Antimicrobial resistance is a major challenge to modern medicine and of grave concern to public health. To monitor resistance, organizations analyze “drug/bug” collections of clinical assay results with the goal of estimating the wild-type distribution (those strains susceptible to the drug) and its prevalence. The most common assay involves placing a bacterium and two-fold concentrations of a drug in wells of a 96-well plate and checking for growth. This choice of two-fold concentrations enforces interval-censored results.
Various methods have been proposed to estimate the wild-type distribution. They all consider the wild-type distribution to be on the left with contamination by resistant isolates on the right. While all account for the interval-censoring, none of the methods accounts for the inherent variability of the assay, which can have a profound effect on the estimated distribution. We propose a Bayesian semiparametric method to address this limitation as well as to quantify changes in the prevalence over time. The improved estimation of the wild-type distribution is demonstrated through a series of simulation studies.