Predictive Power of AI-Driven Population Genomic Risk Scores for Acute Decompensation in a Genetically Distinct Regional Heart Failure Cohort

Abstract

Background: The substantial morbidity associated with chronic Heart Failure necessitates advanced, proactive risk stratification tools, particularly in the UAE where Non-Communicable Diseases are a strategic focus. Current clinical models often neglect population-specific genomic variants, a critical factor within the genetically distinct regional cohort. This study validates a novel predictive model utilizing a Deep Learning Neural Network to integrate clinical and genetic data for predicting acute HF decompensation risk.  

Methods: A retrospective, tertiary-center cohort analysis enrolled 1,450 consecutive adults diagnosed with symptomatic chronic HF (NYHA class II-IV). Baseline clinical and laboratory data, harmonized from a regional patient registry, were paired with low-coverage whole-genome sequencing to derive a composite, population-specific Genomic Risk Score. A DLNN was trained to predict the primary endpoint: acute HF-related readmission within 90 days. Performance was rigorously evaluated using the Area Under the Receiver Operating Characteristic Curve and the Multivariate Cox Proportional Hazards model.  

Results: The integrated AI-GRS model demonstrated superior predictive capability for the 90-day primary endpoint (AUROC 0.88; 95% CI: 0.86-0.90), significantly outperforming prediction based on clinical factors alone (AUROC 0.72; p<0.001). Genomic Risk Score was independently and substantially associated with increased risk, showing an adjusted hazard ratio (HR) of 1.45 per one standard deviation increase (p=0.003). Patients categorized as high-risk by the AI-GRS and placed on an enhanced, personalized intervention protocol achieved a measured 45% reduction in all-cause-90-day readmissions (p=0.012). Conversely, readmission rates in the standard-care control group remained stable. Sensitivity analysis confirmed consistent model performance across all key demographic-subgroups.

Conclusion: Integrating AI-driven genomic risk stratification significantly improves the prediction and clinical management of acute HF exacerbations within the local population. These findings validate a critical, scalable-tool for enhancing patient outcomes and supporting the DHA’s vision for an integrated care delivery model of the future. Deployment of this model enables proactive-intervention strategies aligned with regional-digital-transformation mandates.