arXiv cs.AI· Chaitali Deb Purkayastha, Bharath Kumar Bolla, Vishnu Surya Reddy Nandi·· 3 天前AI 评分29
临床 LLM 评估不止准确率：反事实脆弱性与人口统计学偏差研究
Beyond Accuracy: Counterfactual Fragility and Demographic Bias in Clinical Evaluation of LLMs
AI 导读
研究用两项自动化扰动测试评估六个 LLM 在 150 道 MedQA USMLE 题上的表现：900 次反事实有效性（CFV）尝试中仅 228 次（25.3%）有效，5,400 次人口统计学对比中 1,097 个答案被改变（20.3%），并识别出 3,128 个刻板印象证据标记。
正文 · 原文
Abstract:Clinical LLM evaluation often emphasizes answer accuracy; however, accuracy alone does not test counterfactual consistency or demographic robustness. We evaluated six LLMs on 150 MedQA USMLE questions using two automated perturbation tests to assess their performance. The counterfactual validity (CFV) test asked each model to make a minimal, plausible clinical change that would make a different answer correct. The demographic robustness test added six demographic prefixes to the same vignette and compared the answers and explanations with a no demographic baseline. Of the 900 CFV attempts, 228 (25.3 %) were valid and 672 were invalid. Across 5,400 demographic comparisons, 1,097 answers were changed (20.3%). Automated judging identified 3,128 stereotype evidence flags, including 1,932 in the broad Other category. MedGemma 27B achieved the highest accuracy (87.1%) and CFV (63.3%), lowest answer change rate (16.0%), and low mean Explanation Demographic Dissonance (EDD) score (0.169). However, its accuracy still exceeded its CFV, indicating that correct answers do not guarantee reliable performance on the counterfactual validity task. OpenBioLLM had the highest answer change rate and EDD, whereas GLM had the highest stereotype flag rate. These findings show that accuracy, CFV, answer stability, EDD, and stereotype evidence capture different evaluation aspects. Because all judgments were automated and no clinician validation was available, the results support safety screening but do not establish clinical deployability of the model.
| Comments: | Accepted into AIKP 2026 Conference |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.32807 [cs.AI] |
| (or arXiv:2609.32807v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.32807 arXiv-issued DOI via DataCite |
Submission history
From: Vishnu Surya Reddy Nandi [view email]
[v1]
Sat, 26 Sep 2026 17:28:46 UTC (679 KB)
[v2]
Sat, 10 Oct 2026 19:28:45 UTC (679 KB)
来源：arXiv cs.AI · arxiv.org