Unlocking Retrospective Prevalent Information in EHRs - a Revisit to the Pairwise Pseudolikelihood

Electronic health records offer abundant data on various diseases and health conditions, enabling researchers to explore the relationship between disease onset age and underlying risk factors. Unlike mortality data, the event of interest is non-terminal, hence individuals can retrospectively report their disease-onset-age upon recruitment to the study. These individuals, diagnosed with the disease before entering the study, are termed “prevalent”. The ascertainment imposes a left truncation condition, also known as a “delayed entry”, because individuals had to survive a certain period before being eligible for enrollment. The standard method to accommodate delayed entry conditions on the entire history up to recruitment, hence the retrospective prevalent failure times are conditioned upon and cannot participate in estimating the disease-onset-age distribution. Other methods that condition on less information and allow the incorporation of the prevalent observations either bring about numerical and computational difficulties or require statistical assumptions that are violated by most biobanks. This work presents a novel estimator of the coefficients in a regression model for the age-at-onset, successfully utilizing the prevalent data. Asymptotic results are provided, and simulations are conducted to showcase the substantial efficiency gain. In particular, the method is highly useful in leveraging large-scale repositories for replication analysis of genetic variants. Indeed, analysis of urinary bladder cancer data reveals that the proposed approach yields about twice as many replicated discoveries compared to the popular approach.