Association Analysis of Multivariate Competing Risks Data

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5). In this proposal, the investigator describes three projects on association analysis of multivariate competing risks data which arise frequently in genetic family studies, demography and other areas. Often one is interested in familial association of the onset time of a certain event, with the presence of competing events which may dependently censor the occurrence of the target event. The usual association methods for multivariate survival data assuming the censoring by competing events independent of the target event may produce biased results. In addition, the marginal distributions of the event of interest are not identifiable. Hence the proposed association analysis for multivariate competing risks data focuses on two important quantities in competing risks literature: cause-specific hazard (CSH) and cumulative incidence functions (CIFs). The investigator develops a series of association analyses of multivariate competing risks data which account for the dependence censoring by competing events appropriately. The first project is related to two equivalent association measures of multivariate competing risks data which are CSH ratios and estimated nonparametrically without smoothing. In the second project, the investigator expands the application of frailty models to association analysis of multivariate competing risks data through an improper random variable and expresses the bivariate CIF in terms of its marginals and an association parameter. To incorporate covariates, in the third project, the investigator develops parametric regression models to investigate covariate effects on marginal CIFs and the indirect effects of covariates on the association analysis. These association methods cover many existing approaches for bivariate data as special cases. The proposal concentrates on modeling familial association in a target event with the presence of competing events where the standard methods may produce biased results. For example, in a large dementia study, family clustering in dementia onset is of interest where the competing event death may preclude the occurrence of dementia. The application of this research to the dementia and other studies in health and medicine is expected to generate novel insights on the association in a target event among members of a cluster, which help individuals and practitioners perceive the familial risks more accurately. The enhanced understanding may lead to better prevention and intervention in the target population who are at elevated risks. The methods can be used in many other applications such as demographic studies of human mortality, extremes in financial assets and returns, genetic evaluation of sires for longevity of dairy cows and annuity valuation with dependent mortality in insurance.