Joint modeling of correlated longitudinal data and bivariate types of recurrent events in the presence of death with application to the long-term care insurance data

Abstract

In many longitudinal studies, we observe several correlated processes: a repeated measures process and a recurrent events process. Occasionally, bivariate types of recurrent events are observed. Furthermore, the follow-up of both processes may be stopped by an informative terminal event. For example, in the Long-term Care Insurance data, higher long-term care level scores for beneficiaries are well associated with higher risk of recurrent hospitalizations or out-patient services. These processes also show strong correlation with mortality.In this thesis, a joint random effects model for the repeated measures and bivariate types of recurrent events precess with the presence of death has been proposed. These relationships are modelled by conditioning of shared random effects. Maximum likelihood estimation and inference are carried out through a Gaussian quadrature technique, assuming piecewise constant baseline hazards for recurrent events and death. The proposed model can be fitted easily by Gaussian quadrature tool, e.g. procedure NLMIXED in SAS.We applied the proposed model to the LTC Insurance data to evaluate which LTC service (institutional care or home care) is better for functional status and survival. We also compared the proposed model with other various joint models. As a result, we can not find significant differences in the effectiveness between institutional care and home care from the proposed model. However, we note that home care is associated with better survival in other joint models. These suggest that ignoring the relationships between dependent processes may lead to biased results.In this research, the proposed model greatly facilitates the application of joint random effects models with SAS NLMIXED. So we expect that the joint model can be more easily applied to real data through our thesis. And we can extend the proposed model to multiple types

for terminal event.