International Conference EFMI-STC 2026, Athens 2026

​

Synergistic Polypharmacy Optimization Using Multi-Agent AI and Structured Decision Support

Adnan KULENOVIC and Azra LAGUMDZIJA-KULENOVIC 
Absolute Information Age, Inc., Toronto, Canada
 

Abstract. Polypharmacy management requires both clinical reasoning across multiple conditions and systematic evaluation of adverse drug reactions and interactions among treatment components. This study investigates a synergistic approach that integrates a multi-agent system (MAS) of specialist large language model (LLM) agents with PM-TOM (Personalized Medicine–Treatment Optimization Method), a structured polypharmacy analysis and optimization system. The study hypothesis is that combining agentic clinical reasoning, structured analysis and optimization, and human-guided refinement can identify safer candidate polypharmacy regimens while compensating for key limitations of each approach. This study was based on a published deprescribing case of a 79-year-old man with multiple conditions and medication-related problems. We compared the published discharge treatment (V1) with an AI agent-generated replacement treatment (V2), a PM-TOM-optimized treatment selected from AI-generated drug alternatives (V3), and an AI-assisted clinically refined treatment (V4). PM-TOM evaluated patient-specific drug–adverse drug reactions (ADR), drug–drug interactions (DDIs), drug–condition interactions (DCIs), and Beers Criteria risks. Compared with V1, V4 reduced ADR associations from 38 to 27, critical DDIs from 11 to 4, DCIs from 10 to 7, and combined Beers risk indicators from 9 to 2. The study shows how agentic clinical reasoning, deterministic analysis and optimization, and human judgment can complement each other to improve the safety of complex polypharmacy treatments.
 

Keywords. Polypharmacy, Clinical Decision Support, Multi-Agent Systems, Large Language Models, PM-TOM, Adverse Drug Reactions, Beers Criteria.