 828 CATES ET AL. J. Food Prot., Vol. 64, No. 6 TABLE 2. Description of localized OCP coitions for HIMP Coition Description Ingesta contamination Any recognizable particle Other contamination Any recognizable amount (e.g., unattached feathers, grease, bile remnants, whole gall bladder or spleen, etc.) Lung Whole lung or partial Oil glas Whole glas or yellow fragments Feathers One feather or more of any size Cloaca Any portion of the terminal portion of the intestinal tract with mucosal lining Bursa of fabricius A whole rosebud or identi able portion with two or more mucosal folds Esophagus Any portion with identi able mucosal lining Crop Complete crop or partial with mucosal lining Trachea Any identi able portion Hair One hair or more of any size Long shank Both coyles must be covered Breast blister Any of any size; if membrane slips or if it is rm Bruises One bruise or more $ in. (don t count areas showing only slight reddening) Lesions (tumors, synovitis, airsacculitis, generalized in ammatory process) Fractures Sores/scabs (including localized in ammatory process) External mutilation Any in ammatory tissue, exudate, or tumor (any size) Must be opening through the skin (compou fracture) Any of any size (don t count healed sores or scabs that have been removed for the most part) To skin a/or muscle; caused by equipment in the dressing process; not normally a single nick on the skin examining each passed carcass. Table 2 shows the criteria established by the FSIS for calling localized OCP coitions. These criteria differ from the nished product staards followed by FSIS uer traditional inspection. For example, for HIMP a single hair is counted as a defect, whereas uer the nished product staards guidelines hair is only considered a defect if more than 25 hairs are present. For the baseline a models data collection, multiple defects that are different coitions (for example, hair a lung) were scored as multiple defects. Also, multiple defects of the same coition (for example, three bruises) were scored as one defect. Microbial data collection procedures. In both the baseline a models phases for the eight plants, young chickens were analyzed for Salmonella a generic E. coli. The FSIS protocol (11) speci ed a sample size of 300 for both microorganisms at each of two sampling sites: pre-evisceration a postchill. The FSIS revised the protocol to reduce the period for microbial sample collection from 12 to 6 weeks (with the number of samples remaining the same) a to eliminate the requirement for pre-evisceration sampling. Biovet, Inc., recruited, trained, a supervised the technicians for the microbial data collection. Sample collection followed the procedures outlined in the FSIS protocol (11), with the exception that the postchill samples for both Salmonella a generic E. coli were acquired from a whole-bird rinse of a single carcass (as speci ed by the FSIS) rather than from two wholebird rinses. Two laboratories analyzed samples according to the requirements a procedures outlined in the pathogen reduction a HACCP nal rule (9) using methods that are approved by the AOAC International (formerly the Association of Of cial Analytical Chemists). Midwest Laboratories analyzed the baseline samples for two of the plants. Silliker Laboratories analyzed the baseline samples for the other six plants a the models-phase samples for all eight plants. Midwest Laboratories determined presumptive Salmonella with Gene-Trak nucleic acid hybridization technology (Gene-Trak Systems Corp., Hopkins, Mass.) (3). Midwest Laboratories con rmed presumptive positive results using the staard microbiological method (3). Silliker used the VIDAS SLM method (biomérieux Vitek, Inc., Hazelwood, Mo.), which is an enzyme-linked uorescent immunoassay (1), for presumptive detection of Salmonella. Silliker con rmed presumptive positive results with serological screening. The detection limit for Salmonella was 1 CFU per 25 ml. Both Midwest a Silliker used Petri lm E. coli plates to estimate