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

arXiv: https://arxiv.org/abs/2603.25960 · HTML: https://mlsanity.com/paper/2603.25960/

## Key facts

- Cited by 1 papers 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.

## Placed in

- LLM reasoning / [Clinical and biomedical applications](https://mlsanity.com/lr/clinical-and-biomedical-applications/)

## Builds on (most cited first, 7 of 7)

- [Capabilities of GPT-4 on Medical Challenge Problems](https://mlsanity.com/paper/2303.13375/) (2023)
- [Towards Expert-Level Medical Question Answering with Large Language Models](https://mlsanity.com/paper/2305.09617/) (2023)
- [BioMistral: A Collection of Open-Source Pretrained Large Language Models for Medical Domains](https://mlsanity.com/paper/2402.10373/) (2024)
- [Seven Failure Points When Engineering a Retrieval Augmented Generation System](https://mlsanity.com/paper/2401.05856/) (2024)
- [ClashEval: Quantifying the tug-of-war between an LLM's internal prior and external evidence](https://mlsanity.com/paper/2404.10198/) (2024)
- [Mind Your Step (by Step): Chain-of-Thought can Reduce Performance on Tasks where Thinking Makes Humans Worse](https://mlsanity.com/paper/2410.21333/) (2024)
- [Rationale-Guided Retrieval Augmented Generation for Medical Question Answering](https://mlsanity.com/paper/2411.00300/) (2024)

## Built on by (most cited first, 1 of 1)

- [Is There Knowledge Left to Extract? Evidence of Fragility in Medically Fine-Tuned Vision-Language Models](https://mlsanity.com/paper/2604.09841/) (2026)

## 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.