AI-Driven Personalization of Dual Antiplatelet Therapy Duration Post-PCI: A Novel Approach Balancing Ischemic and Bleeding Risks in the UAE Population

Background: Determining the optimal dual antiplatelet therapy (DAPT) duration remains a pivotal concern in managing patients following percutaneous coronary intervention (PCI). While current guidelines emphasize risk stratification, the integration of artificial intelligence (AI)-driven models, including LightGBM, random forest, and logistic regression, for personalized treatment recommendations has not been extensively explored. This study develops and validates an AI-driven framework that leverages UAE-specific and global datasets to refine DAPT duration and optimize post-PCI outcomes. Methods: Patient data from the Bayanat Data Portal (UAE) and the global MIMIC-IV PhysioNet database were analyzed. Baseline characteristics, ischemic and bleeding events, and long-term clinical outcomes were assessed over a 37-month follow-up period. Among the tested AI models, LightGBM demonstrated the highest predictive accuracy compared to conventional DAPT risk scores. Kaplan-Meier survival analysis, receiver operating characteristic (ROC) curves, and feature importance analysis were used to compare risk-adjusted DAPT strategies. Cost-effectiveness was evaluated based on healthcare resource utilization and quality-adjusted life years (QALYs). Results: Among 5,000 patients, factors such as obesity, prior myocardial infarction (MI), and genetic predispositions significantly influenced DAPT-related outcomes. LightGBM achieved an area under the curve (AUC) of 0.89, surpassing conventional risk scores (AUC: 0.75, p<0.001). Kaplan-Meier curves revealed a significant survival advantage for AI-personalized DAPT (log-rank p<0.01). Shorter DAPT durations increased ischemic risk in high-risk patients, while longer therapy heightened bleeding complications. AI-driven risk stratification reduced unnecessary medication exposure,translating into improved cost-effectiveness and optimized treatment outcomes. Conclusion: AI-based DAPT personalization significantly enhances risk prediction and clinical decision-making, outperforming traditional models. Integrating UAE-specific data ensures regional applicability, reinforcing the need for precision-driven post-PCI management. These findings support AI-powered decision support systems as a transformative approach to improving cardiovascular outcomes, warranting further validation in prospective trial