Systematic handling of environmental fate data for model development – illustrated for the case of biodegradation half-life data

Environmental hazard endpoints describing persistence, mobility, toxicity, or bioaccumulation of chemicals are often associated with high variability in experimental outcomes. The assessment of persistence in the environment is particularly affected due to a multitude of influencing environmental factors, including the taxonomic composition and physiological state of the microbial community present. Biotransformation experiments may therefore result in half-lives spanning several orders of magnitude for the same substance tested with different environmental samples. Due to experimental limitations, values may further be beyond the limits of reliable half-life quantification (i.e., censored data points), and the number of data points per substance may vary considerably. However, reliable data describing average half-lives and their natural variability are an important foundation for building predictive models for environmental hazard endpoints, which are urgently needed by regulatory authorities to manage existing chemicals and by industry for the design of benign, non-persistent chemicals. Here, we propose the application of Bayesian inference to characterize the uncertainty of reported half-lives and to include censored data points to maximize the information extracted from experimental data. Our model estimates the true mean and standard deviation from a set of reported half-lives experimentally obtained for a single substance. Including censored data increases the available data volume, and reporting uncertainties helps estimating the reliability of the half-life data. We apply the inference model to 893 substances with experimental soil half-lives of varying data quantity and quality, and we estimate the true half-life distribution for each compound. Our approach can be easily adapted and applied to other environmental hazard endpoints to estimate uncertainty and to improve data quality for model development.