Evaluation of an Artificial intelligence (AI)–Assisted Decision Support for Medication Supply in Stepdown Care Transitions

Background: Healthcare transitions from acute hospitals to step-down care facilities require careful coordination of medication supply, balancing continuity of care with resource optimization.Traditional approaches that provide full medication courses often lead to much waste when receiving facilities maintain their own formularies.While facility-specific supply strategies help prevent oversupply, they introduce workflow complexities that require manual cross-checking of medication orders against institution-specific formularies-a process that is time-consuming and error-prone.This study evaluated the technical performance, processing efficiency, and user experience of an artificial intelligence (AI) assistant for medication supply decisions during stepdown care transitions. Methods:We conducted a prospective evaluation comprising three components: a technical evaluation of 100 discharge prescriptions (1,373 medication orders) with pharmacist validation; a time-motion analysis of 25 prescriptions comparing manual versus AI-assisted processing; and a user experience survey of pharmacy staff.The evaluation was repeated after transitioning to a new dispensing system using 100 prescriptions (1,124 medication orders). Results:The AI assistant achieved 99.5% overall accuracy in the former system and 99.4% in the new system when cross-checking medications against an institution-specific formulary.Sensitivity was 99.3% and 100%, respectively, with false negatives decreasing from three to zero cases.Time-motion analysis demonstrated a 68.5% reduction in processing time per medication (median 5.4 to 1.7 seconds), with greater efficiency gains for longer prescriptions.Among 13 survey respondents, 92.3% reported satisfaction, 84.6% rated recommendations as mostly or very accurate, and 92.3% perceived time savings exceeding 3 minutes per prescription despite measured savings of approximately 48 seconds for typical prescriptions, suggesting a substantial reduction in cognitive burden. Conclusion:This AI assistant demonstrates high accuracy and improves efficiency in medication supply decisions during care transitions.The solution effectively processes unstructured prescription data while supporting workflow optimization and sustainability initiatives by reducing medication oversupply.However, given the clinical significance of potential errors, particularly false negatives, the system must remain a decision support tool with mandatory human oversight.