European Journal of Public Health · Oct 2026
clinicalAFT: From Routine Biomarkers to Predicted Life Expectancy Changes
By Rebelo T, Yoo D, Patel C et al.
Abstract Patients often present with multiple abnormal biomarkers, yet clinicians lack intuitive tools to prioritize which to address first. Translating biomarker and physiological indicator data into expected years of life gained or lost could improve decision support and patient communication. We analyzed 25,217 U.S. adults from 30 years of the National Health and Nutrition Examination Survey (NHANES 1988-2018) linked to mortality outcomes through the National Death Index. Models included 19 physiological indicators, serum cotinine and demographic factors, with adjustment for age at study entry and pre-existing diagnosed conditions. Data were split into training and test sets. We developed a penalized Weibull accelerated failure time (AFT) model with squared terms to capture non-linear effects and compared it with a deepAFT model. Performance was assessed using C-index. We estimated changes in life expectancy across percentiles of each physiological indicator to quantify their marginal contribution to lifespan. The penalized Weibull AFT model achieved a C-index of 0.815, while deepAFT achieved 0.839. Variable importance rankings were broadly consistent across models, with age at study entry contributing most to lifespan prediction. The highest values of C-reactive protein, glycohaemoglobin and systolic blood pressure were associated with approximately 1.0, 1.75 and 1.5 fewer years of life, respectively. Hemoglobin showed an inverse U-shaped association: moderate values were associated with a 0.5-year gain from median lifespan, whereas very low values were associated with a 1-year reduction. Routinely collected biomarker and physiological indicator data can be translated into predicted years of life gained or lost, offering an intuitive alternative to hazard ratios. By ranking abnormal biomarkers by their estimated lifespan impact, this approach could support clinical prioritization, prevention and more intuitive communication. Key messages • Biomarkers and physiological indicator data can be translated into predicted years of life gained or lost. • Survival models can rank abnormal biomarkers to support clinical prioritization.