A novel plasma proteomic signature predicts 10-year incident heart failure risk among persons living with HIV

Background Effectively treated and virally suppressed persons living with HIV (PLWH) have an increased risk of heart failure compared with the general population. Currently, no methods to predict heart failure risk are available in PLWH. Purpose To create a proteomics-based model of incident heart failure risk in a longitudinal cohort of PLWH in the Veterans Aging Cohort Study Biomarker Cohort. Methods We characterized the baseline plasma proteome of 1,398 PLWH using an aptamer-based platform. Traditional risk factors and HIV-specific variables were incorporated into Cox proportional hazards models for incident heart failure. For risk models involving proteins, least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation was used to identify the optimal set of predictors. Model performance was evaluated using Harrell’s Concordance Index (C-index). Results Among 1,398 PLWH, 165 (11.8%) developed incident heart failure over a mean follow-up period of 9.6 years. The sample was divided randomly 2:1 into training (n = 932) and testing (n = 466) sets. In the training set, 4,248 proteins were screened, identifying 774 significant markers (false discovery rate- [FDR] adjusted p-value < 0.05); 492 of these were validated in the test set (p < 0.05) and were included as candidate predictors in downstream LASSO models. LASSO-Cox regression was used to identify a subset of predictor proteins in combination with clinical and laboratory risk factors, as well as protein-only models. Test concordance indices (C-index) in the validation data set are presented in Table 1. Among the clinical risk factor models, individual traditional risk factors in combination with HIV disease markers (CD4+ T-cell count, viral load) exhibited the best predictive performance (C-index 0.634; standard error [SE] 0.038). Addition of serum inflammatory and coagulation markers (soluble CD-14 [sCD14], interleukin-6 [IL-6], D-dimer) did not substantially improve performance (C-index 0.611, SE 0.042). By contrast, a protein-only model incorporating 11 markers achieved an improved C-index of 0.704 (SE 0.061), while the addition of CD4+ count and viral load performed similarly (C-index 0.708, SE 0.062). The protein-only risk model demonstrated improved risk discrimination compared with clinical risk factors, particularly in the highest-risk quartile (Figure 1; quartile-based C-indices 0.604 (SE 0.042) and 0.672 (0.036) for clinical and protein models, respectively). Conclusion An 11-protein signature demonstrated superior predictive performance for 10-year incident heart failure risk among PLWH compared with baseline traditional risk factors and inflammatory markers. If these findings can be externally validated, they highlight the potential of a proteomics-based approach for risk stratification in this population, which provides additive value beyond traditional or HIV-specific clinical risk factors.