Statistics & Data Analysis | Stasiewicz Food Safety Laboratory | Vision: A world where everyone, everywhere is food secure

Salmonellosis Risk Assessment for Comminuted Turkey under Different Specificities of Concentration-based and Virulence-based Final Product Standards

Highlights

More specifically targeting products with risky contamination is efficient
Concentration- and virulence-based standards capture more risk in less products
Genomic clustering more efficiently than serotype identifies risky contamination

Abstract

Prevalence-based performance standards have guided Salmonella control in poultry industry, but concentration- and virulence-based final product standards could target the most risky contamination more specifically. We adapted our previous risk assessment for chicken parts to comminuted turkey to assess the risk in products implicated by different final product standards, incorporating assumptions from FSIS 2024 risk assessments. We simulated the attributable fraction of illnesses from products contaminated over three level thresholds (0.0031 CFU/g, 1 CFU/g, and 10 CFU/g) and/or containing a serotype in three lists (“Top 3 most prevalent higher-virulence serotypes”, “All higher-virulence serotypes”, and “Higher-virulence proportion of each serotype”). Results showed that 87% of illnesses were attributed to the 0.56% of products with Salmonella exceeding 10 CFU/g. Under more specific criteria of level “AND” serotype, 60% of illnesses were attributed to the 0.14% of products contaminated with Salmonella exceeding 10 CFU/g and one of the three most prevalent higher-virulence serotypes. Further, applying genomic-based clustering information, 75% of illnesses were attributed to slightly more products (0.19%) containing Salmonella exceeding 10 CFU/g and higher-virulence proportion of each serotype. Under the less specific standard, however, 99% of illnesses were attributed to the 5.7% of products containing Salmonella exceeding 10 CFU/g “OR” one of the three most prevalent higher-virulence serotypes. Our study demonstrated that most salmonellosis risk is concentrated in comminuted turkey products with high levels of higher-virulence contaminations. Importantly, more specifically targeting those products could efficiently reduce public health risk while minimizing products implicated.

DOI

https://doi.org/10.1016/j.mran.2025.100359

Quantitative model predicts implementing school cafeteria share tables will not compromise milk safety

Share table model framework, where the cafeteria system consists of (1) the service, (2) the break (or breaks), and (3) overnight refrigeration. A typical day of school cafeteria meal services has n number of services, with n − 1 breaks. All milk cartons served during service or services on a given day are assumed to be in the defined storage condition for the entire day (i.e., throughout services and breaks) until a milk carton is consumed or the day is over and the milk cartons are moved to overnight refrigeration. At the end of each day, leftover milk cartons on the service line or share table can be reserviced on the following day. Each day was simulated 5 times to represent a school week. At the end of the school week, all leftover milk cartons are discarded. Each week is simulated 37 times (a typical school year). This process is repeated 50 times to simulate 50 school years of meal services in cafeterias with share tables.

Abstract

School cafeteria share tables can address food waste and improve food security by allowing students to share unopened items, such as milk. However, unresolved safety concerns present a barrier to recovering milk cartons on share tables. We adapted our previous share table model to study Listeria monocytogenes in pasteurized milk, assuming a concentration distribution that reflects the realistically low prevalence of the pathogen. Student sharing behavior was simulated for 50 yr of school weeks (5 d/wk over 37 wk/yr). Milk safety is assessed by quantifying (1) time to L. monocytogenes growth of 1 log₁₀, (2) L. monocytogenes concentration at consumption, and (3) listeriosis risk. We compare these measures across 23 what-if scenarios to identify potential risk factors and mitigation strategies. Under the baseline scenario (with no share table temperature management), L. monocytogenes increases by 1 log₁₀ after 1 reservice (after service 2). With share table temperature management, improved overnight refrigeration, or shorter services, L. monocytogenes did not increase by 1 log₁₀ until after 2 d of reservice (after service 3). Under excessive time-temperature abuse (inadequate overnight refrigeration or long services), L. monocytogenes increases by 1 log₁₀ before the first day of reservice (before service 2). Comparing the baseline scenarios of share table and no share table, L. monocytogenes concentration at consumption did not substantially differ. Importantly, L. monocytogenes concentration at consumption never exceeded 100 cfu/mL, except under the longest (266 min) service scenario, for only 0.0006% of milk cartons (11/1,794,887). The mean probability of illness per serving (P_Illness) was low across all scenarios. Comparing the baseline share table and no share table scenarios, P_Illness was 3.32 × 10⁻¹³ and 2.72 × 10⁻¹³, respectively, translating to 1 listeriosis illness in every 2,100 (2,000–2,400) and 3,000 (2,700–3,400) yr across all public schools in the United States. These results demonstrate the extremely small predicted risk of listeriosis from consuming milk cartons in cafeterias with share tables, providing an evidence base to support school nutrition staff in conversations with health departments during the approval of share table management practices.

