Development of a dynamic risk prediction tool for identifying individuals at high risk of cardiovascular diseases: A post-hoc analysis of the multi-ethnic study of atherosclerosis.
Abstract
Background
AND
Aims
Cardiovascular disease (CVD) remain the leading cause of mortality and disability. Most risk prediction tools rely on risk factors measured at a single time point, ignoring changes in biomarkers over time. We aimed to develop and validate a dynamic CVD risk prediction tool using joint modelling that incorporates repeated measurements of cardiovascular risk factors.
Methods
We analysed 6637 participants aged 45-84 years from the Multi-Ethnic Study of Atherosclerosis. A Bayesian joint model was used to simultaneously analyze longitudinal risk factor trajectories and time to CVD. Repeated measurements of total cholesterol, systolic blood pressure, HDL-C, and use of lipid-lowering and antihypertensive medications were incorporated as time-varying covariates. Predictive performance was assessed using discrimination, calibration, overall accuracy, and reclassification measures.
Results
The joint model identified significant temporal trends, including declining cholesterol (-1.14 mg/dL/year) and increasing HDL-C (0.43 mg/dL/year), while systolic blood pressure remained stable. All risk factors were independently associated with CVD risk. Compared with a conventional model based on a single baseline measurement, the joint model demonstrated improved discrimination, calibration, overall accuracy, and reclassification, with gains becoming more pronounced as longer longitudinal histories were incorporated. A web-based application was developed to support individualized dynamic CVD risk prediction.
Conclusions
Incorporating longitudinal risk factor trajectories provides more accurate CVD risk estimates than conventional models. This approach improves risk stratification and may enhance clinical decision-making for preventive interventions. To our knowledge, this is among the first studies to translate joint modelling methodology into a clinically applicable CVD risk prediction tool.