Decomposing Physician Disagreement in HealthBench

• URL: http://arxiv.org/abs/2602.22758v1
• Date: Thu, 26 Feb 2026 08:47:42 GMT
• Title: Decomposing Physician Disagreement in HealthBench
• Abstract summary: We decompose physician disagreement in the HealthBench medical AI evaluation dataset to understand where variance resides and what observable features can explain it.<n>The dominant 81.8% case-level residual is not reduced by HealthBench's metadata labels.<n>Disagreement follows an inverted-U with completion quality, confirming physicians agree on clearly good or bad outputs but split on borderline cases.
• Score: 0.0687531213383208
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We decompose physician disagreement in the HealthBench medical AI evaluation dataset to understand where variance resides and what observable features can explain it. Rubric identity accounts for 15.8% of met/not-met label variance but only 3.6-6.9% of disagreement variance; physician identity accounts for just 2.4%. The dominant 81.8% case-level residual is not reduced by HealthBench's metadata labels (z = -0.22, p = 0.83), normative rubric language (pseudo R^2 = 1.2%), medical specialty (0/300 Tukey pairs significant), surface-feature triage (AUC = 0.58), or embeddings (AUC = 0.485). Disagreement follows an inverted-U with completion quality (AUC = 0.689), confirming physicians agree on clearly good or bad outputs but split on borderline cases. Physician-validated uncertainty categories reveal that reducible uncertainty (missing context, ambiguous phrasing) more than doubles disagreement odds (OR = 2.55, p < 10^(-24)), while irreducible uncertainty (genuine medical ambiguity) has no effect (OR = 1.01, p = 0.90), though even the former explains only ~3% of total variance. The agreement ceiling in medical AI evaluation is thus largely structural, but the reducible/irreducible dissociation suggests that closing information gaps in evaluation scenarios could lower disagreement where inherent clinical ambiguity does not, pointing toward actionable evaluation design improvements.