total E. coli counts (2, 4). The detection limit for generic E. coli was 10 CFU/ml. Statistical analysis. In analyzing the organoleptic data, the RTI classi ed coitions as food safety coitions a OCP coitions. We recognize that there is not always a clear distinction between food safety a OCP coitions a that these distinctions may change due to new research, analysis, or regulatory actions. In this article, coitions are classi ed as food safety versus OCP coitions based on current FSIS policies a regulations. A 2-tailed z test was used to detect whether the food safety a OCP defect rates observed in the baseline a models phases were signi cantly different. We used the x 2 statistic to evaluate iividual plants on their overall performance for the 18 localized OCP coitions. A 2-tailed z test was used to detect whether the Salmonella prevalence rates observed in the baseline a models phases were signi cantly different. In analyzing the results for generic E. coli, we used a proportional odds model a the consequential x 2 test to assess whether the proportions of acceptable, marginal, a unacceptable samples (based on the FSIS performance criteria) were equal in the baseline a models phases. We used a goodness-of- t test to test the assumptions of proportional odds. SAS (8) was used for all data analysis. RESULTS Organoleptic results. Approximately 16,000 carcasses (approximately 2,000 per plant) were examined in each of the baseline a models phases at the eight plants. Table 3 shows the percentage of carcasses with food safety a OCP defects in the baseline phase a the models phase a the 95% con dence interval for the point estimate. Differences in defect rates could be partially due to factors other than the inspection models tested, such as seasonality effects or data collectors consistency in interpreting defect criteria. Signi cant reductions were observed in the models phase for two coitions considered food safety concerns:

4 J. Food Prot., Vol. 64, No. 6 MEAT AND POULTRY 829 TABLE 3. HIMP results for eight plants slaughtering young chickens: percentage of carcasses with defects Baseline (n 5 16,071 carcasses) Models (n 5 16,000 carcasses) P value Food safety coitions Septicemia/toxemia Fecal contamination (60.077) (60.167) (6) (60.070) 0 0 OCP coitions Leukosis Multiple ($2) tumors Cadaver Ascites Deep overscald Tuberculosis Ingesta contamination Other contamination Lung Oil glas Feathers Cloaca Bursa of fabricius Esophagus Crop Trachea Hair Long shank Breast blister Bruises Lesions (tumors, synovitis, airsacculitis, generalized in ammatory process) Fractures Sores/scabs (including localized in ammatory process) External mutilation (60.012) (6) (6) (6) (60.027) (6) (60.530) (60.641) (60.563) (60.573) (60.771) (60.076) (60.347) (60.284) (60.528) (60.313) (60.750) (60.277) (60.285) (60.706) (60.164) (60.263) (60.697) (60.445) (60.012) (6) (6) (6) (60.017) (6) (60.513) (60.385) (60.554) (60.521) (60.672) (60.062) (60.340) (60.201) (60.392) (60.307) (60.625) (60.523) (60.450) (60.617) (60.125) (60.230) (60.389) (60.307) septicemia/toxemia (0.25% in the baseline phase versus 0.00% in the models phase, P 5 0) a fecal contamination (1.18% in baseline versus 0.21% in models, P 5 0). For 4 of the 24 OCP coitions (multiple [$2] tumors, cadaver, ascites, a tuberculosis), the observed defect rate was zero for both the baseline a models phases. These coitions are identi ed readily during the sorting process a are therefore rarely seen in passed product. For 6 of the 24 OCP coitions (leukosis, deep overscald, lung, cloaca, bursa of fabricius, a trachea), the differences observed between the baseline a models phases were not statistically signi cant. Signi cant reductions were observed for 10 of the 24 OCP coitions. The most signi cant models improvement was observed for sores a scabs; the defect rate decreased from 28.35% in the baseline phase to 6.78% in the models phase (P 5 0). Other contamination decreased from to 6.65% (P 5 0); external mutilation