Categorical versus continuous risk factors and the calculation of potential impact fractions

Abstract

Background: The potential impact fraction is a measure of effect that calculates the proportional change in disease risk after a change in the exposure of a related risk factor. Potential impact fractions are increasingly used to calculate attributable fractions when the lowest exposure is non-zero. Methods: Risk-factor exposure can be expressed as a categorical or a continuous variable. For a categorical risk factor, a change in risk-factor exposure can be expressed as a change in the proportion of the population in each category (‘proportions shift’). For a continuous risk factor, the change is expressed as a change in its parameters (‘distribution shift’). A third method (‘RR shift’) takes elements of both the categorical and the continuous approach. We compare the three calculation methods using hypothetical data on BMI and an intervention that affects the obese category. Results: The ‘proportion shift’ calculation produces non-linear artefacts and is best avoided. The ‘RR shift’ and ‘distribution shift’ calculation require the estimation of an RR function to describe excess risk, but perform much better. Conclusion: The ‘proportion shift’ calculation is best avoided. The ‘RR shift’ and ‘distribution shift’ calculation produce virtually the same results. For evaluating high-risk strategies, the ‘RR shift’ calculation is the simplest and therefore preferred. The ‘distribution shift’ is best suited for evaluating population strategies.