Estimating Significant Differences in Serotype Prevalence from Salmonella Survival Studies Using a Four-serotype Cocktail in Low-aw Whey Protein Powder at Different Temperatures and aw

Introduction: The use of serotype cocktails is standard for inactivation studies to collect data on the worst-case scenario. Statistically significant differences in serotype survival can be obtained through rapid screening and statistical analyses. Regression models associate a response with a group of fixed effects. One can also use a multinomial mixed effects logistic model including random and fixed treatment effects. A non-parametric alternative is a chi square test comparing observed and expected values under the assumption that study conditions are not related to serotype prevalence.

Purpose: The aim of this study was to evaluate the suitability of three statistical approaches in estimating significant differences in the prevalence of four Salmonella serotypes in low-a_w whey protein powder.

Methods: Whey protein powder was equilibrated to a_w 0.18 and 0.54, inoculated with a cocktail of S. Agona, S. Tennessee, S. Montevideo and S. Typhimurium and stored at 36°C for six months and 70°C for 48 hours. PCR Multiplex serotyping was used to identify 306 randomly picked isolates. A multiple linear regression model, a multinomial mixed logistic model and a chi square test were used to identify significant differences in serotype relative prevalence.

Results: Regression models and the chi square test were not appropriate, as data were not normally distributed and observations were not independent. The most suitable model was the multinomial mixed logistic model accounting for variability among replications and the nature of the data. S. Tennessee showed to survive better than S. Montevideo and S. Typhimurium at higher temperatures and higher a_w levels (P < 0.05). The relative prevalence of S. Agona to S. Tennessee did not change significantly with increasing temperature (P = 0.211) or a_w (P = 0.453).

Significance: The statistical approach presented in this study may be applied to future survival studies that employ serotype cocktails.