Fidelity of Medical Reasoning in Large Language Models

2025/08/08 by Suhana Bedi, Yixing Jiang, Philip Chung +2 · 2 voices · 34 citations
Computer Science · Medicine · Psychology · #Artificial Intelligence in Healthcare and Education #Artificial intelligence #Cognition #Computer science #Explainable Artificial Intelligence (XAI) #Fidelity #Logical reasoning #Machine Learning in Healthcare #Natural language processing #Psychology #Verbal reasoning

paper · pdf · doi:10.1001/jamanetworkopen.2025.26021

published in JAMA Network Open 8(8), e2526021 (American Medical Association)

openalex publication_date 2025/08/08 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/30

Abstract

This cross-sectional study evaluates whether the performance of large language models on medical benchmarks reflects logical reasoning or pattern recognition.

Cited by

• VeriLLMed: Interactive Visual Debugging of Medical Large Language Models with Knowledge Graphs
• A Real-World Evaluation of LLM Medication Safety Reviews in NHS Primary Care
• The Ouroboros of Benchmarking: Reasoning Evaluation in an Era of Saturation
• Measuring Language Model Hallucinations Through Distributional Correctness
• Performance Assessment Strategies for Generative AI Applications in Healthcare
• MindBenchAI: An Actionable Platform to Evaluate the Profile and Performance of Large Language Models in a Mental Healthcare Context
• Do explanations generalize across large reasoning models?
• What Does the AI Doctor Value? Auditing Pluralism in the Clinical Ethics of Language Models
• Mindbench.ai: an actionable platform to evaluate the profile and performance of large language models in a mental healthcare context. [europepmc]
• Disclaimers and Referral Patterns for Medical Advice Across Urgency Levels: Large Language Model Evaluation Study. [europepmc]
• Development and prospective shadow evaluation of a domain-specific large language model for emergency neurological diagnosis. [europepmc]
• Performance of Large Language Models in Oral Health Consultations and the Consistency of the 'AI-as-a-Judge' Framework. [europepmc]
• Physician-Reported Safety Outcomes of AI-Generated Hospital Course Summaries. [europepmc]
• Use of Generative Artificial Intelligence in the Management of Low Back Pain: a Scoping Review. [europepmc]
• Medical errors in large language models revealed using 1,000 synthetic clinical transcripts [europepmc]
• On the robustness of medical term representations in locally deployable language models [europepmc]

Discussions

• (small) study eliminates correct multiple choice answer and replaces it with ‘None of the above’, and LLMs flounder I didn’t know they were getting multiple choice to begin with, bit of a head-start [bsky, 5 points, 0 comments]
• 🖥️ **Fidelity of Medical Reasoning in Large Language Models** "_When forced to reason beyond familiar answer patterns, all models demonstrate declines in accuracy, challenging claims of artificial in [mastodon, 0 points, 0 comments]

Related

• Disentangling Reasoning and Knowledge in Medical Large Language Models
• Performance of Large Language Models in Supporting Medical Diagnosis and Treatment
• Performance of a large language model on the reasoning tasks of a physician
• Automating Expert-Level Medical Reasoning Evaluation of Large Language Models
• Large Language Model Influence on Diagnostic Reasoning