Statistical models for recurrent events during alternating restraint and non-restraint periods
In longitudinal follow-up studies, during the observation period one subject or unit may experience multiple events, such as tumor recurrences, admissions to hospital, episodes of epileptic seizures or asthma, recidivism, equipment repairs and insurance claims. The multiple events can be either recurrent events of the same type or events of different types. In the last decades, such data have been extensively studied and a variety of statistical methods have been developed. However, statistical research on recurrent events of the same type during different types of time periods is sparse. The unique data structure requires specific considerations on the different types of time periods in statistical modeling. This research was motivated by a study on juveniles' recidivism, where the probationers were followed during alternating placement periods and free-time periods. These two types of periods may have different influence on the re-offenses and should be treated differently. The recurrent events that occur during alternating restraint and non-restraint periods also arise in many biomedical scenarios, such as tumor metastases during chemotherapy and chemo-free periods. In the first paper, we proposed a joint frailty model that accounts for the different types of time periods, as well as the dependence between the intensities of the recurrent events during different types of time periods. In the second paper, we jointly modeled two types of such recurrent events and developed a dynamic prediction tool to predict the risk of the next event of a severe type within a future time window using parameter estimates from the joint model and the historical information of these two types of events. Through simulation studies, we showed that the proposed methods outperformed the existing methods with smaller bias and better statistical efficiency. In the third paper, we applied the proposed methods to the juvenile probationers' re-offense data to evaluate the effectiveness of a community-based intervention program in reducing the rate of re-offending.
Li, Xiaoqi, "Statistical models for recurrent events during alternating restraint and non-restraint periods" (2015). Texas Medical Center Dissertations (via ProQuest). AAI10027792.