A0197
Title: A new class of mark-specific proportional hazards models for recurrent events
Authors: Yi Li - University of Michigan (United States) [presenting]
Abstract: Recurrent events can have continuously varying `marks,' such as dosage levels in prescription refills. In a study on post-surgical opioid prescription refills, identifying patient factors associated with the time between refills is crucial for improving pain management and reducing opioid misuse. As associations may differ across dosage levels, existing methods struggle to capture these dynamics. A new class of mark-specific proportional hazards models is proposed for recurrent events to model hazards of refill at given dosages. A marginal model is developed on the gap time scale with a novel dual-weighting scheme to allow for events' information to be weighted depending on their proximity to the mark of interest while addressing the informative nature of each subject's number of recurrences. Consistency and asymptotic normality of the estimator are established, and a sandwich variance estimator for robust inference is developed. Simulations show that the method performs better than competing methods in finite samples. The method is used to study post-surgical opioid prescription refills, identifying factors associated with opioid prescription refill recurrence corresponding to dosages of interest.
