# Biological Age Development and Validation of an Accessible Tiered Biological Age Model for Population Health First Author: *Serene TOH, BSc.* — Global Lead Data Science and Modeling, Elfie Inc. Corresponding Author: *Jean-Francois LEGOURD, MSc.* — Chief Product Officer, Elfie Inc. Affiliations: Elfie Inc., Global Health Data Science Division ## **Abstract** ### **Background** Current biological age models often fail to convey risk in ways that are interpretable and actionable for the general population. A holistic measure that integrates lifestyle, physiological, and biomarker factors is needed to motivate prevention. Using NHANES linked mortality data, we developed and validated an interpretable "Biological Age" metric derived from behavioral, anthropometric, and biomarker data, scalable across settings with varying data availability. ### **Methods and Findings** We analyzed NHANES 2005–2009 with linkage to mortality through 2019 (Centers for Disease Control and Prevention [CDC], 2025). Adults aged ≥20 years with complete data on behavioral, physiological, and biochemical risk factors were included. Survey-weighted Cox proportional hazards models estimated associations with non-accidental mortality, accounting for the complex sampling design (Cox, 1972; Lumley, 2004; Therneau & Grambsch, 2000). Both the simplified and full models demonstrated strong discrimination and calibration when validated in NHANES 2010–2011. The **full model** included blood sugar (HbA1c), eGFR, blood pressure, smoking, sleep hours, total metabolic expenditure per week, heart disease history, sex and age (analytic sample: 7,775 adults; 1,106 deaths). ### **Conclusions** A Biological Age metric can be derived either from a biomarker-rich model or a minimal model using only readily available variables. External validation confirms generalizability. The two-tier framework balances precision and feasibility, supporting use in population health, patient counseling, or public health surveillance. ## **Introduction** Chronological age is a cornerstone metric in epidemiology and clinical practice, yet it fails to fully capture heterogeneity in biological risk. Two 60-year-olds may have vastly different trajectories of morbidity and mortality depending on lifestyles, comorbidities, and underlying biology. The concept of "biological age" aims to capture this divergence by aggregating multiple risk factors into a single metric that more closely reflects the individual’s physiological state (Klemera & Doubal, 2006; Levine, 2013; Liu et al., 2018). We propose a two-tier Biological Age framework: a **full model** that incorporates biomarker measures when available, and a **simplified model** that relies only on easily obtained variables. Our goal is to produce a robust, interpretable, and generalizable Biological Age measure suitable for broad use. ## **Methods** ### **Study Population & Data Sources** We analyzed data from the National Health and Nutrition Examination Survey, a nationally representative survey of the U.S. civilian, non-institutionalized population conducted by the National Center for Health Statistics (NCHS), Centers for Disease Control and Prevention (CDC, 2025). NHANES collected information through structured interviews, physical examinations, and laboratory testing performed in mobile examination centers. - **Modeling cohort:** NHANES 2005–2009 with mortality follow-up via linkage to the National Death Index through December 31, 2019. (CDC/NCHS) - **Validation cohort:** NHANES 2010–2011, harmonized to the same variable definitions and exclusions as NHANES 2005–2009, with mortality linkage. All NHANES protocols were approved by the NCHS Research Ethics Review Board, and written informed consent was obtained from all participants. ### **Statistical Analysis** Continuous variables were expressed as means (±SD), and categorical variables as weighted percentages. Survival time was defined as months from baseline examination to death or censoring on December 31, 2019. Non-accidental mortality was modeled using survey-weighted Cox proportional hazards regression, svycoxph in R. Candidate predictors included age, sex, smoking, total metabolic expenditure per week, average sleeping hours per day, systolic blood pressure (SBP), diastolic blood pressure (DBP), blood sugar (HbA1c), eGFR and self-reported history of diabetes and heart disease. ## **Results** ### **Simplified Model** In the model including age, sex, smoking, average sleeping hours per day, total metabolic expenditure per week and heart disease history, all predictors were significantly associated with mortality risk. - Age and smoking are the dominant predictors of mortality. - Male sex, history of heart disease and low sleep hours significantly elevate risk. ### **Full Analytic Model** The full model incorporated spline terms for blood sugar (HbA1c) and demonstrated significant nonlinearity. Age was the dominant predictor of mortality. Smoking, low sleeping hours and Male sex were strongly associated with mortality. ### **Validation Results** Both the simplified and full Biological Age models demonstrated excellent discrimination and satisfactory calibration in the independent NHANES 2010–2011 cohort, confirming robustness and generalizability. ## **Discussion** We derived and validated Biological Age, a transparent mortality risk score expressed in age-equivalent terms. The simple model is parsimonious and interpretable, while the complex model integrates biomarkers for slightly improved performance. ### **Methodological limitations** Residual confounding and measurement error are possible, particularly for self-reported exposures. Future work could apply multiple imputation. ### **Implications and Future Directions** Biological Age offers a powerful communication tool for personalized risk feedback. Further validation of Biological Age across contemporary cohorts will be important to establish its generalizability. Integration into digital health platforms may also enable real-time personalized risk feedback at scale.