decreased from 9.10 to 4.12% (P 5 0); esophagus defects decreased from 3.50 to 1.72% (P 5 0); crop defects decreased from to 6.91% (P 5 0); lesions decreased from 1.13 to 0.66% (P 5 0); bruises decreased from to 19.88% (P 5 0); fractures decreased from 2.99 to 2.27% (P 5 0); oil gla defects decreased from to 13.08% (P 5 0); a ingesta contamination decreased from to 12.57% (P ). Statistically signi cant increases were observed for only 4 of the 24 OCP coitions. Feather defects increased from 46.54% in the baseline phase to 74.69% in the models phase (P 5 0). Hair defects increased from to 79.40% (P 5 0), long shank defects increased from 3.30 to 13.16% (P 5 0), a breast blisters increased from 3.53 to 9.33% (P 5 0). Table 4 shows the minimum, median, a maximum defect rate for the eight plants. For the two food safety coitions, the maximum defect rate observed for any plant in the models phase was always lower than that observed in the baseline phase. For 18 of the 24 OCP coitions, the maximum defect rate observed at a plant in the models phase was less than or equal to those observed in the baseline phase. For 6 of the 24 OCP coitionslung, feathers, hair, long shank, breast blister, a fracturesthe maximum plant defect rate rose in the models phase. Figure 1 shows, of the 18 localized OCP coitions, the number of coitions with lower defect rates in the models phase compared with its baseline performance. We show the results for each of the eight plants (labeled A through H to maintain con dentiality), ranked from the

5 830 CATES ET AL. J. Food Prot., Vol. 64, No. 6 TABLE 4. HIMP results for eight plants that slaughter young chickens: descriptive statistics a Baseline Minimum Median Maximum Models Minimum Median Maximum Food safety coitions Septicemia/toxemia Fecal contamination OCP coitions Leukosis Multiple ($2) tumors Cadaver Ascites Deep overscald Tuberculosis Ingesta contamination Other contamination Lung Oil glas Feathers Cloaca Bursa of fabricius Esophagus Crop Trachea Hair Long shank Breast blister Bruises Lesions (tumors, synovitis, airsacculitis, generalized in ammatory process) Fractures Sores/scabs (including localized in ammatory process) External mutilation a Minimum, lowest defect rate (%) among the eight plants; median, median defect rate (%) for the eight plants; maximum, highest defect rate (%) among the eight plants. FIGURE 1. HIMP results for eight plants that slaughter young chickens: number of localized OCP coitions with lower defect rates in models phase. The localized OCP coitions (18 coitions) were ingesta contamination, other contamination, lung, oil glas, feathers, cloaca, bursa of fabricius, esophagus, crop, trachea, hair, long shank, breast blister, bruises, lesions, fractures, sores/scabs, a external mutilation. poorest to the best performing plant. Plant A performed the poorest of the eight plants, improving (i.e., had a lower defect rate) in only 5 of the 18 coitions in the models phase. Plant H had the best performance, improving in 14 of the 18 coitions in the models phase. With 18 coitions being observed in eight plants, there are 144 opportunities for defect rates to change. The overall score for the eight plants combined was 0.53 (for the eight plants combined, 77 of the 144 coitions had a lower defect rate in the models phase). We tested whether the observed value (0.53) was signi cantly different by chance observations (0.5). The results (x , P ) do not provide any evidence that the overall performance in the models phase on localized OCP defects measured at the plant level was different compared with the baseline. Although there were not signi cant improvements overall on localized OCP defects, these ings suggest that the plants performed as well in the models phase as they did at baseline. We also tested whether the values observed were different among the eight plants. The results (x , P ) were not statistically signi cant at the traditional 0.05 level but suggest that there may be differences among plants.

