Different methods for determining the thermal inactivation kinetics of microorganisms can result in discrepancies in thermal resistance values. In this study, thermal resistance of Salmonella in whole milk powder was determined with three methods: thermal death time (TDT) disk in water bath, pouches in water bath, and the TDT Sandwich system. Samples from three production lots of whole milk powder were inoculated with a five-strain Salmonella cocktail and equilibrated to a water activity of 0.20. The samples were then subjected to three isothermal treatments at 75, 80, or 85°C. Samples were removed at six time points and cultures were enumerated for survivors. The inactivation data were fitted to two consolidated models: two primary models (log linear and Weibull) and one secondary model (Bigelow). Normality testing indicated that all the model parameters were normally distributed. None of the model parameters for both consolidated models were significantly different (α = 0.05). The amount of inactivation during the come-up time was also not significantly different among the methods (α = 0.05). However, the TDT Sandwich resulted in less inactivation during the come-up time and overall less variation in model parameters. The survivor data from all three methods were combined and fitted to both consolidated models. The Weibull had a lower root mean square error and a better fit, according to the corrected Akaike's information criterion. The three thermal treatment methods produced results that were not significantly different; thus, the methods are interchangeable, at least for Salmonella in whole milk powder. Comparisons with more methods, other microorganisms, and larger varieties of food products using the same framework presented in this study could provide guidance for standardizing thermal inactivation kinetics studies for microorganisms in foods.
Three thermal treatment methods were compared for evaluation of Salmonella in milk powder.
No significant differences between the methods were observed for model parameters.
The TDT Sandwich produced model parameters with less variation.
The Weibull model fitted the inactivation data better than did the log-linear model.