# Clinical subgroup-stratified plasma proteomic signatures improve risk prediction for myocardial infarction: SCORE2-Pro

## TL;DR

A new model called SCORE2-Pro uses blood proteins to better predict heart attack risk in different groups of people, improving accuracy over existing methods.

## Contribution

Introduces a subgroup-specific plasma proteomic model for myocardial infarction risk prediction that outperforms traditional clinical models.

## Key findings

- A 9-protein model improved risk prediction in females with a 0.061 AUC increase and 0.125 NRI improvement.
- A 7-protein model enhanced prediction in middle-aged individuals with a 0.036 AUC increase and 0.127 NRI improvement.
- The model significantly improved reclassification in high-risk populations with an NRI of +0.121.

## Abstract

Myocardial infarction (MI) remains a leading cause of global mortality, with risk varying substantially across demographic and clinical subgroups. Although SCORE2 is widely implemented for cardiovascular risk stratification, the extent to which clinical subgroup specific plasma proteomics can further refine personalized MI risk prediction remains uncertain.

SCORE2-Pro, a clinical subgroup-stratified plasma proteome prediction model was built stratified by sex, age, smoking status, non-high-density lipoprotein (non-HDL) cholesterol, and systolic blood pressure. In 51,010 UK Biobank participants (aged 40–69 years; 54.9% female) without MI at baseline, 70% were used for model development, and the remaining 30% for an internal hold-out validation. We used light gradient boosting machine classifiers and Cox proportional hazards models to identify top-predictive protein combinations and stratification strategies for MI.

The SCORE2-Pro model revealed distinct and highly effective protein panels for each subgroup. Compared with the clinical model, SCORE2-Pro remarkably enhanced predictive performance across demographic and clinical subgroups. A 9-protein model in females improved AUC by + 0.061 (P = 1.51 × 10−3), with a net reclassification index (NRI) of + 0.125 (P = 2.69 × 10−6). Similarly, a 7-protein model for middle-aged subpopulation demonstrated an AUC improvement of + 0.036 (P = 0.033) with an NRI of + 0.127 (P = 8.09 × 10−11). Notably, in high-risk populations, SCORE2-Pro model significantly improved reclassification performance compared with the clinical model (NRI: + 0.121, P = 0.020).

By adopting a clinical subgroup stratification approach using factors derived from SCORE2, we identified subgroup-specific proteomic signatures for MI that considerably improve predictive accuracy and reclassification beyond traditional clinical models.

The online version contains supplementary material available at 10.1186/s12933-025-03050-7.