Latent Class Model of Recurrent Events

• Activity Number: 59
• Type: Topic Contributed
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Yue Shentu*+ and Yabing Mai and Jiajun Liu and Yue Liu
• Companies: Merck and Merck Research Laboratories and Merck Research Laboratories and University of Virginia
• Keywords: latent class ; recurrent events ; gap-time

Abstract

A class of recurrent events data, such as the incidence of hypoglycemia in a clinical trial of Type II diabetes treatments, can be naturally modeled by renewal process of gap times. A feature often seen in such data is the alternating pattern of frequent and infrequent occurrence of events within the same subject, as well as notable heterogeneity in event recurrence between subjects. Previous methodologies often fail to account for both sources of variability. We propose a new latent class gap time model for this problem. The frequent and infrequent recurrence of events are modeled by two latent classes of correlated gap time distributions, and the random switching between the two latent classes in each subject are dictated by the probability of having an elevated hazard of event recurrence after the previous event. This latent class model provides an alternative framework of analyzing complicated recurrent event data, which in the hypoglycemia example enables us to differentiate two diabetes treatments in two aspects: the propensity of experiencing frequent hypoglycemia, and the hazard of the next hypoglycemia incidence.