Effects of Antimicrobial Interventions on Indicator Organisms during Beef Carcass Dressing

Of 672 federally inspected slaughter establishments initially identified as producing hide-off beef and/or veal, 179 were chosen for sampling based on a slaughter production volume of at least 1,000 head per year. These establishments were assigned to strata according to slaughter volume: stratum 1 included those slaughtering more than 1 million head per year, stratum 2 slaughtered 100,000 to 1 million head per year, and stratum 3 slaughtered 1,000 to 99,999 head per year. Samples were collected between August 2014 and December 2015. Throughout the study period, many establishments changed slaughter volume. These establishments and their respective data were eliminated from the baseline study. At the end of the study, 139 (78%) establishments remained: 13 in stratum 1, 30 in stratum 2, and 96 in stratum 3.

The original design of the BVCBS study included veal in a single stratum containing 274 pairs of sample sets. Results on pathogen prevalence and indicator numbers in veal were significantly higher than those in beef (40), so beef and veal data were not combined. For veal as a separate commodity, differences among effects of interventions failed to achieve statistical significance, and we have not reported the comparisons here. This lack of significance results from lower statistical power derived from a small sample size.

FSIS inspectors collected swab (sponge) samples from beef carcasses in a standardized area. Each swab sampled 4,000 cm² of carcass surface area. For each carcass, a set of two swab samples was taken: one at the posterior area and the other at the anterior area. Thus, each set of swabs represented 8,000 cm². For each set, the two swabs were placed into the same plastic bag for overnight shipping on cold packs to the FSIS Midwestern Laboratory (St. Louis, MO). In this article, each combined set of anterior and posterior swabs is considered one sample.

For each carcass randomly selected for sampling, FSIS inspectors collected a set (posterior and anterior carcass regions) of swab samples at each of two points in the slaughter process: post–hide removal and prechill. Random selection was performed using a random number generator or other procedure to prevent selection bias. Post–hide removal samples were taken from the leading side of the carcass, after the hide was removed, and before evisceration. Prechill samples were taken from the trailing side of the carcass, after evisceration, 1 to 5 min after the last antimicrobial intervention on the slaughter floor was applied (to allow liquids to drip off), and no later than 1 h inside the blast chiller, commonly known as the hot box. This resulted in four separate swabs from each carcass: one anterior at post–hide removal, one posterior at post–hide removal, one anterior at prechill, and one posterior at prechill.

Sample sets were coded to allow matching of data for post–hide removal and prechill paired samples taken from the same carcass. The data originally reported for the BVCBS included samples taken from 1,368 beef carcasses (40). Because of missing information, we could only match 1,135 coded pairs. This manuscript describes the analysis of data for those 1,135 carcasses. This included 220 sample pairs from stratum 1, 379 sample pairs from stratum 2, and 536 sample pairs from stratum 3.

In the establishments, FSIS inspectors documented the application of various interventions to beef carcasses at post–hide removal and at prechill for each sample collected. These included a temperature-specific handheld steam vacuum, lactic acid carcass wash cabinet, acetic acid carcass wash cabinet, PAA carcass wash cabinet, hypobromous acid carcass wash cabinet, lactic acid carcass handheld wash, acetic acid carcass handheld wash, PAA handheld wash, hypobromous acid handheld wash, other antimicrobial carcass wash (cabinet or handheld), steam cabinets, temperature-specific hot water carcass wash, trimming, chlorinated water, no intervention, and other interventions. Interventions were recorded for each sample collected, at post–hide removal, and prechill.

Samples were analyzed for pathogens, including Salmonella, E. coli O157:H7, and non-O157 STEC (O26, O45, O103, O111, O121, and O145). Samples were analyzed qualitatively for pathogen occurrence according to the corresponding chapters of the Microbiology Laboratory Guidebook (MLG; current versions available online at: https://www.fsis.usda.gov/wps/portal/fsis/topics/science/laboratories-and-procedures/guidebooks-and-methods/microbiology-laboratory-guidebook/microbiology-laboratory-guidebook): Salmonella MLG Chapter 4.08, E. coli O157:H7 MLG Chapter 5.09 (archived), and STEC MLG Chapter 5B.05 (archived). Samples were also analyzed quantitatively for four indicator organisms: AC, Enterobacteriaceae, coliforms, and E. coli, using bioMérieux TEMPO (Firenze, Italy) as provided in MLG Chapter 3.12, with results reported as MPN. The TEMPO system is a fully automated platform designed to estimate indicator organism concentrations. Indicator organism data were additionally log transformed (log MPN/100 cm²).

