[Population attributable fraction: estimation and interpretation]

The goal of this paper is to revise the concept, estimation methods, and interpretations of the population attributable fraction. From the usual formula of the population attributable fraction: (Ip ­ I₀) / Ip, where Ip is the cumulative incidence in the overall population, and I₀ is the cumulative incidence in the non-exposed group, other formulae are presented for use in exposures with more than two levels, and in the presence of confounding factors. Differences in estimation methods between cohort and case-control studies are discussed, and equations to estimate confidence intervals are displayed. Finally, some interpretations -­including the concepts of «etiologic case» and «case in excess», suggested by Greenland and Robbins-­, and some habitual errors are discussed.