Project Summary/Abstract Motivated by the repeated and mixed types of post stroke readmission events, this proposal addresses the development and application of novel statistical approaches for the analysis of multivariate recurrent event data. It also develops a predictive model which allows the incorporation of multiple short-term events data for the prediction of survival outcomes. Existing approaches in stroke application usually (1) failed to consider the recurrent nature of the readmission data, or (2) considered only type of event or a composite endpoint by combining preventable and unpreventable events for analysis. These simplified approaches result in biased assessment of true disease burden. Therefore, to provide accurate understanding, it is important to study different causes of readmission events simultaneously. The commonly used models are shared random effect models with an assumption of constant dependence between different types of recurrent events over time. However, this assumption is not satisfied and subsequently the existing models are not adequate to model the data with time-varying dependence. To address this challenge, in Aim 1, we plan to develop an innovative joint modeling approach for multivariate recurrent event data allowing for time-varying dependence over time. To improve the prediction accuracy of long-term survival outcomes, it would be desirable to incorporate short-term event outcomes along with biological markers for risk prediction. Existing methods can only handle a single short-term event, which may have poor prediction performance when multiple short-term events are available and associated with survival outcomes. In Aim 2, we plan to develop a novel predictive tool which quantifies the risk of long- term outcomes by incorporating multiple short-term outcomes into prediction framework. These proposed approaches will address a gap in statistical analysis in the context of multivariate recurrent event data and provide a better predictive tool for survival outcomes.