Statistical analyses were conducted using JMP software version 13.1.0 (SAS Institute, Cary, NC). Logistic regression was used to investigate the association between indicator organisms and pathogen occurrence. Salmonella, E. coli O157:H7, and non-O157 STEC occurrence were considered dependent variables and characterized as binary (i.e., positive or negative), whereas the log-transformed indicator organisms were independent variables and continuous. To investigate the effect of interventions on indicator organisms, a two-tailed t test was used to compare the mean value of the log change in indicator organism levels to zero or another category. For all statistical analyses, P < 0.05 was considered a statistically significant result. Sample size was an important consideration in our analysis. When the number of observations (sample size) for a given intervention was under 24, we did not pursue the analysis because of the limited number of observations.

We used JMP software version 13.1.0 to create modified violin plots to graphically represent the effects of slaughter interventions on indicator organisms (26). Violin plots are similar to box-and-whisker plots, but the shape of the violin is particularly useful for showing the distribution of the data, especially the concentration of observations around the median and outliers. Wider sections represent more observations, whereas narrower sections indicate fewer observations. As in the more familiar box-and-whisker plots, we represent the median of the data with a horizontal line segment; however, we substituted individual observations as points on the plots instead of the traditional box-and-whisker plots. In our case, this helps to compare the differing number of observations among plots, as well as to visualize outliers.

All assayed pathogens occurred at low rates at prechill. A summary is shown in Table 1. Some numbers are slightly different from those originally reported for the BVCBS (40), because the numbers listed here describe only the 1,135 matched, coded sample pairs.

Prevalence is based on prechill data. The estimated pathogen prevalence differs substantially from measured occurrence, because prevalence is weighted by production volume. National prevalence was calculated during the FSIS BVCBS (40). For such qualitative assays, a negative result for a sample reports that the number of target organisms present is below the lower limit of detection (LLOD) for the assay (0.28 CFU/100 cm²). The kits used for this study are commercially available to be used with the BAX system (Hygiena, Camarillo, CA). The LLOD was defined by the manufacturer and validated by Association of Official Analytical Chemists and FSIS. Indicator organisms were present with higher occurrence, as summarized in Table 2.

The quantitative data (MPN) are interval censored, because quantitative assays exhibit both lower and upper limits. Thus, where an indicator was not detected, instead of zero, we assigned the data a value equal to one-half the lower limit of quantitation (LLOQ) (43). The laboratory assay LLOQ for indicators is around 10 MPN/mL (the number varies among target indicators), and half that value equates to 0.49 log MPN/100 cm².

When we compared pathogen occurrence among the various interventions, no differences achieved statistical significance (data not shown). This is due to the relatively low occurrence of pathogens at prechill, with only 16 STEC (the total of E. coli O157:H7 and non-O157 STEC) and 38 Salmonella positives in 1,135 samples, divided among numerous interventions. We attempted to enumerate pathogens, but few samples had pathogens at levels greater than or equal to the assay LLOQ (40). We had anticipated these results, based on preliminary data, and we proceeded to analyze indicator data. Notwithstanding arguments about the relevance of various indicators, we chose to focus our analysis on AC, because it provided the largest range of values above LLOQ.

Using logistic regression, the log values of AC were associated with the presence of Salmonella and non-O157 STEC at prechill and hide removal and with the presence of E. coli O157:H7 at prechill. These relationships, along with previous research, suggest that indicator data can be used to investigate the effectiveness of interventions on pathogens. An example visualization of the relationship at hide removal is shown in Figure 1. Median log AC for Salmonella-negative samples was 3.9, whereas median log AC for Salmonella-positive samples was 4.2. These data are widely distributed, with AC for both Salmonella-positive and Salmonella-negative samples ranging from almost 0 to 8 log. Thus, it is possible to have a Salmonella positive with a low log AC (lower right) or a Salmonella negative with a high log AC (upper left).

Williams et al. (44) have suggested that the level of indicators is not necessarily a good measure of process control and that reduction in indicators is a better reflection. This is because variation in incoming microbial load can affect final microbial load more than interventions. Accordingly, we analyzed paired samples to measure the reduction in indicators afforded by the process. We calculated AC reduction as log AC at post–hide removal (before evisceration and interventions) minus log AC at prechill (after evisceration and final antimicrobial interventions) for each paired sample. Differences of less than zero occasionally resulted for carcasses that had higher AC after application of interventions.

According to the questionnaire completed by the inspectors, few slaughter operations reported no intervention use at prechill, and the number of samples that we obtained for zero interventions was small and statistically weak. We assumed that with no interventions, there was no change in log AC (i.e., the difference is equal to zero).