DOI

https://doi.org/10.3168/jds.2025-27069

Development of a flexible produce supply chain food safety risk model: Comparing tradeoffs between improved process controls and additional product testing for leafy greens as a test case

Model and scenario analysis framework. The model consists of five main process stages, an initial contamination event, reductions and/or increases (Δ), and a retail sample as the risk output test. Mean (µ), standard deviation (σ), and the probability of occurrence (PO) are defined for each process stage where contamination or increase/reduction occurs. When a product test is implemented, the mass, grabs, number of tests, and probability of occurrence are defined. For the analysis, two baseline contamination scenarios were developed, high and low variability. In addition, two industry-relevant management scenarios were evaluated, improved process controls and additional product testing.

Highlights

A flexible supply chain microbial risk model for fresh produce was developed.
Probability of a positive test at retail was used as a food safety risk measure.
Leafy greens contaminated with Shiga-toxin-producing E. coli were modeled.
Improved process controls better-reduced recall risk vs. more product testing.
Additional product testing would reject lots of potentially low public health risk.

Abstract

The produce industry needs a tool to evaluate food safety interventions and prioritize investments and future research. A model was developed in R for a generic produce supply chain and made accessible via Shiny. Microbial contamination events, increases, reductions, and testing can be modeled. The output for each lot was the risk of one, 300-gram sample testing positive, described by two industry-relevant risk metrics, the overall risk of a positive test (proxy for recall risk) and the number of lots with the highest risk (>1 in 10 chance) of testing positive (proxy for public health risk). A leafy green supply chain contaminated with Shiga-toxin-producing Escherichia coli was modeled with a mean of 1 pathogen cell per pound (µ=1 CFU/lb or -2.65 Log(CFU/g)) under high (σ=0.8 Log(CFU/g)) and low (σ=0.2 Log(CFU/g)) variability. Baseline risk of a positive test in the low-variability scenario (1 in 20,000) was lower than for high-variability (1 in 4,500), showing rare high-level contamination drives risk. To evaluate tradeoffs, we modeled two well-studied, frequently used interventions: additional product testing (8 of 375-gram tests/lot) and improved process controls (additional -0.87±0.32 Log(CFU/g) reduction). Improved process controls better reduced recall risk (to 1 in 115,000 and 1 in 26,000 for low- and high-variability, respectively), compared to additional product testing (to 1 in 21,000 and 1 in 11,000 for low- and high-variability, respectively). For low variability contamination, no highest risk lots existed. Under high variability contamination, both interventions removed all highest risk lots (about 0.05% of total). Yet, additional product testing rejected more lower-risk lots (about 1% of total), suggesting meaningful food waste tradeoffs. This model evaluates tradeoffs between interventions using industry-relevant risk metrics to support decision-making and can be adapted to assess other commodities, process stages, and less-studied interventions.

