UC Irvine

The analysis of time-to-event outcomes, or survival analysis, is common in many areas of biomedical research. In a large cohort study where the event of interest may have lower rates and the predictor of interest is expensive to measure, the case-cohort design can be applied to save the valuable covariate while maintaining valid inference, by performing analysis on a random sample from the full cohort and the unsampled cases. In Alzheimer's disease (AD) biomarker research, where many current and future biomarkers can be discovered from cerebrospinal fluid samples, the case-cohort design can be very useful. We propose several methodological developments within univariate and multivariate case-cohort designs, aimed to assist in solving challenges in AD biomarker discovery. We address some of the case-cohort design's shortcomings in its potential lack of power when performing analysis on a subset of the full cohort data. First, when there is at most one event per subject, we demonstrate the potential lack of power when assessing differential biomarker effects across rare subpopulations. We propose a stratified case-cohort estimator for assessing the interaction of the covariate of interest and an effect modifier, which is shown to be more efficient for comparing subgroup differences while maintaining valid marginal inference. Turning to the analysis of recurrent events, which is motivated by the need to perform analysis on multiple disease progressions and multiple covariate measurements, we develop a sampling and analysis framework and propose three types of case-cohort sampling to obtain an analysis set for efficiency gains. Lastly, to assist in planning for a recurrent events case-cohort design, we develop a log rank test statistic and derive the required sample size for a desired level of power.