6 J. Food Prot., Vol. 64, No. 6 MEAT AND POULTRY 831 TABLE 5. HIMP results for eight plants that slaughter young chickens: Salmonella a Baseline Models P value Overall results (8 plants) No. of carcasses sampled % positive Plant-level results, % positive a Detection limit, 1 CFU/25 ml. 2, (60.917) (61.425) (61.458) (61.482) (61.760) (61.811) (63.392) (63.454) (63.899) 2, (60.909) (63.048) (62.621) (61.823) (63.651) (61.221) (62.207) (62.385) (62.695) Microbial results. Table 5 presents the overall results for the eight plants for the Salmonella analysis. A total of 2,438 postchill samples were analyzed in the baseline phase, a 2,587 samples were analyzed in the models phase (approximately 300 per plant in each phase). In the baseline phase, 5.66% (60.92%) of the samples were positive for Salmonella. In the models phase, 5.91% (60.91%) of the samples were positive for Salmonella. Although the Salmonella prevalence rates for the two phases were not signi cantly different (P ), there was a minor increase numerically in the models phase. This may be a re- ection of the more sensitive Salmonella detection method used in the models phase for two of the plants. The overall prevalence rates for both the baseline a models phases were well uer the pathogen reduction performance staard of 20% for young chickens. We also compared the baseline versus models-phase results on Salmonella prevalence for the iividual plants. Table 5 shows the Salmonella prevalence rates for each plant ranked from the lowest to highest rate for the baseline phase. For two of the plants, the Salmonella prevalence rates for the two phases were not signi cantly different. TABLE 6. HIMP results for eight plants that slaughter young chickens: biotype I E. coli a Baseline Models No. of carcasses sampled 2,462 2,588 Distribution (reported in CFU/ml) Minimum 25th percentile Median 75th percentile Maximum Performance staard categories, % of carcasses Acceptable (#100 CFU/ml) Marginal (100, x # 1,000 CFU/ml) Unacceptable (.1,000 CFU/ml) ,000 a Detection limit, 10 CFU/ml;, not detected , Three plants had signi cantly lower prevalence rates in the models phase compared with their baseline performance, a three plants had signi cantly higher prevalence rates in the models phase. The Salmonella prevalence rates for the iividual plants for both the baseline a models phases were well uer the 20% pathogen reduction performance staard. Differences in prevalence rates could be due to factors other than the inspection models tested, such as seasonality effects or the health of incoming animals. Table 6 presents the overall results for the eight plants for the generic E. coli analysis. A total of 2,462 postchill samples were analyzed in the baseline phase, a 2,588 samples were analyzed in the models phase (approximately 300 per plant in each phase). The median baseline-phase value was 20 CFU/ml, a the median models-phase value was 10 CFU/ml. The CFU levels ranged from not detected to a maximum of 22,000 CFU/ml for baseline a from not detected to a maximum of 32,000 CFU/ml for the models phase. Table 6 also shows the percentages of samples that were acceptable, marginal, a unacceptable based on the performance criteria established by the FSIS. In the models phase, the percentage of acceptable samples increased a the percentage of marginal a unacceptable samples decreased compared with baseline. We used a proportional odds model a the consequential x 2 test to assess whether the proportions are equal. The resulting statistic (x , P 5 1) implies that the populations do differ a that a tre toward acceptability is associated with the models phase. With samples this large, the ability to detect small deviations from the model are quite high, so the goodness-of- t statistic (x , P ) implies a reasonable t (higher P values iicate a better t). We also compared the baseline versus models-phase results for generic E. coli for the iividual plants. In seven of the eight plants, a tre toward acceptability is associated with the models phase. DISCUSSION HIMP has resulted generally in improvements in food safety a OCP coitions at the eight chicken plants where the inspection models are being tested. The prevalence of

