Abstract
Drug-drug interactions (DDIs) represent a major preventable etiology of adverse drug events (ADEs) in inpatient hospital settings, particularly among multimorbid patients subjected to complex polypharmacy. Conventional computerized provider order entry (CPOE) systems utilize static, rule-based screening tools characterized by excessive sensitivity and poor specificity, precipitating profound alert fatigue and high override rates (>90%). In this study, we implemented and prospectively evaluated a machine learning-augmented, context-aware DDI screening engine embedded within the electronic health record (EHR) of an 800-bed academic medical center. Over a 12-month pre-post interventional study spanning 377,412 medication orders, the artificial intelligence (AI) tool parsed patient-specific clinical parameters—including renal biomarkers, hepatic panel trends, serum electrolytes, and chronological administration intervals—to dynamically triage drug interaction risks. Implementation of the AI screening tool resulted in a 62.4% reduction in total alert volume and an increase in clinical pharmacist intervention acceptance rates from 68.2% to 91.5% (p < 0.001). Concurrently, verified prescribing errors involving high-risk DDIs decreased by 41.8% (from 14.8 to 8.6 per 1,000 patient-days; relative risk: 0.58, 95% CI: 0.51–0.67). These findings demonstrate that context-aware AI architecture substantially mitigates alert fatigue, accelerates pharmacy verification workflows, and significantly improves medication safety in institutional pharmacy practice.
