Personalized feature threshold estimation in joint modelling of longitudinal and time-to-event data

2025-06-21

AI Summary

This study addresses the limitation of static thresholds in cardiovascular disease (CVD) risk prediction, which fail to account for inter-individual heterogeneity. We propose a variable-threshold joint modeling framework that simultaneously models longitudinal trajectories of biomarkers—including BMI, blood pressure, and blood glucose—and time-to-event outcomes, enabling adaptive threshold estimation stratified by sex, race, and biomarker type. Leveraging data from the Atherosclerosis Risk in Communities (ARIC) cohort, our approach integrates multivariate longitudinal trajectory modeling with survival analysis to derive individualized risk thresholds. Compared to conventional fixed-threshold methods, the proposed framework significantly improves sensitivity and specificity in identifying high-risk individuals, enhances risk stratification accuracy, and increases clinical interpretability. By providing statistically rigorous, personalized threshold estimation, this work establishes both a methodological foundation and a practical tool for precision prevention of CVD.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Graphical Models
• Constraint Satisfaction and Optimization: Mixed Discrete/Continuous Optimization

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Cardiovascular disease (CVD) cohort studies collect longitudinal data on numerous CVD risk factors including body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), glucose, and total cholesterol. The commonly used threshold values for identifying subjects at high risk are 30 kg/$m^2$ for BMI, 120 mmHg for SBP, 80 mmHg for DBP, 126 mg/dL for glucose, and 230 mg/dL for total cholesterol. When studying the association between features of longitudinal risk factors and time to a CVD event, an important research question is whether these CVD risk factor thresholds should vary based on individual characteristics as well as the type of longitudinal feature being considered. Using data from the Atherosclerosis Risk in Communities (ARIC) Study, we develop methods to estimate risk factor thresholds in joint models with multiple features for each longitudinal risk factor. These thresholds are allowed to vary by sex, race, and type of feature. Our methods have the potential for personalized CVD prevention strategies as well as better estimates of CVD risk.

Problem

• Estimating personalized thresholds for CVD risk factors
• Assessing variability by sex, race, and feature type
• Improving CVD risk prediction and prevention strategies

Innovation

• Personalized thresholds for CVD risk factors
• Joint modeling of longitudinal and event data
• Thresholds vary by sex, race, feature type