Shou-En Lu, Johns Hopkins University Department of Biostatistics
In this talk, two methodologies will be introduced: a marginal inference on clustered failure time data, and a case-control design and analysis for a cohort of individuals clustered into small groups. In the first methodology, we assume that the marginal hazard of failure for individuals is specified by a proportional hazards model and the structure of dependence between individuals within a cluster is left unspecified. An estimation procedure is developed based on a pseudo-likelihood. With this methodology, a risk set sampling method is proposed for the construction of the pseudo-likelihood. The principle is to construct an unbiased risk set consisting of independent individuals at each distinct failure time. In the second methodology, a case-control design for a cohort of individuals clustered into small groups will be introduced. With an extension of the risk set sampling method proposed in the first methodology, a small number of controls per case from the nonfailures are sampled in the full cohort at each distinct failure time. Instead of collecting covariate information from all cohort members, only those from the cases and the controls sampled at each distinct failure time are needed for the analysis. A data set from NNIPS (Nepal Nutrition Intervention Project-Sarlahi) will be given to illustrate the use of both methodologies, and simulation studies will be presented to demonstrate the performances of the proposed estimators.
Return to Longitudinal/Survival Working Group List | Return to Home Page