Toward expert-level medical text validation with language models
摘要
With the growing use of language models (LMs) in clinical environments, there is an immediate need to evaluate the accuracy of LMs. Detecting errors in LM-generated text is challenging because (1) manual review is costly and (2) expert-composed reference outputs are often unavailable in real-world settings. While the “LLM-as-a-judge” paradigm (a LM evaluating another LM) offers scalable evaluation, even frontier LMs can miss subtle but clinically significant errors. To address these challenges, we propose MedVAL, a novel, self-supervised, data-efficient distillation method that leverages synthetic data to train evaluator LMs to assess whether LM-generated medical outputs are factually consistent with inputs, without requiring physician labels or reference outputs. To evaluate LM performance, we introduce MedVAL-Bench, a dataset of 840 physician-annotated outputs across 6 diverse clinical use cases capturing real-world challenges. Each output is reviewed following a physician-defined taxonomy of risk levels and error categories, enabling evaluation of LMs in making deployment safety decisions. Across 10 state-of-the-art LMs spanning open-source, proprietary, and medically adapted models, MedVAL distillation significantly improves (p < 0.001) alignment with physicians across seen and unseen tasks, increasing average F1 scores from 66% to 83%. Despite strong baseline performance, MedVAL improves the best-performing proprietary LM (GPT-4o) by 8% without training on physician-labeled data, demonstrating a performance statistically non-inferior to a single human expert on a subset annotated by multiple physicians (p < 0.001). To support a scalable pathway towards clinical integration, we open-source: (1) Codebase, (2) MedVAL-Bench, (3) MedVAL-4B. Across evaluated tasks, our study provides evidence of LMs approaching expert-level ability in risk-aware validation of LM-generated medical text.