Effect Modification and Its Impact on Preventable and Attributable Fractions in the Potential Outcomes Framework

Policy decisions should be guided by measures that capture the impact of exposures on outcomes and that explicitly account for present-day exposure distribution. Both the preventable and attributable fractions have been used for this purpose; however, exposure effects can vary across subpopulations, and when this occurs, appropriate interpretation of these measures should be facilitated by a discussion of the contributions of different subpopulations.

We analyze preventable and attributable fractions in the presence of effect modification. In particular, we use potential outcomes to formally define these quantities and to clarify the weighting of different strata in the total population measures.

Our derivations show that stratum-specific preventable and attributable fractions are weighted in proportion to the relative frequencies of effect modifiers among individuals with the outcome of interest. We also demonstrate that these weights are valid for the related quantities, preventable and attributable proportions. Finally, we present an example that illustrates how effect modification affects interpretation of these measures.

In sum, when effect modification is present, investigators should consider reporting these measures by the relevant population strata, and information that would allow quantification of their implicit weights in the total population estimate. Our study provides a formal justification for this approach.