Figure 2 illustrates how production volume (stratum) varied with reduction in log AC. Compared with large (stratum 1) and medium-sized (stratum 2) establishments, the smallest volume establishments (stratum 3) achieved smaller reductions in log AC through interventions. Although stratum 1 (median log reduction of AC 2.6) was not statistically different from stratum 2 (median 2.7), the difference between these two strata and stratum 3 (median 1.2) was statistically significant (P < 0.0001). Stratum 1 and 2 establishments had lower levels of log AC at prechill and post–hide removal and the greatest log reduction after all interventions were applied. In agreement with these AC reduction data, we observed that most pathogen positives found in this study were associated with production volume stratum 3. Specifically, 89% of Salmonella, 100% of E. coli O157:H7, and 90% of non-O157 STEC samples came from stratum 3 establishments (40).

Most establishments apply several interventions in a multihurdle approach (47). We observed that samples subjected to a larger number of interventions were associated with increased log AC reduction (Fig. 3). As in Figure 2, we assumed that with zero interventions, there was no reduction in log AC. With one intervention, median AC reduction was 1.2 log lower than for zero interventions. With two interventions, median AC reduction was 1.6 log lower than for zero interventions. With three interventions, the difference was 2.3 log, and with four or more interventions, median AC reduction was 2.5 log lower than for zero interventions. Comparing these measurements pairwise, we observed significant improvement in AC reduction as the number of interventions increased from zero to one, from one to two, and from two to three (P < 0.01). The difference between three and four interventions was not statistically different. We were not able to discern from the data whether a particular multihurdle intervention combination was most effective.

We found that the number of interventions at prechill increased with production volume stratum, with higher-volume establishments more likely to provide a larger number of interventions. Similarly, although many establishments did not apply a hide-on intervention, we found that higher-volume establishments were more likely to apply at least one intervention before hide removal.

Notwithstanding that most interventions were used in a multihurdle approach, some establishments used some interventions singly. This allowed us to investigate the correlation between log AC reduction and individual interventions. To do this we took the subset of data represented by the first grouping in Figure 3 (for a single intervention) and plotted the data according to the specific intervention applied (Fig. 4). One effective intervention, hot water, is simple to apply and relatively commonly used. A temperature-controlled hot water carcass wash produced a median 1.9 log decrease in log AC. Other effective interventions were lactic acid cabinet wash (median 1.4 log AC reduction), and lactic acid handheld wash (median 1.2 log AC reduction), all with P < 0.0001 (pairwise t test against zero interventions). Although these results were statistically significant compared with zero, they were not statistically different from one another.

Although this study included data for 1,135 samples, few samples had been subjected to a single intervention. Thus, we lacked the sample size to determine effectiveness of many interventions, because they were infrequently used singly. Specifically, we collected fewer than 24 data points for bromine, chlorine, PAA, steam cabinet, steam vacuum, trimming, and other antimicrobials. The other category (median 0.6 log AC reduction) included various cocktails containing lactic and citric acids with other components. When any individual mixture is considered an intervention, statistical power is lost because of the small sample size. Among the interventions for which we had at least 24 data points, only acetic acid handheld wash was found to not be statistically different from zero when used singly (median 0.4 log AC reduction).

As indicated earlier, most of our data represented multiple hurdle approaches for process control. Although measuring the effectiveness of multiple interventions provides less specific information, it is practical to assess slaughter process interventions as practiced. Figure 5 illustrates the effectiveness of multiple interventions. In Figure 5, samples may be represented multiple times: once for each intervention used to produce the sample. In addition, these data do not represent individual contributions of each intervention toward the total effect; rather, they represent the total effect of all interventions applied, including the one listed. We found that almost all multiple interventions were effective (P < 0.0001) compared against zero improvement in log AC, but we could not determine whether a particular multihurdle intervention combination was most effective (Table 3). There were two exceptions. Interventions containing an acetic acid wash and the hypobromous handheld wash did not have an adequate sample size for statistical analysis.

Hot water and lactic acid were among the effective single interventions, whereas multihurdle interventions were associated with a reduction in AC of 1.6 to 2.9 log. Scientific literature that shows a correlation between indicators and pathogens has not always been consistent (41, 44). This reflects a challenge when comparing quantitative indicator data with qualitative pathogen data, which is likely a result in proper sanitary dressing procedures in U.S. facilities that significantly limit the amount of cross-contamination from hides and ingesta. Establishments that demonstrate a lower log reduction in AC were more likely to have a Salmonella-positive rate (P < 0.0001). That is, the interventions may have large effects on the Salmonella count while rarely killing all Salmonella organisms (data not shown).

