A two-stage pseudo likelihood approach to estimation and inference for alternating recurrent events data

Hospitalizations in the United States cost almost 2 trillion dollars every year and represents one-third of the annual total healthcare cost in the nation. It has been nationally recognized that a large percentage of hospital readmissions can be avoided or prevented and consequently the national financial burden attributed to healthcare reduced. At the University of Iowa Hospitals and Clinics (UIHC), a nurse-led Transitional Care Team (TCT) intervention is deployed to attempt to prevent unnecessary hospital readmissions. The UIHC TCT intervention is designed in a way to provide patients with disease self-management, medical education, and clear instructions regarding discharge and hospital revisits in order to optimize cost.

This thesis work explores the UIHC TCT intervention effect by using an innovative statistical approach seen in the subject-area of recurrent events analysis. The dataset used to explore this intervention effect has been case-control-matched using another statistical technique called propensity score matching. The Center for Medicare and Medicaid Services (CMS) requires that 30-day and 90-day readmission rates be used as first-hand evidence of intervention effect. However, these measures have some serious limitations as they ignore the effects of multiple readmissions, time between readmissions, and potential correlation between readmissions within the same subject. We built a modeling framework that accounts for these factors.

The study population is inpatient care patients that were treated at the hospital for some time before returning to their homes. The time spent at the hospital receiving care may influence their subsequent return visits. This makes length of stay at hospital an important factor to overlook in determining future readmissions. We extended our model to include the estimation of intervention effect for both in-hospital length of stay and readmissions simultaneously. We call this analysis “alternating recurrent events analysis." We conduct simulation studies to examine the performance of our proposed model in the presence of multiple readmissions and within-subject correlation. We discovered a slight intervention effect which suggests that TCT should be encouraged as means to reduce cost and prevent unnecessary hospital readmissions.