Modeling Salmonella Inactivation in Low Moisture Foods: Using Parameter Estimation to Improve Model Performance

Validating Salmonella inactivation processes for low moisture foods is a critically important food safety requirement, due to Salmonella persistence in these systems. Application of microbial inactivation models for this purpose is complicated by critical interactions between product water content and activity, temperature, and process humidity. Several models have been proposed; however, very few can handle or have been tested under dynamic conditions. One previously published model accounted for product surface temperature and process dew point, to predict Salmonella inactivation on almonds, but did not incorporate dynamic water activity. The goal of this study was to apply improved parameter estimation techniques to reduce correlation and relative standard errors of the parameters (RSEP), and to propose a more robust model for this application. Model fitting was performed using nonlinear regression, and the root mean squared error (RMSE), RSEP, variance-covariance matrix (VCM), and scaled sensitivity coefficients (SSC) were used to evaluate model performance in terms of parameter quality and robustness. Results indicated a reasonable performance of the model (RMSE = 1.6 log), with RSEP below 7.5%. However, VCM and SSC indicated correlation among the parameters. Therefore, multivariate optimization was applied to minimize the correlation, with the sum of the RSEP used as the objective function. Two of the elements on the VCM were reduced from around -0.5 to < 0.1, and the RSEP of the associated parameter also reduced from ∼7.5% to < 3.5%. The remaining matrix elements did not change, which indicates an inherently larger correlation among those parameters (0.91). Post-fitting analysis of estimated parameters and optimization of reference values for inactivation models are useful to improve model performance and reliability. An attempt to reparametrize the correlated parameters, accounting for the effect of product water activity, is underway. This modification accounts for process conditions, product characteristics, and interactions with product surface temperature.