Improving Benchmark Dose Estimation for Sodium Reduction

A recent series of experiments aimed to determine the amount salt can be reduced in food products without losing too much of their original taste; this is called the benchmark dose. In this study, a logistic model was used, however, it did not to include some features of the experiment and also used the data in an inefficient way. To improve the model, we borrowed the idea of risk assessment from toxicology and employed a log-logistic model with independent background risk. We also explored a more efficient way to analyze the experiment data by using a beta-binomial model. Compared to the beta-binomial model, the conventional logistic model overestimates the benchmark dose. The beta-binomial model is recommended for future use. I. Background and Motivation To respond to industry interest in salt reduction, Nuala Bobowski from the Department of Food Science and Nutrition at the University of Minnesota designed a series of experiments to determine how much sodium can be reduced with minimal adverse effects (Bobowski, 2012). An ‘adverse effect’ occurs when the difference between a normal food product and an identical product with same ingredients but reduced sodium is noticed by tasters half the time. Our goal is to find how much reduction in salt is possible without losing the traditional flavor of the food. This