In addition, the value of the before-after difference in AC ranges between −4 and 6 log. This is partly because AC reduction represents the difference between two sets of data that each range between 0 and 8 log (i.e., post–hide removal and prechill). Regardless of the inherent imprecision of microbiology, differences in the effectiveness of dressing interventions, and unavoidability of random noise in experimental design, such a large range in data suggests poor control of experimental variables. Establishments are required under FSIS regulations to maintain scientific and in-plant data to support the procedures they use, including application of antimicrobial interventions, and to use them according to the supporting science while assessing all pertinent critical operational parameters. However, control of intervention application is not the role of FSIS, and this experiment represents field-collected rather than experimentally generated data. Because each of the tested intervention strategies produced such a range of effectiveness in controlling indicator levels, it seems likely that how interventions are applied, including coverage, sequence, and concentration, may be as important as which interventions (and combinations of interventions) are applied.

Interventions may be applied at any point during slaughter and processing. Those performed on live animals, for reasons of humane handling, are limited. Low-pressure water wash at room temperature and squeegee use are effective for breaking down gross contamination from mud and feces. Bacteriophage may be used safely on live animals (3). Phages are only effective as targeted interventions to address a particular pathogen, and effects are generally minimal (39).

After stunning and before hide removal, additional interventions may be performed with the hide on. With hide-on status, we recorded dehairing; bacteriophages; washing with hot, medium, or cold water; caustic soda; chlorine; lactic acid carcass wash; acetic acid carcass wash; PAA carcass wash; and other interventions, in addition to no intervention. Interventions performed at this stage help to reduce microbial count of carcasses entering the relatively high cross-contamination hazard slaughter phase of hide removal. Because we did not perform sampling of carcasses with the hide on (after these interventions but before hide removal), the effects of hide-on interventions were additive with those performed at post–hide removal (preevisceration) and could not be separately determined. Çalicioğlu et al. (14) found that interventions vary in their effect on indicators on hides, from 0 to 3.6 log MPN/cm², but had no significant effect on carcasses. This is likely because cattle hides are dirty, with a potential for large reduction in microbial load. In contrast, hide-off carcasses are less contaminated than cattle hides, if the slaughter establishment follows proper sanitary dressing procedures. Proper sanitary dressing prevents most contamination in beef slaughter. When contamination does occur, interventions may be used to decrease the incidence of pathogens. In addition, using indicator reduction as a measure of process control becomes a less effective strategy as the starting concentration of the indicator organism decreases. Thus, the additive effectiveness of sanitary dressing, hide-on interventions, and post–hide removal interventions decreased the sensitivity of our measurements.

Figure 4 provides evidence that simple hot water and lactic acid, when used as a single intervention, are effective post–harvest interventions for beef carcasses. We did not observe a statistically significant difference between the results for hot water and those for lactic acid on log AC difference. In addition, the minimum temperature and pressure for application of hot water were poorly defined in this experiment. Many others have found hot water wash to be effective (4, 7, 18, 20, 24, 36, 46). Some operations may find it difficult to manage the necessary volume of very hot water, because this requires significant space, heating-and-cooling capacity, fresh water, and wastewater. For these establishments, lactic acid may be another choice, whether applied using a hand sprayer or applied using a cabinet (15). Although acetic acid is low cost and easy to apply, acetic acid was not shown in this study to be as effective in reducing log AC (25). Acetic acid was used only by establishments in stratum 3, with the lowest production volume.

Spray cabinets provide consistent application of antimicrobial intervention, hands-off automation, and recycling of spray. However, they require space and significant investment and may be more suitable for larger establishments. Limited to hand application systems, smaller operations can find that reproducibility is challenging. Hand-spray operators must apply a sufficient quantity of spray to wet all surfaces of the carcass and achieve full coverage. For training and verification of consistent application, an indicator paper, such as pH paper, may be used to verify presence of sufficient antimicrobial volume and coverage.

In conclusion, hot water and lactic acid, when used as a single intervention, performed well among antimicrobial interventions analyzed in this study. We observed that indicator organisms were associated with the presence of pathogens. Measurement of paired samples taken from the two halves of the same carcass allowed us to assess intervention effectiveness according to the reduction in log AC. This is a better measure of process control than any single-point measurement. Although most antimicrobials studied were effective, hot water and lactic acid produced the greatest reduction in AC when used as a single intervention. Most establishments applied a multihurdle approach, and their use of multiple interventions lowered AC more effectively than did fewer interventions. Finally, we noted a large range in effectiveness for all interventions. This suggests an important aspect of process control, which is that how interventions were applied was at least as important as which interventions were applied.

The authors acknowledge the many dedicated and effective FSIS staff involved in conducting the BVCBS, specifically the inspectors who collected the samples and the laboratory scientists who performed the work to produce these data.