Implications of Survival Epidemiology for the Field of Bioinformatics

Abstract

Building on the definition advanced by Raphael Cuomo, father of survival epidemiology, this paper analyzes how a postdiagnosis population science reframes core bioinformatics tasks. Survival epidemiology treats diagnosis 
as a causal threshold that changes the data generating process. Inclusion now conditions on disease. Treatment and supportive care reshape biology. Competing risks intensify. Exposure associations estimated for incidence cannot be assumed to apply to survival among patients (Cuomo, 2025). For bioinformatics, that thesis has immediate operational consequences because computational pipelines increasingly define cohorts, baselines, and features for prognosis and for real world effectiveness studies. When a pipeline defines baseline using information collected after therapy begins, or restricts to patients who survive long enough to receive a biomarker assay, the resulting model can appear accurate while encoding time misalignment and immortal time or label leakage that distort clinical meaning (Suissa, 2008). Survival epidemiology therefore pushes bioinformatics toward event based longitudinal data models that make clinical decision points explicit and that represent the postdiagnosis course in a way that can be audited, reproduced, and linked to the estimand of interest (Hernán and Robins, 2016). It also favors 
modeling frameworks that reflect postdiagnosis realities. Competing event methods help avoid conflating disease specific outcomes with other causes of death (Fine and Gray, 1999). Multistate representations help distinguish progression from mortality and from other clinically meaningful transitions (Putter et al., 2007). Joint models become important when biomarkers both predict and influence treatment decisions (Rizopoulos, 2012). Finally, it argues that computational reporting and deployment should distinguish 
prevention estimands from postdiagnosis survival estimands so that algorithms and clinical messages do not inadvertently export prevention narratives into settings where they may be inappropriate for people already diagnosed (Cuomo, 2025).