DOI

https://doi.org/10.1016/j.jfp.2024.100393

Milk spoilage model predicts that share tables would not meaningfully increase spoilage, and improved storage systems can reduce spoilage

Experimental setups of data collection for milk internal temperature under different storage systems and refrigeration condition and external temperature. Both external and internal milk temperature and experimental setup were measured for an hour in duplicate (2 probes) over 4 d resulting in a total of 4 replicates.

Abstract

School share tables offer opportunities for food recovery and increased access to healthy foods by allowing students to donate or consume unopened items, such as cartons of milk. However, stakeholders have concerns about temperature abuse potentially causing premature milk spoilage. Although previous research showed short ambient temperature abuse of milk (under conditions representing share tables) does not meaningfully affect microbial milk quality, differences across school cafeterias (e.g., ambient temperatures, storage systems, bell schedules, refrigeration temperature) may limit the generalizability of this conclusion. To address this, the overnight refrigeration temperature and the milk’s initial contamination were predicted to be the main drivers for milk spoilage. Share tables were predicted to only cause inconsequential microbial quality changes (4 spoiled milk per million served, which would be ≤2 milk cartons spoiled per school year) under short and medium bell schedules (≤125 min of total service), even without temperature control during the lunch period. Under long (221 min) and very long (266 min) bell schedules, share tables with ambient temperature storage were predicted to have higher milk spoilage (19 and 42 spoiled milk cartons per million served, respectively), and adding storage systems was predicted to reduce the decline in milk quality (12 and 24 spoiled milk cartons per million served, respectively). These results provide a resource to support science-based decision making for the inclusion of milk in school cafeteria share tables, ultimately working to reduce food waste and address food insecurity.

DOI

https://doi.org/10.3168/jds.2024-25189

Risk Assessment Predicts Most of the Salmonellosis Risk in Raw Chicken Parts is Concentrated in Those Few Products with High Levels of High-Virulence Serotypes of Salmonella

Summary of collected USDA-FSIS publicly available datasets used in this project to characterize Salmonella contamination in chicken parts by presence, level, and serotype

Highlights

Chicken parts are rarely contaminated with high Salmonella levels, above 1 CFU/g.
High-virulence serotypes are frequently found in US chicken parts.
Most salmonellosis risk is concentrated in chicken part products above 1 CFU/g.
Most salmonellosis risk is in products with high levels of high-virulence serotypes.

Abstract

Salmonella prevalence declined in U.S. raw poultry products since adopting prevalence-based Salmonella performance standards, but human illnesses did not reduce proportionally. We used Quantitative Microbial Risk Assessment (QMRA) to evaluate public health risks of raw chicken parts contaminated with different levels of all Salmonella and specific high- and low-virulence serotypes. Lognormal Salmonella level distributions were fitted to 2012 USDA-FSIS Baseline parts survey and 2023 USDA-FSIS HACCP verification sampling data. Three different Dose-Response (DR) approaches included (i) a single DR for all serotypes, (ii) DR that reduces Salmonella Kentucky ST152 virulence, and (iii) multiple serotype-specific DR models. All scenarios found risk concentrated in the few products with high Salmonella levels. Using a single DR model with Baseline data (μ = -3.19, σ = 1.29 Log CFU/g), 68% and 37% of illnesses were attributed to the 0.7% and 0.06% of products with >1 and >10 CFU/g Salmonella, respectively. Using distributions from 2023 HACCP data (μ = -5.53, σ = 2.45), 99.8% and 99.0% of illnesses were attributed to the 1.3% and 0.4% of products with >1 and >10 CFU/g Salmonella, respectively. Scenarios with serotype-specific DR models showed more concentrated risk at higher levels. Baseline data showed 92% and 67% and HACCP data showed >99.99% and 99.96% of illnesses attributed to products with >1 and >10 CFU/g Salmonella, respectively. Regarding serotypes using Baseline or HACCP input data, 0.002% and 0.1% of illnesses were attributed to the 0.2% and 0.4% of products with >1 CFU/g of Kentucky ST152, respectively, while 69% and 83% of illnesses were attributed to the 0.3% and 0.6% of products with >1 CFU/g of Enteritidis, Infantis, or Typhimurium, respectively. Therefore, public health risk in chicken parts is concentrated in finished products with high levels and specifically high levels of high-virulence serotypes. Low-virulence serotypes like Kentucky contribute few human cases.

