OECD Document 54 (“Current Approaches in the Statistical Analysis of Ecotoxicity Data: A Guidance to Application”), first released in 2006, is being revised and will also include Bayesian methods alongside traditional frequentist approaches. The OECD document 54 annexes provide example datasets from OECD test guidelines 201 (algal growth inhibition), 202 (Daphnia acute immobilisation), 211 (Daphnia reproduction), and 215 (fish juvenile growth). The revision reflects a broader shift in life sciences toward Bayesian statistics, valued for handling hierarchical, nonlinear, and data-limited scenarios, while offering transparent assumptions, improved uncertainty quantification, and reproducible workflows. Criticisms of Bayesian methods (higher data demand, greater uncertainty, and subjectivity due to priors) will be addressed through intuitive workflows, checklists, and comparisons with frequentist results. For concentration-response analysis, log-normal, log-logistic, and Weibull type I and II models were applied, with model choice guided by Bayesian cross-validation. ECX values and credible intervals were derived from posterior distributions. Hypothesis testing used Bayesian hierarchical regression (analogous to ANOVA), with partial pooling to reduce false positives/negatives and effect significance assessed via posterior draws and the ROPE principle, yielding de facto NOEC/LOEC values. Validation included prior-posterior and posterior-predictive checks, plus sensitivity analyses to detect prior-likelihood conflicts. Results showed Bayesian methods robustly produced EC/LCX and NOEC/LOEC values across datasets, with credible intervals automatically capturing uncertainty from data variability, sample size, and priors. While strong priors influenced some parameters, final endpoints remained unaffected. Overall, Bayesian analysis matched frequentist outcomes in most cases and offered advantages in challenging contexts (e.g., unwanted mortality, small sample sizes). The methodology meets regulatory requirements, demonstrating that Bayesian approaches are not more data-demanding, uncertain, or subjective than frequentist ones; and can even reduce uncertainty in some cases. The study encourages scientists and regulators to adopt Bayesian methods for standard ecotoxicity data analysis, highlighting their robustness, transparency, and potential to improve regulatory statistics.
Wolf et al. (Thu,) studied this question.