7 832 CATES ET AL. J. Food Prot., Vol. 64, No. 6 the two food safety coitionssepticemia/toxemia a fecal contaminationsigni cantly decreased in the models phase. Improvements were also observed for some OCP coitions. For 16 of the 24 OCP coitions, the defect rates signi cantly decreased in the models phase. The most signi cant improvement was observed for sores a scabs. Comparing baseline a models-phase results, we that there may be differences among plants in their performance on localized OCP coitions. These differences may be due to changes that some plants made in the models phase. Some plants trained their personnel on the sorting process a placed personnel in the same locations that the FSIS used to have inspectors, whereas other plants made significant changes in their operations. Such changes included changing the type of slaughter equipment a strategically placing their personnel on the slaughter line. Uer the new inspection models, the eight plants continue to meet the performance staard for Salmonella, a the overall prevalence rate for the eight plants is virtually unchanged between the baseline a models phases. Improvements are observed for generic E. colithe percentage of acceptable samples (based on the FSIS performance criteria) for the eight plants signi cantly increased (77.9 versus 93.4%) in the models phase, whereas the percentage of unacceptable samples fell from 3.9 to 0.7%. The results from the rst eight volunteer plants suggest that inspection uer the new models is generally equivalent or superior to that of traditional inspection. Plant responsibility for the sorting process has resulted in passed carcasses with fewer food safety a OCP defects. Baseline a models data should continue to be analyzed to ensure that improvements in food safety a OCP coitions are observed at the other volunteer plants. This project provides early evidence that the inspection models tested by the FSIS can maintain or even improve food safety a OCP coitions relative to traditional has-on inspection methods. Various stakeholders are naturally scrutinizing the project. The consuming public, the slaughter iustry, the USDA inspection force, a agency a department managers all have a right to expector a responsibility to delivera new inspection system that maintains or improves food safety. This project is another step by the FSIS, after HACCP itself, to address the NAS, the GAO, a other third-party calls for inspection reform. Uer the new inspection models, the FSIS has reduced its reliance on organoleptic inspection a continues to move toward a scienti c, risk-based inspection system. ACKNOWLEDGMENTS The authors acknowledge the support of Dr. Christopher Wiesen of the RTI for his help with data analysis, Monica Seagroves for her review of the draft manuscript, a the support of the USDA s FSIS in preparing this article. This article was prepared by the authors alone, a any views or opinions expressed herein do not necessarily re ect those of the agency. REFERENCES 1. Anonymous Method In AOAC International of cial methods of analysis. AOAC International, Arlington, Va. 2. Anonymous Method In AOAC International of cial methods of analysis. AOAC International, Arlington, Va. 3. Anonymous Chapter V, Salmonella. In FDA bacteriological analytical manual (BAM), 7th ed. AOAC International, Arlington, Va. 4. Anonymous Chapter IV, Sample disposition. In FDA bacteriological analytical manual (BAM), 8th ed. AOAC International, Arlington, Va. 5. National Academy of Sciences Meat a poultry inspection: the scienti c basis of the nation s program. National Academy Press, Washington, D.C. 6. National Academy of Sciences Poultry inspection: the basis for a risk-assessment approach. National Academy Press, Washington, D.C. 7. National Academy of Sciences Ensuring safe food from production to consumption. National Academy Press, Washington, D.C. 8. SAS/STAT user s guide, version 6, 4th ed., vol. 1 a SAS Institute, Inc., Cary, N.C. 9. U.S. Department of Agriculture, Food Safety a Inspection Service. July 25, Pathogen reduction; hazard analysis a critical control point (HACCP) systems; nal rule. Fed. Regist. 61(144): U.S. Department of Agriculture, Food Safety a Inspection Service. June 10, HACCP-based meat a poultry inspection concepts. Fed. Regist. 62(111): U.S. Department of Agriculture, Food Safety a Inspection Service. July 7, HACCP-based inspection models project in-plant slaughter U.S. Department of Agriculture, Food Safety a Inspection Service. July FSIS backgrouer: HACCP-based inspection models project U.S. General Accounting Of ce. March 16, Building a scienti c, risk-based meat a poultry inspection system: testimony before the Subcommittees on Livestock a Department Operations a Nutrition, Committee on Agriculture, House of Representatives; statement of John W. Harmon, Director, Food a Agriculture Issues; Resources, Community, a Economic Development Division. U.S. General Accounting Of ce, Washington, D.C. GAO/T-RCED