Statistical Analysis for Mixed Outcome Measures in Recurrent Event Studies

项目总结/文摘:

Project Summary/Abstract: This application addresses the development and application of new analytic strategies for mixed panel-binary/panel- ordinal/panel-count/recurrent-event data in event history studies. While these 4 data types are all generated from recurrent- event processes, they have different endpoints. Recurrent-event data record the occurring time points of each event, panel- count data record the number of events since the last observation, panel-ordinal data record the number of events categorically, and panel-binary data record if any event has happened since the last observation. At times we must deal with mixed data as different endpoints may be collected for the same variable in multiple observations. Among the 4 data types, recurrent-event data offer the greatest amount of relevant information, followed by panel-count, ordinal, and binary data. The statistical literature on mixed panel-binary/panel-ordinal/panel-count/recurrent-event data is sparse though examples of these data types exist abundantly in cancer and non-cancer studies. The renowned longitudinal Childhood Cancer Survivor Study (CCSS) is an example. Standard multivariate or longitudinal methods cannot reflect the special structure of the event process underlying the panel-binary or ordinal data. There is an urgent need to develop intuitive, efficient, and computationally feasible methods for analyzing complex data in event history studies. In this proposal, we plan to: 1) develop a likelihood-based semiparametric estimation method for regression analysis of mixed panel-binary and panel-count data; 2) develop a likelihood-based semiparametric estimation method for regression analysis of mixed panel-binary data, panel-count data, and recurrent-event data and apply that method to the CCSS data; 3) develop likelihood-based semiparametric estimation methods for regression analysis of mixed panel-binary/pane-ordinal/panel-count/recurrent-event data. These approaches will address a gap in statistical analysis in the context of recurrent events and potentially have strong statistical and clinical relevance for the study of complex event history data.