DOI

https://doi.org/10.1016/j.jfp.2024.100304

Time and temperature abuse of milk in conditions representing a school cafeteria share table does not meaningfully reduce microbial quality

Abstract

Share tables (ST) are tables or stations in school cafeterias where students can return unopened foods and beverages, providing an opportunity to access these items at no cost. Currently, research suggests that milk is among the most wasted items in breakfast and lunch programs in the United States. Share tables present a simple solution for reducing milk waste, but research is needed to understand the microbial spoilage potential of milk in ST. To this end, uninoculated milk cartons and milk cartons inoculated with 2 to 3 log₁₀(cfu/mL) Pseudomonas poae, a fast-growing psychrotroph, was exposed to ambient temperature during winter (mean temperature = 20.3°C) and summer (23.1°C) for 125 min, repeated over 5 d (the length of a school week). Microbial counts in the inoculated milk cartons increased linearly, exceeding the spoilage threshold of 6.0 log₁₀(cfu/mL) after d 3 and after d 4 in the winter and summer season trials, respectively. In the winter trial, the microbial counts for uninoculated milk cartons never exceeded the lower limit of detection, 2.31 log₁₀(cfu/mL), and in the summer trials, microbial counts never reached the spoilage threshold, indicating that initial contamination is a driving factor of microbial milk spoilage. Regardless of sharing status or seasonality, the greatest changes in counts for inoculated milk cartons occurred during overnight refrigeration, ranging from 0.56 to 1.4 log₁₀(cfu/mL), while during the share table ranged from no observable change up to 0.29 log₁₀(cfu/mL), emphasizing that school nutrition personnel should focus efforts on tightly controlling refrigeration temperatures and returning milk to refrigeration as soon as possible. A previously developed model for school cafeteria share tables was adapted to understand the typical residence time of milk in a simulated cafeteria with an ambient temperature share table for the summer and winter seasons over 1,000 wk. Milk was predicted to have a very short mean residence time (85 min) regardless of sharing status or season, with 99.8% of milk consumed, discarded, or donated within the first 2 d. As a result, only 3 out of 451,410 and 6 out of 451,410 simulated milks spoiled in the winter and summer seasons, respectively. The data generated here can be used to inform science-based decision-making for including milk in share tables, or applied to any system where one might have to accept short-term unrefrigerated storage of milk to meet a waste reduction or food security goal.

DOI

https://doi.org/10.3168/jds.2023-24175

Modeling Preharvest Cyclospora cayetanensis Sampling and Testing for Various Water and Produce Sampling Plans

Scenario Analysis Framework, a 45-day growth cycle was simulated. Water and produce testing occurred as described in Table 3. Two contamination level scenarios (high and low level) were simulated. Two contamination frequencies (daily or once per season) were simulated. This resulted in a total of 48 combinations.

Highlights

Assay performance was modeled for detecting contamination during cilantro production.
Frequent water testing most powerfully detects water contamination.
End of season produce testing most powerfully detects in-field contamination.

