AI-Driven Pharmacogenomic Optimization of Drug Therapy in Patients With Type 2 Diabetes Mellitus and Cardiovascular Diseases

Background: The co-occurrence of type 2 diabetes mellitus (T2DM) and cardiovascular diseases (CVD) represents one of the most clinically complex and pharmacologically challenging comorbidity profiles encountered in contemporary medicine. Pharmacogenomics offers a precision medicine framework for individualizing drug therapy according to patientspecific genetic variants in drug-metabolising enzymes, transporters, and pharmacodynamic targets. However, the translational implementation of pharmacogenomic data into clinical decision-making for T2DM-CVD patients remains limited by the complexity of multi-drug regimens, variant interpretation, and real-world scalability. Objective: To develop and prospectively validate an artificial intelligence (AI)-driven pharmacogenomic decision support system for the optimization of drug therapy in patients with concurrent T2DM and CVD, and to evaluate its impact on glycaemic control, cardiovascular outcomes, and adverse drug reaction (ADR) rates at 6-month follow-up. Methods: A prospective, randomised, parallel-arm controlled trial was conducted at two tertiary cardio-endocrinological centres (January 2022–June 2024), enrolling 620 patients aged 40–80 years with confirmed T2DM (HbA1c ≥7.0%) and established CVD (coronary artery disease, heart failure, or atrial fibrillation). Participants were randomised 1:1 to an AIguided pharmacogenomic group (n=310) or standard care group (n=310). All enrolled patients underwent comprehensive pharmacogenomic profiling (CYP2C19, CYP2D6, CYP2C9, SLCO1B1, ABCB1, UGT1A1, VKORC1). An XGBoostbased AI decision support model integrated genomic, clinical, and laboratory data to generate individualised drug-dose recommendations in the intervention group. Primary outcomes were HbA1c change and major adverse cardiovascular events (MACE) at 6 months. Secondary outcomes included LDL-C, systolic blood pressure (SBP), fasting plasma glucose (FPG), and ADR rates. Results: The AI-guided group achieved significantly greater HbA1c reduction (−1.5±0.4% vs. −0.8±0.5%; p<0.001), greater LDL-C reduction (−1.2±0.4 vs. −0.7±0.4 mmol/L; p<0.001), and greater SBP reduction (−20.1±6.3 vs. −11.4±7.1 mmHg; p<0.001). MACE occurred in 4.4% of the AI-guided group versus 9.3% of the standard care group (HR 0.46, 95% CI 0.28–0.76; p=0.002). The composite ADR rate was 18.5% versus 31.6% (p<0.001). The XGBoost model demonstrated an AUC of 0.887 for MACE prediction and 0.863 for ADR prediction. Actionable pharmacogenomic variants were identified in 68.4% of patients, with CYP2C19 poor metaboliser status conferring the highest predictive weight for adverse clopidogrel outcomes. Conclusion: AI-driven pharmacogenomic decision support significantly improves glycaemic control, cardiovascular risk reduction, and drug safety in patients with concurrent T2DM and CVD. Integration of genomic data into AI-powered clinical decision support systems represents a scalable and clinically impactful precision medicine strategy for this highrisk population.

Ҳали таржима қилинмаган