PhysioRisk: an interpretable survival- modeling framework for separating demographic and physiological mortality risk
Year: 2026
Type: preprint
Abstract: Background Biomarker-based mortality risk models often show strong predictive performance but remain highly correlated with chronological age, limiting interpretation of whether risk reflects demographic mortality structure or physiological deviation. We developed and evaluated a survival-based framework that separates age- and sex-based demographic mortality risk from an age- and sex-residualized physiological risk component. Methods Using NHANES 1999–2016 mortality-linked data, we estimated DemoRisk as the age- and sex-based component of a Cox proportional hazards model with spline-transformed age. Clinical biomarkers were residualized with respect to age and sex, aggregated into physiological axes, and entered into a penalized Cox model to estimate PhysioRisk. TotalRisk was defined as the additive recombination of DemoRisk and PhysioRisk. Performance was evaluated in a temporal test cohort using concordance index, hazard ratios per standard deviation, Kaplan–Meier stratification, and correlation with chronological age. Results were benchmarked against PhenoAge no-CRP proxy (PhenoAge NCP) and PhenoAge NCP acceleration. Results In the temporal test cohort (n = 6,739; deaths = 416), DemoRisk showed strong discrimination but high age correlation. PhenoAge NCP achieved similar discrimination and remained strongly age-correlated. TotalRisk matched comparator discrimination while preserving separation of demographic and physiological components. PhysioRisk showed modest standalone discrimination but low age correlation and a stronger hazard gradient than PhenoAge NCP acceleration. Conclusions Mortality risk in NHANES can be separated into a dominant demographic component and a mortality-relevant physiological component derived from age- and sex-residualized biomarkers. PhysioRisk is not intended to replace established mortality-risk scores, but to provide an interpretable decomposition of demographic and physiological contributions to risk.
Source: Research Square
Author: Gregory S. Allen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7163550907