Abstract

As of August 2023, the two U.S. Food and Drug Administration (FDA) official detection methods for C. cayetanensis are outlined in the FDA Bacteriological Analytical Manual (BAM) Chapters 19b (produce testing) and 19c (agricultural water testing). These newly developed detection methods have been shown to not always detect contamination when present at low levels. Yet, industry and regulators may choose to use these methods as part of their monitoring and verification activities while detection methods continue to be improved. This study uses simulation to better understand the performance of these methods for various produce and water sampling plans. To do so, we used published FDA test validation data to fit a logistic regression model that predicts the methods’ detection rate given the number of oocysts present in a 10-L agricultural water or 25 g produce sample. By doing so, we were able to determine contamination thresholds at which different numbers of samples (n = 1, 2, 4, 8, 16, and 32) would be adequate for detecting contamination. Furthermore, to evaluate sampling plans in use cases, a simulation was developed to represent C. cayetanensis contamination in agricultural water and on cilantro throughout a 45-day growth cycle. The model included uncertainty around the contamination sources, including scenarios of unintentionally contaminated irrigation water or in-field contamination. The results demonstrate that in cases where irrigation water was the contamination source, frequent water testing proved to be more powerful than produce testing. In scenarios where contamination occurred in-field, conducting frequent produce testing or testing produce toward the end of the season more reliably detected contamination. This study models the power of C. cayetanensis detection methods to understand the sampling plan performance and how these methods can be better used to monitor this emerging food safety hazard.

DOI

https://doi.org/10.1016/j.jfp.2023.100161

Multi-Spectral Sorting Based on Visibly High-Risk Kernels Sourced from Another Country Reduces Fumonisin and Toxigenic Fusarium on Maize Kernels

Highlights

Fumonisin levels in Kenyan maize were significantly reduced after spectral sorting calibrated to high-risk kernels from Ghana
Spectral sorting reduced toxigenic Fusarium populations in maize kernels
Sorting based on visibly high-risk kernels represents a potentially generalizable approach to physical remediation of fumonisin

Abstract

Fusarium species infect maize crops leading to production of fumonisin by their toxigenic members. Elimination of microbes is critical in mitigating further post-harvest spoilage and toxin accumulation. The current study investigates the efficacy of a previously described multi-spectral sorting technique to analyze the reduction of fumonisin and toxigenic Fusarium species found contaminating maize kernels in Kenya. Maize samples (n = 99) were collected from six mycotoxin hotspot counties in Kenya (Embu, Meru, Tharaka Nithi, Machakos, Makueni and Kitui County) and analyzed for aflatoxin and fumonisin using commercial ELISA kits. Aflatoxin levels in majority (91%) of the samples were below the 10 ng/g threshold set by the Kenya Bureau of Standards and therefore not studied further. The 23/99 samples that had > 2,000 ng/g of fumonisin were selected for sorting. The sorter was calibrated using kernels sourced from Ghana to reject visibly high-risk kernels for fumonisin contamination using reflectance at nine distinct wavelengths (470-1,550 nm). Accepted and rejected streams were tested for fumonisin using ELISA, and the presence of toxigenic Fusarium using qPCR. After sorting, there was a significant (p<0.001) reduction of fumonisin, by an average of 1.8 log ng/g (98%) and ranging between 0.14 to 2.7 log ng/g reduction (28-99.8%) with a median mass rejection rate of 1.9% (ranged 0% to 48%). The fumonisin rejection rate ranged between 0 to 99.8% with a median of 77%. There was also a significant reduction (p=0.005) in the proportion of DNA represented by toxigenic Fusarium, from a mean of 30 to 1.4%. This study demonstrates the use of multi-spectral sorting as a potential post-harvest intervention tool for the reduction of Fusarium species and pre-formed fumonisin. The spectral sorting approach of this study suggests that classification algorithms based on high-risk visual features associated with mycotoxin can be applied across different sources of maize to reduce fumonisin.

DOI

https://doi.org/10.1016/j.jfp.2023.100142

Simulation Evaluation of Power of Sampling Plans to Detect Cronobacter in Powdered Infant Formula Production

Highlights

Sampling by Codex guidelines detects Cronobacter in a recalled PIF batch profile.
Sampling would not reliably detect Cronobacter in a non-recalled batch profile.
Sampling PIF with stratification is potentially more powerful than random sampling.
Taking more samples, even if smaller, increases the power to detect contamination.

