Hypothesis-Based Weight of Evidence: An Approach to Assessing Causation and its Application to Regulatory Toxicology

Other papers in this symposium focus on combining direct observations or measurements of a phenomenon of interest. Here, I consider the distinct problem of integrating diverse kinds of data to address the scientific case for toxicological causation in view of information that usually contains gaps and outright contradictions. Existing weight-of-evidence approaches have been criticized as either too formulaic or too vague, simply calling for professional judgment that is hard to trace to its scientific basis. I discuss an approach-hypothesis-based weight of evidence-that emphasizes articulation of the hypothesized generalizations, their basis, and span of applicability. Hypothesized common processes should be expected to act elsewhere as well-in different species or different tissues-and so outcomes that ought to be affected become part of the evidence evaluation. A compelling hypothesis is one that provides a common unified explanation for observed results. Any apparent exceptions and failures to account for some data must be plausibly explained. Ad hoc additions to the explanations introduced to "save" hypotheses from apparent contradiction weaken the degree to which available data test causal propositions. In the end, we need an "account" of all the results at hand, specifying what is ascribed to hypothesized common causal processes and what to special exceptions, chance, or other factors. Evidence is weighed by considering comparative plausibility of an account including the proposed causal effect versus an alternative that explains all of the results at hand otherwise.