mlsanityData to 14 Oct 2026

When Chain-of-Thought Backfires: Evaluating Prompt Sensitivity in Medical Language Models

arXiv 2603.25960 ↗First posted 26 Mar 2026cs.CL cs.AIcited by 1

On the map: LLM reasoning → Clinical and biomedical applications

Abstract (arXiv)

Large Language Models (LLMs) are increasingly deployed in medical settings, yet their sensitivity to prompt formatting remains poorly characterized. We evaluate MedGemma (4B and 27B parameters) on MedMCQA (4,183 questions) and PubMedQA (1,000 questions) across a broad suite of robustness tests. Our experiments reveal several concerning findings. Chain-of-Thought (CoT) prompting decreases accuracy by 5.7% compared to direct answering. Few-shot examples degrade performance by 11.9% while increasing position bias from 0.14 to 0.47. Shuffling answer options causes the model to change predictions 59.1% of the time, with accuracy dropping up to 27.4 percentage points. Front-truncating context to 50% causes accuracy to plummet below the no-context baseline, yet back-truncation preserves 97% of full-context accuracy. We further show that cloze scoring (selecting the highest log-probability option token) achieves 51.8% (4B) and 64.5% (27B), surpassing all prompting strategies and revealing that models "know" more than their generated text shows. Permutation voting recovers 4 percentage points over single-ordering inference. These results demonstrate that prompt engineering techniques validated on general-purpose models do not transfer to domain-specific medical LLMs, and that reliable alternatives exist.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Clinical and biomedical applications, one of 17 directions in LLM reasoning. Builds on 7 of its 22 references that are in the map. First posted on arXiv 26 Mar 2026.

Builds on

7 of 22 references.

References · 7

  1. 2023Capabilities of GPT-4 on Medical Challenge Problemscited by 638
  2. 2023Towards Expert-Level Medical Question Answering with Large Language Modelscited by 384
  3. 2024BioMistral: A Collection of Open-Source Pretrained Large Language Models for Medical Domainscited by 301
  4. 2024Seven Failure Points When Engineering a Retrieval Augmented Generation Systemcited by 139
  5. 2024ClashEval: Quantifying the tug-of-war between an LLM's internal prior and external evidencecited by 113
  6. 2024Mind Your Step (by Step): Chain-of-Thought can Reduce Performance on Tasks where Thinking Makes Humans Worsecited by 109
  7. 2024Rationale-Guided Retrieval Augmented Generation for Medical Question Answeringcited by 41

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Is There Knowledge Left to Extract? Evidence of Fragility in Medically Fine-Tuned Vision-Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026To Reason or Not to: Selective Chain-of-Thought in Medical Question Answering
  2. 2026Better Eyes, Better Thoughts: Why Vision Chain-of-Thought Fails in Medicine
  3. 2026When Large Language Models Fail in Healthcare: Evaluating Sensitivity to Prompt Variations
  4. 2025Why Chain of Thought Fails in Clinical Text Understanding
  5. 2024The Limited Impact of Medical Adaptation of Large Language and Vision-Language Models
  6. 2025Instruction Tuning and CoT Prompting for Contextual Medical QA with LLMs
  7. 2025Dissecting Clinical Reasoning in Language Models: A Comparative Study of Prompts and Model Adaptation Strategies
  8. 2025On the Robustness of Answer Formats in Medical Reasoning Models