Development and validation of a 2-year MACCE prediction model incorporating routine biomarkers in young patients with premature coronary heart disease: a multicenter study
J Xu, Y Song, Z Ma, Y C Hao, L Jiang, X Ding, Z Gao, Y Zhang, X Y Zhao, L Song, J Q YuanAbstract
Background
Young patients with premature coronary heart disease (PCHD) are at high lifetime cardiovascular risk, yet existing risk scores are poorly calibrated for this population. Although inflammation and lipid accumulation are central to PCHD pathogenesis, the predictive value of LDL-C and hs-CRP for short-term adverse events in patients under 45 remains unclear. As a prelude to our ongoing National Science and Technology Major Project-funded cohort study, we aimed to develop a prediction model using existing multicenter data to establish a methodological framework and identify key predictors.
Purpose
To develop and validate a 2-year MACCE prediction model for PCHD patients aged ≤45 years and evaluate the incremental value of adding LDL-C and hs-CRP to conventional clinical factors.
Methods
We derived a 2-year MACCE prediction model from the multicenter PROMISE study (2015–2019; n=1091; PCHD patients aged ≤45 years) and validated it in an independent single-center cohort (2013; n=840). MACCE included all-cause death, myocardial infarction, revascularization, or stroke. Predictors included age, sex, BMI, hypertension, diabetes, prior revascularization, SYNTAX score, lipoprotein(a), uric acid, eGFR, LDL-C, and hs-CRP, selected by clinical consensus. Cox and Lasso-Cox models were developed with 10-fold cross-validation; hyperparameters were tuned via nested cross-validation. Model performance was assessed using the C-index with bootstrapped 95% CIs. Analyses used R 4.5.2.
Results
The derivation cohort included 1,091 patients (mean age 40.4±4.2 years; 94.4% male), with 86 MACCE cases (7.9%) during 2-year follow-up. The validation cohort included 840 patients (40.9±4.0 years; 94.5% male), with 86 MACCE cases (10.2%). The clinical-only Cox model achieved a cross-validated C-index (CV-Cindex) of 0.614 and external C-index of 0.659. The Lasso-Cox model demonstrated improved stability (CV-Cindex: 0.653; external C-index: 0.660). Adding LDL-C or hs-CRP individually yielded modest improvements: for the Cox model, CV-Cindex increased to 0.624 (LDL-C) and 0.627 (hs-CRP), with external C-indices of 0.654 and 0.662, respectively. The combination of both biomarkers yielded the highest performance in both the Cox model (CV-Cindex: 0.634; external C-index: 0.656) and Lasso-Cox model (CV-Cindex: 0.652; external C-index: 0.661) (Table 1). All models demonstrated statistical significance (p < 0.001).
Conclusion
In young PCHD patients lacking validated risk tools, a simple model with routine LDL-C and hs-CRP improved 2-year MACCE prediction. These findings support combined biomarker–clinical stratification in this understudied high-risk population and will inform variable selection for our ongoing National Science and Technology Major Project-supported cohort study.Table 1.Model PerformanceFigure 1 C-index of Prediction Models