A simulation model to quantify the efficacy of dry cleaning interventions on a contaminated milk powder line

Abstract

Outbreaks of Salmonella in low moisture foods have been caused by cross-contamination from the processing environment into product. We used Monte Carlo simulations to model the impact of hypothetical cross-contamination scenarios of Salmonella from production equipment into milk powder. Model outputs included the quantity and extent of the contaminated product from a production line. Outputs were used to compare the efficacy of cleaning interventions. Cross-contamination of potential dry cleaning surrogates was also modeled. Input parameters for the model included log reductions from wiping an inoculated surface with a dry towel and transfer coefficients from an inoculated surface to milk powder. After a 2-log CFU contamination breach (Salmonella introduced to an 8.4 cm² stainless-steel surface on the processing line before production), the number of consumer-sized milk powder units (300 g) contaminated with Salmonella was 72 [24, 96] (median [p5, p95] across 1,000 simulation iterations). The average concentration of Salmonella within contaminated units was -2.33-log CFU/g [-2.46, -1.86]. Wiping the contaminated surface with a dry towel before the production of milk powder reduced the number of contaminated units to 26 [12, 64]. Flushing the contaminated surface with 150 kg of milk powder prior to milk powder production reduced the number of contaminated units to 0 [0, 41]. Flushing with 300 kg of milk powder further reduced the number of contaminated milk powder units to 0 [0, 16]. Enterococcus faecium resulted in a similar number of contaminated units (74 [44, 93]) compared with Salmonella (72 [24, 96]) after a 2-log CFU contamination breach.

Importance: This work demonstrates the utility of modeling as a decision support tool to (i) estimate Salmonella cross-contamination into product under different scenarios, (ii) compare different cleaning interventions, and (iii) help inform the selection of a Salmonella surrogate for cleaning validation studies. Risk models can describe the tradeoffs associated with different dry cleaning strategies in low moisture food environments. For example, the model presented in this study can estimate the differences in product contamination as a consequence of flushing a processing line with increasing quantities of material. Additionally, outputs from this model can be used to evaluate the risk of cross-contamination from a contaminated dry cleaning tool. Finally, comparing outputs from a simulation model is an alternative method for comparing Salmonella surrogates used in dry cleaning validation. Simulation model outputs (i.e., prevalence and concentration of contaminated units) may be more broadly interpretable than comparing transfer coefficients alone, enhancing decision support.

Keywords: Monte Carlo simulation; computer modeling; dry cleaning; dry sanitation; flushing; food-borne pathogens; low moisture food; milk powder.

Fig 1

The different scenarios considered in the model. Scenario 1 (cross-contamination from a contaminated surface into milk powder), scenario 2 (cross-contamination from a contaminated surface into milk powder after a flushing intervention), scenario 3 (cross-contamination from a contaminated surface into milk powder after a dry towel wiping intervention), scenario 4 (cross-contamination from a contaminated towel to a stainless-steel surface, and then into milk powder), and scenario 5 (repeated towel wiping of a contaminated stainless-steel surface without production of milk powder).

Fig 2

Partial rank correlation coefficients for all model input parameters were correlated with the prevalence and concentration of contaminated units (scenario 3). The transfer coefficient was highly influential on concentration and prevalence.

Fig 3

The results from 1,000 simulated production runs (scenario 1: no cleaning) are plotted for 2-log CFU (A and B) and 6-log CFU (C and D) starting contamination levels. (A and C) Each data point represents the number of milk powder units (300 g) contaminated with Salmonella in a simulated production run. (B and D) Each data point represents the average concentration of Salmonella within the contaminated milk powder units of a simulated production run.

Fig 4

The median contamination concentration (CFU/g) in the n-th milk powder unit (i.e., from the 1st unit produced to the 500th) is graphed in black with 5th and 95th percentiles graphed in red. The concentration of contamination within product units decreased exponentially as sequential units were produced in a simulated production run (scenario 1).

Fig 5

The results of 1,000 simulated production runs (scenario 2) with increasing amounts of material used in flushing before production begins (30, 150, and 300 kg). Increasing the amount of flushed material resulted in fewer product units contaminated with Salmonella.

Fig 6

The results of 1,000 simulated production runs for scenario 1, 2, and 3 runs are plotted for each starting contamination level (2-, 4-, and 6-log CFU). (A) Each data point represents the number of milk powder units (300 g) contaminated with Salmonella in a simulated production run. (B) Each data point represents the average concentration of Salmonella contamination within the contaminated milk powder units of a simulated production run.

Fig 7

(A) Each data point (n = 1,000 iterations) represents the number of successive cleanings with a dry towel required to remove all surface contamination (4 log) as modeled in scenario 4. The number of cleanings required was the most similar between L. innocua and S. enterica. (B) Across 1,000 simulated production runs (scenario 1, 2-log CFU), the median contamination concentration (CFU/g) in the n-th sequential milk powder unit produced during a production run is graphed for each surrogate. S. enterica and E. faecium had the most similar diffusion of contamination into product, with less contamination being transferred into each unit, but more units becoming contaminated.

Fig 8

Density estimate of the prevalence of contaminated milk powder units (A) and the mean concentration within contaminated units (B) across all simulated production runs (n = 1,000 iterations) for each organism after a 2-log CFU contamination breach. Higher peaks mean more simulations fell into the range measured on the x-axis. In terms of contaminated units, Salmonella and E. faecium were the most similar in the absence of a towel-wiping intervention, but L. innocua was more similar to Salmonella when towel wiping was included.