COMBINING INCERTITUDE, CAUSATION, AND GROUP JUDGMENT IN SCIENCE-POLICY

Background: Predicting disease burdens from low environmental or occupational exposures is essential correctly to inform stakeholders and decision-makers about how much, and at what cost, reducing exposure decreases the burden of multifactorial diseases such as cancer. Public health policy combines scientific knowledge, through expert group agreement, to resolve conflicts and enhance the legitimacy of its choices. Hence, scientific group conclusions should be formal, to be consistent with the tenets of scientific reasoning. However, particularly at low exposures, heterogeneous and conflicting evidence affect both causes and effects. Correct causal models link events that are site of action dependent (e.g., genes, proteins, cells, tissues, and organs, to results in whole individuals (clinical) and groups of individuals (epidemiological)), through sets of known, partially understood, and assumed aggregated physical, chemical, and biological mechanisms. Although only one causal model is eventually used by a public agency, its choice should account for the combination of several risk factors and adequately and replicably balance conflicting theories and results. Method: We formally aggregate aspects of: i) probabilistic evidence (i.e., uncertainty), ii) alternative forms of uncertainty, (i.e., incertitude) correctly to represent the semantic meaning of vague modifiers (e.g., using fuzzy or possibilistic measures) beyond the crisp confines of probability measures, and iv) assess competing individual expert opinions about causal assumptions, models, theories, and results through formal group decision-making methods. We discuss three overlapping areas. The first regards the combination of heterogeneous numerical or lexical information through compensatory and non-compensatory averages, fuzzy integrals and Dempster-Shafer combination rules. The second propagates and fuses uncertain knowledge, exemplified by exposure to inorganic arsenic and cancer using Monte Carlo simulations, and probabilistic and fuzzy causal networks of disease sub-processes. These formulate causation at different granularities (e.g., genomic and epidemiological) and integrate them into a causal disease process. The third addresses expert group decision-making, as may be adopted by scientific advisory groups, to inform national policy. Results: We develop simple examples using several measures of incertitude and formal combination rules that we integrate in developing causation at low exposures. Voting and ranking methods can result in paradoxes and impossibilities. We use the discursive dilemma, which may go unnoticed when simple majority or consensus, to identify a preferred causal model. Conclusions: It is essential for society to be fully informed in a replicable way about the aggregate judgments that are generated by panels to inform public policy and we provide such guidance.