Rethinking the Meaning of Concentration–Response Functions and the Estimated Burden of Adverse Health Effects Attributed to Exposure Concentrations

Four articles by Anenberg et al., Fann et al., Shin et al., and Smith contribute valuable perspectives and syntheses to a large and growing literature that estimates the burden of mortality risks attributed to fine particulate matter (PM2.5) based on estimated epidemiological associations, summarized as concentration–response (C–R) relations. This comment questions the use of C–R relations to predict or estimate how changing exposure concentrations would change responses in a population. C–R associations typically reflect modeling choices, and equally good choices can commonly lead to conflicting conclusions about the signs, significance, and magnitudes of C–R relations and regression coefficients. This indicates that C–R relations do not necessarily reflect underlying stable causal laws useful for making risk predictions, but only choices about how to describe past data, with no uniquely correct choice being determined by the data. Similarly, currently available C–R data typically do not suffice to make valid predictions about how future changes in concentrations will affect responses. These difficulties can be substantially overcome by model ensemble and causal graph modeling and time series methods, but these require different data and knowledge—for example, knowledge of how multiple variables depend on each other, rather than only of how one dependent variable is associated with multiple explanatory variables—than that captured by traditional C–R models or expressible by any single C–R coefficient or curve.