Adaptive system improves medication advice using patient history
ARMED: Adaptive Retrieval-Augmented Medication Recommendation
Machine LearningArtificial IntelligenceComputation and LanguageInformation Retrieval
Summary
Making good medication recommendations for patients is complicated because doctors need to consider many factors like diagnoses, other health issues, and how different treatments interact. The authors present ARMED, a system that improves this process by smartly finding and using the most relevant past patient information. Unlike older methods that always search for data the same way, ARMED changes how it looks for information based on how much it already found and how closely it relates to the current patient visit. Tests show ARMED gives better medication suggestions without needing large external AI models or special training.
What this means in practice
- •For hospital data teams: Integrate adaptive retrieval methods to improve medication suggestions from patient records during clinical workflows.
- •For healthcare software developers: Build medication recommendation tools using locally runnable small language models without relying on external APIs or fine-tuning.$Commercial implications: Enables development of autonomous medication recommendation products suitable for clinical deployments with privacy and cost advantages.
Tested on one dataset.
Authors
Kyungkeon Chung, Se Un Park, Hyejin Lee, Junyoung Park
Abstract
Medication recommendation requires jointly considering diagnoses, comorbidities, longitudinal patient history, and treatment interactions. Although recent retrieval-augmented generation approaches using language models have shown promise for clinical recommendation, most existing systems rely on fixed retrieval pipelines that cannot adapt to diverse clinical retrieval scenarios and may introduce noisy or irrelevant evidence. We propose ARMED, an adaptive clinical RAG framework that dynamically expands retrieval according to retrieval sufficiency and current-admission relevance. ARMED first observes the initial retrieval outcome and selectively performs supplementary or fallback multi-query retrieval when additional evidence is needed. The framework also incorporates a relevance gate that activates only clinically relevant same-patient history and organizes retrieved evidence into structured query-level clinical evidence. Experiments on MIMIC-IV demonstrate that ARMED consistently outperforms previous static retrieval baselines and improves overall medication recommendation quality. By combining locally runnable small language models with adaptive retrieval, ARMED supports medication recommendation without relying on external language model APIs or task-specific fine-tuning.