MedUPS: Towards Diagnostic Assistance in Uncommon Medical Cases with Large Language Models

2026-08-02

AI Summary

This study addresses the challenge of clinical decision support in rare or non-guideline-adherent cases, where existing evaluations of large language models predominantly focus on final diagnoses while neglecting intermediate reasoning under diagnostic uncertainty. To bridge this gap, the authors propose the MedUPS alignment framework and construct the MedUPSQA dataset, comprising 21,874 intermediate decision points derived from temporally segmented clinical cases. Leveraging supervised fine-tuning (SFT) and group relative policy optimization (GRPO), augmented with an LLM-as-a-Judge mechanism to provide temporally aligned reward signals, the approach explicitly models sequential clinical reasoning. Evaluated on Qwen3.6-27B, Qwen3.5-9B, and HuatuoGPT-3-8B, the method achieves next-step decision accuracies of 66.7%, 57.8%, and 44.4%, respectively—significantly outperforming baseline approaches, with some smaller models even surpassing larger state-of-the-art counterparts.

Abstract

Uncommon and off-guideline cases are difficult for clinical decision support, because physicians must make a series of management decisions under diagnostic uncertainty and rarely see the full case at once. Most large language model (LLM) benchmarks for medicine score only the final diagnosis, yet much of clinical care turns on the next appropriate action: the next test to order, the imaging study to obtain, the specialist to involve, or the differential to pursue. We introduce MedUPSQA, a dataset of 21,874 mid-stream clinical decision points built from 5,535 real case reports, and MedUPS, an alignment framework that supervises models on these intermediate decisions as they unfold along a patient's trajectory. We segment free-text case presentations into chronologically ordered, accumulating clinical chunks and align models to predict the next step with reinforcement learning (GRPO), using an external LLM-as-a-Judge reward. This objective mirrors how clinicians actually meet patients, reasoning forward from accumulating evidence toward the next decision, rather than committing to a final label. Across three backbones, mid-stream alignment raises next-step accuracy from 55.2 to 66.7 for Qwen3.6-27B, from 47.2 to 57.8 for Qwen3.5-9B, and from 37.8 to 44.4 for HuatuoGPT-3-8B, with 95% CI. In several model scales we test the objective improves accuracy more than scale, with smaller models surpassing larger, frontier models we evaluate. We further train supervised fine-tuning (SFT) baselines on the mid-stream task, SFT improves all backbones above base, indicating the target framwork carries signal independently of the optimizer. We release the dataset, code, and aligned checkpoints.

Problem

• uncommon medical cases
• clinical decision support
• diagnostic uncertainty
• next-step prediction
• mid-stream clinical decisions

Innovation

• clinical decision support
• large language models
• reinforcement learning
• next-step prediction
• rare medical cases