Keywords: recurrent event, hypothesis testing, causal inference
In this presentation, I will discuss how recurrent event data behaves in the context of causal inference with observational data. In particular, I focus on recurrent events within a certain time window and interventions to change the rates of recurrence. For instance, how many patrons revisit within a month before and after an intervention?
Real world examples of this type of recurrent event include hospital re-admissions within 30 days, product returns within a return window, and sales promotion by issuing coupons upon check-out (where the coupons are valid for a month after check-out). The intervention or campaign aims to either reduce or promote target events happening within a specified time window that follows the intervention. Note that in this context the intervention or campaign happens only if an index event occurs (e.g., a patient is hospitalized and discharged, or a customer makes a purchase).
First I will briefly introduce the framework of quasi-experiments with observational data to evaluate causal effects of an intervention. Next I will show how this type of recurrent event data behaves in hypothesis testing. For example, in comparing the recurrence rate of two time periods, the results can be misleading when the lengths of the time periods are different. Last I will suggest methodological solutions to analyze recurrent event data. The study results include Monte Carlo simulations as well as a real world application.