B1354
Title: Methods for analyses of recurrent events in cohort studies
Authors: Neil Wright - University of Oxford (United Kingdom) [presenting]
Abstract: Many epidemiological and clinical studies include analyses of events that can reoccur, such as hospitalisations, but utilise only the first occurrence for each participant and disregard subsequent events. Large cohort studies often collect data from routine sources such as electronic health records and health insurance systems, which include details of multiple events for each participant during follow-up. Non-parametric approaches to the analysis of recurrent events include estimation of incidence rates and the mean cumulative count. Poisson or negative binomial regression models can be used to compare incidence rates. There are several extensions to the Cox proportional hazards model for comparison of the hazards of recurrent events. In some studies, there are two or more types of recurrent events, or a recurrent event and a competing terminal event, and joint modelling of hazards or multi-state models are required. These various approaches and available methods for their application will be described. The challenges of applying these methods and communicating results in the context of a large cohort study will also be explored.
