Survival analysis with uncertain endpoints using an internal validation subsample

摘要: When a true survival endpoint cannot be assessed for some subjects, an alternative that measures the with error may collected, which often occurs when is too invasive or costly to obtain. We develop nonparametric and semiparametric estimated likelihood functions incorporate both uncertain endpoints available all participants only subset of participants. propose maximum estimators discrete function time hazard ratio representing effect binary continuous covariate assuming proportional hazards model. show proposed are consistent asymptotically normal analytical forms variance estimators. Through extensive simulations, we also have little bias compared naA ve estimator, uses endpoints, more efficient moderate missingness complete-case endpoints. illustrate method by estimating risk developing Alzheimer's disease using data from Disease Neuroimaging Initiative. Using our optimal study design strategies compare across treatment groups new trial these characteristics. demonstrate how calculate number events in validation set desired power simulated baseline distribution event, size, correlation between outcomes, proportion outcomes missing can pilot studies. sample size formula does not depend on event calculated matches well simulation based results. results Ginkgo Evaluation Memory study, would need observed studies comparing development among those without antihypertensive use, as total subjects recruited trials. Degree Type Dissertation Name Doctor Philosophy (PhD) Graduate Group Epidemiology & Biostatistics First Advisor Sharon X. Xie