Joint Model with Random Changepoint for Longitudinal Measures and Semi-competing Risks

Dementia displays a gradual decline in cognitive abilities, often accompanied by an accelerated cognitive decline preceding diagnosis. Changepoint models are proposed to identify when cognitive decline accelerates and how it progresses. Joint models are developed to further account for dropout due to death or dementia. Cognitive decline in dementia patients may lead to complications that have an impact on their mortality. However, few joint models consider semi-competing risks (i.e., dementia and death) by distinguishing transitions between various health states, i.e., dementia without death, death after dementia, and death without dementia. We proposed a joint model that accounts for both changepoints and semi-competing risks by combining a multivariate random changepoint model for cognitive decline with an illness-death model that estimates health state transitions. We examined the proposed model with two types of random changepoints: one with a smooth change and another with an abrupt change. We also explored a shared random effect structure and a current value structure that connect both longitudinal and survival processes. Two types of cohorts, i.e., a disease cohort and a community cohort, were generated to evaluate the models. Simulation studies showed our proposed models could effectively characterize the influence of the longitudinal process on health state transitions. In addition, the choice of changepoint formulations, association structures, and cohort types impacted model performance. Real data application in the Framingham Heart Study indicated significant associations between changepoints in cognitive trajectories and health states for dementia and death. Our method provides a flexible framework to integrate longitudinal trajectories with changepoints and semi-competing risks.