Work

T2. Multimodal Hazard Score Integrates Genetics, Metabolomics, and Mental Health-related Risk Factors to Improve Cardiovascular Disease Risk Prediction

Year: 2025

Type: article

Abstract: Background Severe mental disorders (SMDs), including schizophrenia, bipolar disorder, and major depression are associated with markedly reduced life expectancy, compared to the general population, and in part driven by elevated risk of cardiovascular disease (CVD). Mounting evidence indicates the role of metabolic dysregulation in SMDs, consistent with abnormalities in lipid, glucose, and inflammatory profiles. Advances in high-throughput plasma metabolomics technologies, such as nuclear magnetic resonance (NMR) spectroscopy, now enable robust quantification of key metabolic biomarkers, which add predictive value beyond traditional biomarkers. While existing CVD risk models focus on single modalities, multimodal hazard scores (MHS) have potential to improve early detection of CVD, through integration of genetic, metabolic, and clinical risk factors, including prior SMD diagnosis and antipsychotic medication use, which may lead to tailored treatment and lifestyle interventions. Methods We developed a polygenic hazard score (PHS) to predict age of onset of coronary artery disease (CAD) using SNPs (p < 10-6) derived from the latest CARDIoGRAMplusC4D genome-wide association study (181,522 cases; 984,168 controls). The PHS was trained and tested (80/20%) in the UK Biobank (UKB: 41,918 CAD cases; 295,028 CAD-free controls) with genotype data, sex, and the first ten genetic principal components, using LASSO-regularized Cox proportional hazard regression models and 10-fold cross-validation. External validation was performed in the independent Hordaland Health Studies (HUSK) cohort (6,249 CAD cases; 26,578 controls). Performance was estimated by 1,000x bootstrapping. We further assessed 15-year CAD risk in a UKB test set (1,391 CAD cases; 27,527 CAD-free controls), by incorporating SMD diagnosis, atypical antipsychotic use, and other risk factors from an established clinical algorithm (QRISK3). We developed a biomarker hazard score (BHS) using 249 NMR-derived plasma metabolites. We then evaluated the predictive improvement of a MHS, which integrated the PHS with the clinical risk score (QRISK3), and the BHS. Results The PHS comprising 1,258 SNPs was a significant predictor of CAD age of onset in the UKB test set cohort (Hazard ratios [HR] for 80th vs. 20th percentile = 4.1; 95% confidence intervals [CIs]: 3.8–4.4) and the HUSK validation cohort (HR₈₀/₂₀ = 4.0; 95% CIs: 3.7–4.4). When predicting 15-year CAD risk in the UKB test set, QRISK3 alone yielded a HR₈₀/₂₀ of 9.3 (95% CIs: 7.4–11.7), which increased to 14.7 (95% CIs: 11.2–19.3), when combined with the PHS. A BHS, consisting of 75 plasma metabolites, yielded a HR₈₀/₂₀ of 16.9 95% CIs (12.3–23.2), which improved to 23.6 (95% CIs: 16.9–32.9), when combined with the PHS and QRISK3. Overall, the MHS model (PHS + QRISK3 + BHS) demonstrated improved time-dependent risk stratification for CAD, compared to single modalities alone. Discussion We present the first polygenic hazard model for CAD age of onset, validated in an independent cohort. Our findings demonstrate that multimodal integration of genetic, metabolic, and clinical risk factors, including prior SMDs diagnoses, as well as psychotropic medication use is promising framework for enhanced CVD risk stratification. The MHS enable identification of high-risk individuals which may inform personalized prevention strategies. Future work will focus on further validation in patients with SMDs and development of decision-support tools for psychiatric care settings. Disclosure: Nothing to disclose.

Source: European Neuropsychopharmacology

DOI: 10.1016/j.euroneuro.2025.08.285