Related papers

• Counterfactual Fairness Audits of Multi-Step Clinical LLM Agents Require a Measured Per-Action Instability Floor [3.0704370207786265]
  We show that a counterfactual fairness estimate reported without a per-action floor beside it cannot be read as evidence of disparity.<n>The measurements were taken with FairMedAgent, an evaluation harness for disparity in the actions of clinical adjudication agents.
  arXiv  Detail & Related papers  (2026-09-02T23:38:17Z)
• Bayesian uncertainty estimation improves clinical decision making in medical AI agents [3.251669068529209]
  We show that Monte Carlo dropout provides an uncertainty signal that tracks generalisation across training-set scales and flags confident yet error-prone predictions.<n>In a controlled 2x2 factorial experiment, a clinical-decision-support agent exploited this uncertainty only when it was delivered as a binary error-risk flag rather than as raw scores.
  arXiv  Detail & Related papers  (2026-07-22T11:54:23Z)
• Measuring Epistemic Resilience of LLMs Under Misleading Medical Context [47.20527783144983]
  Large language models (LLMs) now reach expert-level scores on medical licensing exams.<n>We show this assumption is fragile: when misleading context is injected into questions that LLMs originally answer correctly, they abandon the correct answer.<n>We call the ability to maintain correct judgment under adversarial context epistemic resilience, and introduce MedMisBench to measure it.
  arXiv  Detail & Related papers  (2026-06-10T16:27:26Z)
• ClinicalBench: Stress-Testing Assertion-Aware Retrieval for Cross-Admission Clinical QA on MIMIC-IV [0.0]
  Reasoning benchmarks measure clinical performance on clean inputs.<n>We evaluate the step before reasoning: retrieval over real EHR notes, where negation, temporality, and family-versus-patient attribution can flip a correct answer to a wrong one.<n>EpiKG carries assertion label and a temporality tag with every fact in a patient knowledge graph, then routes retrieval by question intent.
  arXiv  Detail & Related papers  (2026-05-11T18:47:52Z)
• From Black Box to Glass Box: Cross-Model ASR Disagreement to Prioto Review in Ambient AI Scribe Documentation [43.148402136307716]
  Cross-model disagreement among heterogeneous ASR systems can act as a reference-free uncertainty signal.<n>We transcribed 50 publicly available medical audio clips with eight ASR systems spanning commercial APIs and open-source engines.<n>Low-agreement regions were enriched for content disagreements, with the content fraction increasing from 53.9% to 73.9% across quintiles of high-risk mass.
  arXiv  Detail & Related papers  (2026-03-02T13:02:13Z)
• PanCanBench: A Comprehensive Benchmark for Evaluating Large Language Models in Pancreatic Oncology [48.732366302949515]
  Large language models (LLMs) have achieved expert-level performance on standardized examinations, yet multiple-choice accuracy poorly reflects real-world clinical utility and safety.<n>We developed a human-in-the-loop pipeline to create expert rubrics for de-identified patient questions.<n>We evaluated 22 proprietary and open-source LLMs using an LLM-as-a-judge framework, measuring clinical completeness, factual accuracy, and web-search integration.
  arXiv  Detail & Related papers  (2026-03-02T00:50:39Z)
• Mechanistically Guided LoRA Improves Paraphrase Consistency in Medical Vision-Language Models [2.064612766965483]
  Medical Vision-Language Models can give different yes or no answers to rephrasings of the same clinical question.<n>We study this in MedGemma-4B using PSF-Med Sadanandan and Behzadan (2025), which provides paraphrase pairs for systematic consistency evaluation on medical VQA.<n>We fine-tune Low-Rank Adaptation (LoRA) adapters with a combined loss that balances paraphrase consistency with answer accuracy.
  arXiv  Detail & Related papers  (2026-02-24T23:30:50Z)
• Explainable Admission-Level Predictive Modeling for Prolonged Hospital Stay in Elderly Populations: Challenges in Low- and Middle-Income Countries [65.4286079244589]
  Prolonged length of stay (pLoS) is a significant factor associated with the risk of adverse in-hospital events.<n>We develop and explain a predictive model for pLos using admission-level patient and hospital administrative data.
  arXiv  Detail & Related papers  (2026-01-07T23:35:24Z)
• Decomposing Uncertainty in Probabilistic Knowledge Graph Embeddings: Why Entity Variance Is Not Enough [0.0]
  Probabilistic knowledge graph embeddings represent entities as distributions, using learned variances to quantify uncertainty.<n>These variances are relation-agnostic, meaning an entity receives identical uncertainty regardless of relational context.<n>We prove an impossibility result: any uncertainty estimator using only entity-level statistics independent of relation context achieves near-random OOD detection on novel contexts.
  arXiv  Detail & Related papers  (2025-12-26T12:38:43Z)
• Hide-and-Seek Attribution: Weakly Supervised Segmentation of Vertebral Metastases in CT [68.09387763135236]
  We introduce a weakly supervised method trained solely on vertebra-level healthy/malignant labels, without any lesion masks.<n>We achieve strong blastic/lytic performance despite no mask supervision.
  arXiv  Detail & Related papers  (2025-12-07T14:03:28Z)
• Prostate-VarBench: A Benchmark with Interpretable TabNet Framework for Prostate Cancer Variant Classification [14.190646211771073]
  Variants of Uncertain Significance (VUS) limit the clinical utility of prostate cancer genomics.<n>Prostate-VarBench is a curated pipeline for creating prostate-specific benchmarks.
  arXiv  Detail & Related papers  (2025-11-12T06:13:50Z)
• MEDEQUALQA: Evaluating Biases in LLMs with Counterfactual Reasoning [7.167933033102407]
  We introduce MEDEQUALQA, a counterfactual benchmark that perturbs only patient pronouns while holding critical symptoms and conditions constant.<n>We evaluate a GPT-4.1 model and compute Semantic Textual Similarity (STS) between reasoning traces to measure stability across pronoun variants.<n>Our results show overall high similarity (mean STS >0.80), but reveal consistent localized divergences in cited risk factors, guideline anchors, and differential ordering.
  arXiv  Detail & Related papers  (2025-10-09T22:12:58Z)
• Medical Hallucinations in Foundation Models and Their Impact on Healthcare [71.15392179084428]
  Hallucinations in foundation models arise from autoregressive training objectives.<n>Top-performing models exceeded 97% accuracy when augmented with chain-of-thought prompting.
  arXiv  Detail & Related papers  (2025-02-26T02:30:44Z)
• Towards Self-Supervised Covariance Estimation in Deep Heteroscedastic Regression [102.24287051757469]
  We study self-supervised covariance estimation in deep heteroscedastic regression.<n>We derive an upper bound on the 2-Wasserstein distance between normal distributions.<n>Experiments over a wide range of synthetic and real datasets demonstrate that the proposed 2-Wasserstein bound coupled with pseudo label annotations results in a computationally cheaper yet accurate deep heteroscedastic regression.
  arXiv  Detail & Related papers  (2025-02-14T22:37:11Z)
• Evaluating AI systems under uncertain ground truth: a case study in dermatology [43.8328264420381]
  We show that ignoring uncertainty leads to overly optimistic estimates of model performance.<n>In skin condition classification, we find that a large portion of the dataset exhibits significant ground truth uncertainty.
  arXiv  Detail & Related papers  (2023-07-05T10:33:45Z)
• Towards Reliable Medical Image Segmentation by Modeling Evidential Calibrated Uncertainty [57.023423137202485]
  Concerns regarding the reliability of medical image segmentation persist among clinicians.<n>We introduce DEviS, an easily implementable foundational model that seamlessly integrates into various medical image segmentation networks.<n>By leveraging subjective logic theory, we explicitly model probability and uncertainty for medical image segmentation.
  arXiv  Detail & Related papers  (2023-01-01T05:02:46Z)
• Treatment Effect Risk: Bounds and Inference [58.442274475425144]
  Since the average treatment effect measures the change in social welfare, even if positive, there is a risk of negative effect on, say, some 10% of the population. In this paper we consider how to nonetheless assess this important risk measure, formalized as the conditional value at risk (CVaR) of the ITE distribution. Some bounds can also be interpreted as summarizing a complex CATE function into a single metric and are of interest independently of being a bound.
  arXiv  Detail & Related papers  (2022-01-15T17:21:26Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.