Llms struggle to maintain reasoning accuracy in medical risk assessments
Can LLMs Follow Medical Expert Logic? A Benchmark for Hierarchical Logical Consistency in Risk-of-Bias Assessment
Artificial Intelligence
Summary
Assessing the quality of medical studies needs careful logical thinking. This paper introduces a test to see how well large language models (LLMs) follow strict medical reasoning rules. The authors found that even if LLMs get many individual facts right, their overall conclusions often fail logical checks. This reveals that current models might seem accurate but actually make big reasoning mistakes when judging medical studies.
What this means in practice
- •For medical ai developers: Evaluate and improve AI systems for bias assessment by testing logical consistency beyond label accuracy in clinical trial analysis.
- •For clinical ai validation teams: Use the benchmark to verify AI reasoning correctness when deploying language models in medical decision support systems.
Authors
Jiayu Huang, Zichen Tang, Qianhui Ling, Zemin Kuang, Haihong E
Abstract
Evidence-based medicine demands strict logical consistency, yet current evaluations of large language models (LLMs) prioritize superficial label matching over genuine reasoning. We introduce LogiMed-RoB, a benchmark grounded in Cochrane Risk of Bias (RoB) 2.0 expert logic, comprising 860 randomized controlled trials (RCTs) and 14,820 queries. It evaluates models under the Hierarchical Logical Consistency (HLC) framework across four dimensions: Atomic Consistency, Domain Consistency, Aggregation Consistency, and Evidential Faithfulness. Experiments on 10 state-of-the-art LLMs reveal a catastrophic Error Compounding Effect: despite the top model reaching 98.88% Atomic Consistency, its end-to-end consistency collapses to 45.13%, with several open-weight architectures plummeting to nearly 0%. We further uncover a systematic evidence-reasoning gap: even when models retrieve high-quality evidence, they fail to deduce correct outcomes in 18.63-40.05% of cases, while Blind Guess Rates reach 48.28%. LogiMed-RoB demonstrates that high outcome accuracy can conceal critical reasoning flaws, underscoring the necessity of white-box logical verification for clinical deployment.