This project addresses modeling and inference for joint longitudinal and recurrent event outcomes. Although there are a variety of models for analysis of joint longitudinal and (single event) survival outcomes, the recurrent event setting raises unique features that require nontrivial generalization. Among these features is the interplay that may occur between the longitudinal and recurrent event processes due to interventions at each event occurrence. The investigator develops and implements a flexible class of models and associated inferential procedures for this situation, building upon a latent class approach. The properties of these methods are characterized relative to alternative modeling approaches, such as generalizations of selection and shared random effects models often used in the joint longitudinal-survival context. Application to real data demonstrates practicality of the methods and that new insight can be obtained.
This project addresses the situation where a series of event times is observed for each subject under study, together with a series of readings on a quantity (marker) that provides information about the risk for subsequent events. Examples in the biomedical sciences include recurrent cancer and an associated biological marker that conveys information about the risk for subsequent cancers, and recurrent cardiovascular disease events and markers such as stress, cholesterol or blood pressure levels. In engineering and reliability, the event times may be machine breakdowns, with the marker being part fatigue. A context relevant to national defense is the occurrence of terrorist acts with chatter an associated marker. This project develops modeling and prediction methods for this context that allow for the effects, on both the marker and event processes, of interventions following event occurrences. In the example contexts given, such interventions could be medical treatments, machine repairs, and changes in defense postures, respectively. The methods combine information from both the marker and observed event occurrence times to strengthen predictions about each. The new methods are applied to real data to demonstrate their practicality and that new insight can be obtained. The project also contributes to research infrastructure by training graduate students.