Abstract

Cronobacter is a hazard in Powdered Infant Formula (PIF) products that is hard to detect due to localized and low-level contamination. We adapted a previously published sampling simulation to PIF sampling and benchmarked industry-relevant sampling plans across different numbers of grabs, total sample mass, and sampling patterns. We evaluated performance to detect published Cronobacter contamination profiles for a recalled PIF batch [42% prevalence, −1.8 ± 0.7 log(CFU/g)] and a reference, nonrecalled, PIF batch [1% prevalence, −2.4 ± 0.8 log(CFU/g)]. Simulating a range of numbers of grabs [n = 1–22,000 (representing testing every finished package)] with 300 g total composite mass showed that taking 30 or more grabs detected contamination reliably (<1% median probability to accept the recalled batch). Benchmarking representative sampling plans ([n = 30, mass grab = 10g], [n = 30, m = 25g], [n = 60, m = 25g], [n = 180, m = 25g]) showed that all plans would reject the recalled batch (<1% median probability to accept) but would rarely reject the reference batch (>50% median probability of acceptance, all plans). Overall, (i) systematic or stratified random sampling patterns are equal to or more powerful than random sampling of the same sample size and total sampled mass, and, (ii) taking more samples, even if smaller, can increase the power to detect contamination.

DOI

https://doi.org/10.1016/j.jfp.2023.100115

Spectral kernel sorting based on high-risk visual features associated with mycotoxin contamination reduces aflatoxin and fumonisin contamination in maize from Ghana

Comparison between the previous study (left) that first used this single-kernel optical sorter for mycotoxin remediation (Stasiewicz et al., 2017), and this study (right) which adapted that process to sorting based on rejection of kernels with high-risk features associated with mycotoxin contamination.

Highlights

Visual features associated with mycotoxin risk were used to train a multispectral sorter.
Multispectral sorting reduced aflatoxin and fumonisin levels in maize form Ghana.
Sorting removed about half the total aflatoxin mass, most of the fumonisin mass, and less than one quarter of the maize.

Abstract

Rapid single kernel analysis could enable physical sorting to remove mycotoxins from bulk grains. The purpose of this study was to use visual characteristics previously associated with aflatoxin and fumonisin contamination of maize kernels to calibrate a multi-spectral sorter and then sort mycotoxin contaminated lots. A total of 76 corn samples were collected from poultry farmers in the Dorma-Ahenkro area, Ghana. Paired 400-g subsamples were used for bulk analysis and single kernel sorting. Individual kernels were selected from contaminated samples by 2 levels of stratification: visible high-risk kernels (n = 1000) and visible low risk kernels (n = 1000). High-risk kernels had one or more of 3 features: fluorescence under UV (366 nm) light, mold, or brokenness. Kernels were used to calibrate a multi-spectral sorter (individual wavelengths from 470 to 1070 nm) to remove high-risk kernels. Then, kernel samples were sorted and the reject and accept streams individually ground and tested for aflatoxin and fumonisin contamination using ELISA. Bulk sample levels ranged between 0.78 and 67 ppb aflatoxin and <2.5 × 10⁻³ – 5.7 ppm fumonisin. Classification algorithms to reject visible high-risk spectra were 63% sensitive and 90% specific. After sorting, samples showed a significant aflatoxin reduction (p < 0.001, 73/76 samples reduced, mean reduction 31 ppb, range −9.7 – 67 ppb); all samples showed a significant fumonisin reduction (p < 0.001, mean reduction 1.9 ppm, range 9.3 × 10⁻² – 6.1 ppm). From the accepted stream, 61/76 samples tested <15 ppb aflatoxin, significantly more than the 40/76 prior to sorting (p < 0.001); all accepted samples tested <2 ppm fumonisin concentration compared to only 2/76 prior to sorting. From sorting, average mass rejected was 12% (range 1.2%–36%), and that rejected mass contained an average of 46% of the total aflatoxin (range 4.3%–97%) and 88% of the total fumonisin (range 10%–84%). Visual characteristics associated with mycotoxin contamination can inform classification models which can enable sorting contaminated maize to reduce aflatoxin and fumonisin contamination.

DOI

https://doi.org/10.1016/j.foodcont.2023.109788