Regression splines in the time-dependent coefficient rates model for recurrent event data

Abstract

Many epidemiologic studies involve the occurrence of recurrent events and much attention has been given for the development of modeling techniques that take into account the dependence structure of multiple event data. This paper presents a time-dependent coefficient rates model that incorporates regression splines in its estimation procedure. Such methods would be appropriate in situations where the effect of an exposure or covariates changes over time in recurrent event data settings. The finite sample properties of the estimators are studied via simulation. Using data from a randomized community trial that was designed to evaluate the effect of vitamin A supplementation on recurrent diarrheal episodes in small children, we model the functional form of the treatment effect on the time to the occurrence of diarrhea. The results describe how this effect varies over time. In summary, we observed a major impact of the vitamin A supplementation on diarrhea after 2 months of the dosage, with the effect diminishing after the third dosage. The proposed method can be viewed as a flexible alternative to the marginal rates model with constant effect in situations where the effect of interest may vary over time.