Programming challenges of sampling controls to cases from the dynamic risk sets in nested case control studies

Abstract

Pharmacoepidemiological studies based on the cohort design are simpler to analyse and their results easier to interpret. However, these may not reflect real-life drug use which is a major strength of such studies. The nested case–control design is often used instead to avoid the computational burden associated with time-dependent explanatory variables. Unlike the classical case–control design which is generally easy to programme, that of the nested case–control can pose a number of challenges. Subjects can be chosen as controls more than once and a subject who is chosen as a control can later become a case. Indeed controls are chosen from among those in the cohort who are at risk of the event at that time (i.e. we sample from the risk set defined by the case). We highlight the main programming challenges of the design as well as describe and demonstrate approaches for resolution and appropriate implementation.