It is suggested that trustworthy clinical decision support should be evaluated by both average correctness and stability across medically equivalent patient narratives.
Abstract
Medical large language models are often judged by how many clinical questions they answer correctly. That view is useful, but it misses a practical risk. A model may know the right answer and still change its response when the same case is written in a different patient voice. This paper evaluates that risk as SDoH aware narrative anchoring bias. We use NarrativeShield SDoH MedQA, a counterfactual medical question answering dataset in which each case appears in persona based narratives while the answer key remains fixed. The dataset is reshaped from wide format into case grouped persona rows. We evaluate three open source instruction tuned LLMs from the Qwen2.5 family: 1.5B, 3B, and 7B. The final experiment uses 300 clinical cases and produces 8,100 model responses across three prompting conditions. We report persona level accuracy, counterfactual consistency, correct consistency, and narrative sensitivity error. Qwen2.5 7B achieves the best accuracy at 56.33 percent and the best correct consistency at 40.33 percent. Paired McNemar exact tests show significant accuracy gains for 7B over 3B in all prompt settings. Even so, narrative sensitivity remains, with the lowest error still at 31.67 percent. These results suggest that trustworthy clinical decision support should be evaluated by both average correctness and stability across medically equivalent patient narratives.
An evaluation framework based on the MedMCQA dataset consisting of two complementary uncertainty settings is proposed, which introduces linguistic uncertainty cues through prompt modifications to simulate ambiguous clinical contexts and observes significant variation across models in their ability to abstain when the correct answer is unavailable.
Maryam Tahermazandarani, Adnan Mahmood, Fahmida Islam et al.· International Conference on...· 0 citations
Large language models used for clinical diagnostic reasoning are sensitive to sociolinguistic register, not just clinical content. We term this failure mode Narrative Anchoring: identical clinical facts expressed in different registers cause diagnostic outputs to diverge. Unlike prior demographic-bias work, which manipulates explicit identity tokens such as race or income, our benchmark isolates register as the sole channel of variation, with no demographic marker present in any form. We construct a dataset of 1,000 USMLE clinical vignettes, each rewritten into three sociolinguistically distinct personas under an independently audited fact-preservation guarantee, verified by a separate model that never sees the generation prompt. Across seven language models spanning three architecture families and scales, Narrative Anchoring is statistically significant under direct prompting in every model tested, with a Narrative Anchoring Gap of 0.064 to 0.151. Chain-of-thought reasoning and explicit debiasing instructions reduce the bias only partially, and their apparent gains are frequently confounded by accuracy collapse. We introduce NarrativeShield, a three-agent pipeline that structurally extracts and verifies clinical facts before diagnostic reasoning begins, reducing the Narrative Anchoring Gap to near-zero ($-0.004$ to $0.037$) and achieving the lowest rate of severely unstable decisions (DSS $<$ 0.8) of any method across all models, at a modest and mechanistically expected accuracy cost for most models. A stress test using a non-instruction-tuned base model shows that executing a debiasing intervention at all is gated by zero-shot instruction-following ability, not prompt content alone. We release our dataset, human-validated for fact preservation, as a standalone resource for studying register-based clinical bias.
Prabhjot Singh, Pritam Deka, V. Chennareddy· 0 citations
Background: LLM judges increasingly score whether clinical language models give overconfident answers under incomplete evidence, yet whether a measured"safety gain"reflects real behavior change or the judge's calibration is unresolved. Using a structured evidence-sufficiency prompt as a test case, we asked whether it reduces unsafe overconfident answers, how far that effect depends on the scoring judge, and what it costs in helpfulness. Methods: In a retrospective public-data benchmark (Real-POCQi, HealthBench, MedRBench), four models (GPT-5.5, Claude Opus 4.8, Gemini 3.5 Flash, Grok 4.3) answered a fully paired common panel (1,200 cells) with a standard prompt and the wrapper. The pre-specified endpoint was the paired reduction in unsafe overconfidence scored by the primary judge (GPT-5.4-nano); secondary analyses added a different-family judge (Claude Sonnet 5), a correctness judge, matched scaffold controls, and a blinded three-clinician review. Results: Unsafe overconfidence fell from 49.3% to 24.7%, a paired reduction of 24.7 points (95% CI 21.8-27.7; p<0.001), robust in direction across models and paraphrases. Magnitude was judge-dependent: Sonnet agreed on direction but nearly halved the effect (+13.1 points), with one-directional disagreement. Blinded clinicians characterized the primary judge as a high-sensitivity (1.00), low-specificity (0.55) screen, not a calibrated rate. The gain carried a model-specific helpfulness cost (correct diagnosis 80.3% to 50.3%): near-free for GPT-5.5, near-total for Gemini (-58 points). Matched scaffold controls showed genuine behavior change, not judge circularity. Conclusions: LLM-judged clinical safety effects should be reported as directional and relative, anchored to human review and evaluated jointly with helpfulness, not as calibrated absolute rates. This does not establish clinical deployment readiness.
Medical LLMs are often evaluated by whether they select the correct diagnosis, but diagnostic accuracy alone does not show whether the model used the case evidence appropriately. We present a behavioral audit of evidence use in medical diagnosis. For each case, we decompose patient information into evidence units, score candidate diagnoses under controlled evidence subsets, and mine low-order interactions in diagnostic margins. Because medical evidence is diagnosis-relative, the audit separates interaction discovery from failure assignment: large or negative interactions can reflect plausible differential diagnosis, while suspicious interactions require robustness checks and clinical review. We evaluate five open-weight LLMs on DDXPlus, CupCase, and MedCase. Across datasets, faithful support and differential conflict or cancellation account for most interaction strength, showing that many evidence interactions are clinically plausible rather than failures. In a DDXPlus-focused blinded five-reviewer 130-item enriched review sample, invalid or shortcut-like cases concentrate in negated or absent findings and clinically local evidence. These results show that accuracy can hide candidate evidence-use failures and motivate role-aware audits for medical LLM evaluation.
Five widely used prompting strategies across five influential LLMs in the latest medical bias benchmark reveal substantial heterogeneity in both effectiveness and overhead across models, with no strategy proving universally effective and some even exacerbating bias.
Ying Xiao, Zhenpeng Chen, Jie M. Zhang· Philosophical transactions....· 1 citation
Clinical decisions, such as diagnosing conditions, prescribing medications, and recommending procedures, are rarely made with absolute certainty. Instead, they reflect probabilistic judgments shaped by evolving patient information and incomplete evidence. However, current EHR systems and knowledge graphs encode such decisions as deterministic triples, lacking a mechanism to represent the subjective confidence inherent in clinical reasoning. We present JudgEHR, a framework for clinical decision confidence estimation that leverages large language models (LLMs) to perform cohort-based collective inference over structured patient records by representing clinical events as knowledge graph triples and integrating them into LLM prompts. JudgEHR groups related clinical concepts into cohorts using LLM-driven relational inference, and then jointly evaluates the plausibility of all clinical decision triples within each cohort by considering patient visit history and background medical knowledge. We apply our method to the MIMIC-III dataset. Our statistical analysis shows that JudgEHR generates semantically consistent confidence scores, with similar concepts receiving closer values, whereas dissimilar replacements yield large confidence differences. Experiments on the MIMIC-III dataset show that incorporating the confidence into a zero-shot LLM-based pipeline improves relative AUROC by ${1 4. 6 \%}$ and AUPRC by 21.8% on the mortality prediction task.
Kexuan Xin, Guillaume Pelat, Jonathan Vitale et al.· International Conference on...· 0 citations
