Abstract Body: Background While clinical risk factors (CRF) have been used to predict heart failure (HF), an end-stage syndrome with high morbidity and mortality, the role of polygenic risk score (PRS) in identifying HF risk prediction remains unclear.
Method Using genome-wide summary statistics, we constructed 52 PRSs, including 1 for HF, 43 for echocardiographic(echo) traits, and 8 for CRF. Associations with incident HF were evaluated in 31,650 trans-ancestry participants (55% non-White; mean follow-up 16 years) from seven cohorts from the Trans-Omics for Precision Medicine (TOPMed) program, randomly divided into training (80%) and testing (20%) sets. Cox regression models adjusted for demographic and CRF were used to relate PRSs and to incident HF. Least absolute shrinkage and selection operator (LASSO) feature selection was performed on HF-related PRSs to build a multi-trait PRS (mPRS) in the training set. Associations of mPRS with incident HF, HF with preserved and reduced left ventricular (LV) ejection fraction (HFpEF and HFrEF, respectively), were examined in the testing set, with predictive performance evaluated by the C-statistic. Replication was conducted in 510,074 participants from the Million Veteran Program (MVP).
Results In the training set (n=25,320, HF cases=3,380), 17 out of 52 PRSs were associated with incident HF, where PRSs for HF and LV ejection fraction showed the largest effects (Figure 1a, HR_HF-PRS: 2.02,95% CI 1.97-2.08, HR_LVEF-PRS:0.93, 95% CI 0.89-0.97, p-value <0.05). Nine HF-related PRSs (1 HF, 6 echo, 2 CRF) were selected by LASSO to construct a multi-trait PRS (mPRS). In the testing set (n=6330, HF cases=856), per SD increase of mPRS was associated with about twofold increase in the risk of incident HF and its subtypes (Figure 1b, HR_HF: 1.81, 95% CI 1.68–1.96; HR_HFpEF: 2.08, 95% CI 1.78-2.43; HR_HFrEF:1.97, 95% CI 1.68-2.31), and the associations were stronger in White vs Non-white participants. The associations of mPRS with incident HF and its subtypes were replicated in MVP (Figure 1b, HR_HF: 1.13, 95% CI 1.11–1.15; HR_HFpEF: 1.09, 95% CI 1.07-1.11; HR_HFrEF:1.17, 95% CI 1.15-1.19). Adding mPRS over CRF modestly improved HF and HFrEF prediction in the trans-ancestral and white group (Figure 1c, HF_{trans-ancestral}: estimated C: 0.89, delta C=1%, all p-values<0.05).
Conclusion: The component PRS and mPRS were associated with HF risk, and mPRS can improve the prediction of incident HF, highlighting the potential for identifying at-risk populations.