Continuous Covariate Effects in Survival Analysis with Uncertain Outcomes and an Internal Validation Subsample

• Activity Number: 367
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jarcy Zee* and Sharon X. Xie
• Companies: Arbor Research Collaborative for Health and University of Pennsylvania
• Keywords: Survival analysis ; Kernel Smoothing ; Missing Data ; Validation Subsample

Abstract

Survival endpoints are often unavailable for all subjects in a study due to cost restrictions or invasive procedures, but alternative endpoints measure the true outcome with error. We develop a semiparametric estimated likelihood method for survival analysis of studies with uncertain outcomes available for all subjects and true outcomes available for only a subsample. Our proposed method assumes a proportional hazards model and discrete time points. Using kernel smoothing techniques, we estimate a hazard ratio of a continuous covariate. The maximum estimated likelihood estimator is consistent and asymptotically normal and we develop the analytical form of the estimator's variance. Through numerical studies, we show that the proposed method has little bias compared to the naïve estimate, which uses only uncertain endpoints, and is more efficient with moderate missingness compared to the complete estimator, which uses only true endpoints. Finally, we illustrate our proposed method using data from the Alzheimer's Disease Neuroimaging Initiative to estimate the effect of education on time to development of Alzheimer's disease.