Integrating biomarkers and machine learning for early prediction of incident heart failure in older adults: a systematic review and meta-analysis
S Fernandes, S S Kulkarni, S Satishkumar, M A Quamar, M Rasheed Muhammad, S Rasheed Muhammad, M Muhsina, B Tahir, A AhamedAbstract
Background/Purpose
Early identification of older adults at highest risk for heart failure (HF) is an urgent, unmet clinical need. We performed a systematic review and meta-analysis to integrate the prognostic evidence for emerging biomarker and machine learning (ML) risk scores and to derive a standardized, evidence-based prevention framework.
Methods
We conducted a PRISMA 2020-compliant systematic review and meta-analysis, registered with PROSPERO. We searched PubMed, Scopus, and ScienceDirect from January 1, 2020, to November 1, 2025, for observational studies in adults ≥65 years reporting adjusted hazard ratios (HRs) or discrimination metrics (AUC/C-index) for incident HF. Two reviewers independently screened studies, extracted data, and assessed the risk of bias using the PROBAST tool. Random-effects models were used for quantitative synthesis.
Results
After screening 2,350 records, 49 studies met the inclusion criteria. Of these, 11 studies, reporting on 13 distinct risk scores, provided sufficient data for meta-analysis. The pooled hazard ratio (HR) for incident HF, per one-standard-deviation increase in risk score, was 1.48 (95% CI 1.30–1.69), with considerable heterogeneity (I² = 81%) that is expected when synthesizing diverse, multidimensional risk scores.. Restricting the analysis to the 7 scores rated as low risk of bias yielded a stronger pooled HR of 1.60 (95% CI 1.38–1.85). The most predictive scores centered on phenotypes of inflammation, diabetic cardiomyopathy, and lipidomic dysregulation, with HRs ranging from 1.75 to 1.95. Aggregate discrimination was excellent, with a pooled AUC of 0.88 (95% CI 0.79–0.96). Using this synthesis, we propose 4-pillar "HF-AI Risk" framework integrating: 1) Core Biomarkers (e.g., NT-proBNP, hsCRP), 2) Comprehensive Metabolic Panels, 3) Organ-Specific Stress Markers (e.g., GDF-15), and 4) an ML-Integration Layer for dynamic, personalized risk stratification.
Conclusion
This first comprehensive synthesis of contemporary evidence confirms that biomarker and ML-derived risk scores are powerful, independent predictors of incident HF in older adults. The derived HF-AI Risk framework, which prioritizes inflammatory and metabolic pathways, provides an immediate, evidence-based blueprint for clinical trials and health systems. It enables a paradigm shift from reactive HF care to precise, AI-augmented primary preventionFor image description, please refer to the figure legend and surrounding text.