Analyzing Big EHR Data—Optimal Cox Regression Subsampling Procedure with Rare Events

Massive sized survival datasets are becoming increasingly prevalent with the development of healthcare industry. Such pose computational challenges unprecedented in traditional analysis use-cases. A popular way for coping massive is downsampling them to a more manageable size, such that resources can be afforded by researcher. Cox proportional hazards regression has remained one most statistical models data to-date. This work addresses settings right censored and possibly left truncated rare events, observed failure times constitute only small portion overall sample. We propose subsampling-based estimators approximate their full-data partial-likelihood-based counterparts, assigning optimal sampling probabilities observations, including all failures analysis. Asymptotic properties proposed established under suitable regularity conditions, simulation studies carried out evaluate finite sample performance estimators. further apply our procedure on UK-biobank colorectal cancer genetic environmental risk factors.