Researchers Propose Graph-Based Framework for Medication Safety in Older Adults with Multimorbidity

Researchers Propose Graph-Based Framework for Medication Safety in Older Adults with Multimorbidity

Researchers have introduced ATLAS, a coupled graph-policy distillation framework designed to improve medication safety recommendations for older adults with multiple health conditions, according to a paper published on arxiv.org.

According to the research, ATLAS addresses a key challenge with large language model (LLM) agents: safe medication choices for older adults depend on conditions, medications, and geriatric risks that users may omit during consultations. The framework structures guideline evidence as a medication-safety graph, where targeted questions update patient state and distill relevant relations into a patient-specific medication conflict graph (PMCG).

The system employs what the researchers describe as a “risk-first multi-agent policy” that uses the PMCG to screen contraindications, assess cautions and monitoring needs, identify safer alternatives, and verify the final medication plan.

The researchers also introduced GeriMedBench, an interactive benchmark for testing safety-critical information acquisition and evidence-based decision revision. According to the paper, ATLAS achieved the strongest complete-decision performance among compared systems across three benchmarks: a European non-interactive multimorbidity benchmark, an Asian interactive multimorbidity benchmark, and an Asian non-interactive cross-guideline benchmark.

On the European benchmark, ATLAS exceeded the strongest proprietary LLM baseline by 53.73 points in Strict Success Rate and 14.63 points in overall safety reasoning score, with no unsafe recommendations under automated evaluation. A blinded clinician evaluation gave ATLAS higher mean ratings across all five criteria and flagged potentially unsafe recommendations in one ATLAS case and two cases using Gemini.