ABSTRACT
A fundamental tool in pharmacoepidemiology is understanding the determinants of time to the first adverse event. While there is often interest in the number of events during the period of measurement, time to the first event is the purest measure of the safety (or lack there of) of a drug. A major advantage of time to event analysis or survival analysis is that it provides a natural solution to the problem of censoring. In time to event models a subject can either experience the event during the observation period, be measured throughout the entire observation period but fail to experience the event, or have a more limited window of observation due to factors such as change or discontinuation of insurance coverage, loss to follow-up or death. Subjects who are terminated from study without experiencing the event of interest are termed censored. The ability to accommodate censoring in statistical analysis is fundamental in pharmacoepidemiology. Of course an entire book could be devoted to the topic of survival analysis, and indeed many books have been written on the subject (e.g., Kalbfleisch and Prentice 2011, Elandt-Johnson and Johnson 1999, Miller 2011, Hosmer et al. 2008, Fleming and Harrington 2011). In this text we discuss so called “discrete-time survival models” because they can easily accommodate many of the features of time to event data that are characteristic of drug safety studies.
