Applying Bayesian forecasting to predictive toxicology: The probability of innate carcinogenicity to humans of colorants synthesized from benzidine

Abstract

The preclinical identification of health hazards relies on the performance (the historic concordance to the respective gold standard) of regulatorily recommended bioassays. However, any testing with less than 100% sensitivity (or 100% specificity) can deliver false results (outcomes discordant to the respective gold standard). Conversely, the predictive values approach (a.k.a. Bayesian forecasting) weighs (1) the performance of the predictive bioassay (battery, or framework) with (2) the prevalence of -positivity to the respective gold standard- in the most representative category to which the test substance can be allocated. Thus, the predictive values approach (PVA) provides the numeric probability for the toxicity to humans of chemicals that, circumstantially, are evaluable only through nonclinical data. Consequently, the PVA improves the predictivity of nonclinical toxicology, and increases the impact of hazard identifications entirely based on preclinical data. This article aimed to introduce the PVA through a worked example. Due to their toxicological homogeneity and public health relevance, the superfamily of colorants synthesized from benzidine (BZ) or some mutagenic congeners was selected (colorings hereafter mentioned as BZ-related-colorants). Through the PVA, the numeric probability of innate carcinogenicity to humans of 259 BZ-related-colorants was either estimated from rodent carcinogenesis bioassays (RCBs) or predicted from methods alternative to the RCB. A discussion was provided on (1) some limitations and implications of the PVA, and (2) the probable significance of the predictive values figured here for 259 BZ-related-colorings.

Keywords: Bayesian forecasting; Benzidine-based dyes; Predictive toxicology; Predictive values approach; Regulatory toxicology.