Guosheng Yin

Doctoral Student

Department of Biostatistics

University of North Carolina at Chapel Hill

The Mixed Baseline Additive Hazards Model for Multivariate Failure Time Data

Guosheng Yin and Dr. Jianwen Cai

Multivariate failure time data often arise in biomedical and epidemiological studies.  There might be multiple parallel events of interest, while there could be clustered individuals (e.g. siblings) or organs (e.g. teeth) contributing to each event type. Both the between-failure-type correlation and the within–cluster correlation need to be examined to ensure valid statistical estimation and inference.  For such right-censored multivariate survival data, we study the additive hazards model and propose estimating equations for parameter estimation.  The regression coefficient estimates are shown to follow multivariate normal distribution asymptotically where the sandwich-type variance-covariance matrix can be consistently estimated.  Furthermore, jointly across all the failure types, the estimated baseline and subject-specific cumulative hazard processes are shown to converge weakly to a zero-mean Gaussian random field.  The weak convergence properties for the corresponding survival processes are established as well. Through a resampling technique, we propose the procedures to construct simultaneous confidence bands for the survival curve of a given subject.  Based on the score process, we present graphical and numerical methods for the goodness-of-fit test under the additive hazards model. Simulation studies are conducted to assess the finite-sample properties and the new proposal is illustrated with the Framingham Heart Study data set.