Conceptual aspects of attributable risk with recurrent disease events.

The concept of attributable risk is extended to cover disease events which are potentially recurrent. Whereas the conventional attributable risk parameter allows quantification of the proportion of diseased individuals due to a certain risk factor, the parameter introduced here quantifies the proportion of disease events which is attributable to a specific exposure variable. We accomplish this by taking into account not only the probability of ever being affected by the disease but also the exposure-specific number of occurrences of the disease event within a defined time period. Practical examples highlighting the relevance of the new measure are presented. Various representations of the parameter are given and methods are proposed to adjust this new measure for confounding and effect modification. Point estimates of the parameter and their asymptotic variances are derived under different distributional assumptions. Advantages and limitations of this new approach for assessing the public health impact of an exposure for recurrent disease event are discussed.