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

Background

Precision-guided dual antiplatelet therapy (DAPT) duration post-percutaneous coronary intervention (PCI) remains a clinical challenge. Current risk stratification methods lack personalization, underscoring the need for advanced predictive tools. We developed and validated an artificial intelligence (AI) framework (LightGBM, random forest [RF], logistic regression [LR]) to optimize DAPT duration using multinational datasets.

Methods

Structured clinical data from 5,000 patients (Bayanat Data Portal, UAE; MIMIC-IV PhysioNet, global) were analyzed. The dataset was split into training (70%), validation (15%), and test (15%) sets, ensuring balanced outcomes. Primary endpoints were ischemic and bleeding events over 37 months. Models predicted outcomes and recommended patient-specific DAPT durations. Performance was evaluated using area under the curve (AUC), Kaplan-Meier survival analysis, and feature importance. Calibration and cross-validation ensured generalizability, while cost-effectiveness was assessed via healthcare utilization and quality-adjusted life years (QALYs).

Results

Key predictors included obesity, prior myocardial infarction, and CYP2C19 genetic risk. LightGBM outperformed conventional risk scores (AUC 0.89 vs. 0.75; *p* < 0.001). AI-guided DAPT significantly improved event-free survival (log-rank *p* < 0.01), reducing overtreatment in low-risk patients and bleeding in high-risk groups. The AI strategy yielded cost savings (17,150 AED/patient) and superior cost-effectiveness (ICER: –451,315 AED/QALY).

Conclusion

AI-driven DAPT personalization enhances risk prediction, treatment safety, and clinical outcomes post-PCI. This framework demonstrates real-world potential, particularly when incorporating regional data for tailored decisions. Prospective validation is warranted to confirm these findings.

Clinical Perspective

• Therapeutic Gap: Existing DAPT guidelines provide wide-ranging approvals, but personalized therapy remains a limitation due to interpatient unpredictability in ischemic and bleeding risk.

• Novelty of AI Implementation: This study exceptionally integrates AI-driven models trained on both UAE-specific and worldwide datasets to enhance DAPT duration selection.

• Improved Risk Stratification: Machine learning algorithms exhibit conventional risk assessment tools in evaluating ischemic and bleeding measures, allowing a more accurate balance of security and effectiveness.

• Clinical Translational Potential: Using both regional and global data ensures diverse population applicability while preserving culturally and epidemiologically adapted application.

• Economic Influence: Cost-effectiveness analysis validates that AI-driven DAPT modifications reduce unnecessary treatment and health care insignificance while ensuring cardiovascular protection.

Clinical Implications

• Patient-tailored Antiplatelet Regimens: AI-based models qualify individualized clinical management strategy, affecting beyond standardized Fixed-duration DAPT paradigms.

• Attenuation of Complications: Evidence-based treatment span lowers ischemia-vulnerable patients while preserving coagulation stability in those with favorable risk profiles.

• Data-Driven Clinical Optimization: Healthcare providers can utilize computational risk stratification to calibrate pharmacotherapeutic precision, facilitating treatment persistence and efficacy measures.

• Geographically tailored care innovations: Conclusions propose a model for machine learning incorporation in Middle Eastern heart care.

• Prospective Investigative Priorities: Results provide the foundation for longitudinal validations of AI-driven risk assessment in multinational samples.