In a dynamic cohort study evaluating predictive modeling strategies for primary public health services, researchers found that extended Bayesian joint modeling effectively incorporated longitudinal biological age trajectories to predict cardiovascular disease risk among older adults. By simultaneously estimating latent aging trajectories and cardiovascular disease outcomes, the Bayesian joint model achieved numerically superior overall predictive performance compared with two alternative dynamic modeling approaches.
While chronological age has traditionally served as a standard demographic metric, longitudinal biological age more effectively captures individualized physiological deterioration and dynamic aging patterns over time. However, how best to integrate repeated biological age measurements into dynamic cardiovascular prediction models within primary public health settings has remained an open methodological question.
To evaluate dynamic modeling approaches, the researchers leveraged data from the West China Health and Aging Cohort spanning from 2017 to 2024. The study analyzed 264,272 physical examination records collected from 33,034 residents who were followed for an average duration of 7.54 years. Biological age was constructed from seventeen core biomarkers using linear regression and machine learning algorithms, including a random forest framework.
The investigation systematically compared three distinct analytical frameworks for incorporating longitudinal biological age measurements: a two-stage pace-of-aging Cox model that extracted composite aging rates via mixed-effects models before standard Cox regression, a time-varying Cox model that dynamically updated biological age covariates, and an extended Bayesian joint model that simultaneously evaluated latent biological age trajectories alongside cardiovascular risk. For risk prediction, the five-year mark served as a common landmark, with event risks over the subsequent one-year and three-year horizons estimated using only measurements gathered by that landmark.
When quantifying associations with overall cardiovascular disease, random forest-derived biological age was significantly linked to event risk across all three modeling frameworks. The Bayesian joint model demonstrated a hazard ratio of 1.126 with a 95 percent confidence interval of 1.098 to 1.154 (P < 0.001). The time-varying Cox model produced a hazard ratio of 1.115 with a 95 percent confidence interval of 1.086 to 1.144 (P < 0.001), while the two-stage pace-of-aging Cox model yielded a hazard ratio of 1.067 with a 95 percent confidence interval of 1.012 to 1.124 (P = 0.015).
Evaluating discriminatory performance in a hold-out testing set using time-dependent area under the receiver operating characteristic curve, the Bayesian joint model achieved the highest one-year estimate for overall cardiovascular disease at 0.776, exceeding the time-varying Cox model at 0.736 and the two-stage model at 0.735. Across a three-year prediction window, the Bayesian joint model and time-varying Cox model exhibited similar discriminatory power, recording area under the curve values of 0.754 and 0.757, respectively, compared with 0.734 for the two-stage model. Furthermore, compared with a conventional time-varying Cox model utilizing traditional risk markers—which recorded area under the curve estimates of 0.651 at one year and 0.625 at three years—the biological age model demonstrated higher predictive value.
The authors concluded that extended Bayesian joint modeling successfully integrates longitudinal biological age dynamics with survival outcomes and performs well in dynamic cardiovascular disease prediction across subtypes. These findings warrant further evaluation of joint modeling strategies for dynamic cardiovascular risk stratification in older adults, with biomarker selection tailored to target disease processes.
Subject of Research: Dynamic modeling strategies incorporating longitudinal biological age trajectories into cardiovascular disease prediction
Article Title: How to incorporate longitudinal biological age trajectories into cardiovascular diseases prediction for primary public health service? A comparative study of three dynamic modeling strategies
Article References: Gong, P., Han, X., Chang, Z., Jiang, X., Liu, Y., Xia, M., Li, J., & Zhang, T. (2026). How to incorporate longitudinal biological age trajectories into cardiovascular diseases prediction for primary public health service? A comparative study of three dynamic modeling strategies. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03068-9
Image Credits: AI Generated
DOI: 10.1186/s12874-026-03068-9
Keywords: cardiovascular disease, biological age, longitudinal trajectories, Bayesian joint model, time-varying Cox model, two-stage PoA-Cox model, dynamic risk prediction, risk stratification, older adults, biomarkers, random forest, West China Health and Aging Cohort, primary public health, survival analysis, AUC discrimination, pace of aging
News Source: Ophelia Keating. (November 19, 2026). Bayesian Joint Models Improve Cardiovascular Risk Prediction Using Longitudinal Biological Age. Scienmag.



