Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2604.16383 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911602817630208 |
|---|---|
| author | DeLucia, Alexandra Huang, Heyuan Joshi, Sonal Yarmohammadi, Mahsa Hassoon, Ahmed Dredze, Mark |
| author_facet | DeLucia, Alexandra Huang, Heyuan Joshi, Sonal Yarmohammadi, Mahsa Hassoon, Ahmed Dredze, Mark |
| contents | LLM-as-a-Judge frameworks are increasingly trusted to automate evaluation in place of human experts, yet their reliability in high-stakes medical contexts remains unproven. We stress-test this assumption for detecting incomplete patient-facing medical responses, evaluating three rubric granularities (General-Likert, Analytical-Rubric, Dynamic-Checklist) and three backbone models across two clinician-annotated datasets, including HealthBench, the largest publicly available benchmark for medical response evaluation. LLM Judges discriminate complete from incomplete responses at and slightly above near chance (AUC $0.49$--$0.66$); at the threshold required to recall $90\%$ of incomplete responses, clinicians must still review the vast majority of the dataset, offering no triage utility. Even when model and clinician verdicts agree, they rarely cite the same explanation; and when they diverge, false positives stem from over-flagging non-essential gaps while false negatives reflect outright detection failures. These results reveal that LLM Judges and clinicians apply fundamentally different completeness standards; a finding that undermines their use as autonomous evaluators or triage filters in clinical settings. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_16383 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Same Verdict, Different Reasons: LLM-as-a-Judge and Clinician Disagreement on Medical Chatbot Completeness DeLucia, Alexandra Huang, Heyuan Joshi, Sonal Yarmohammadi, Mahsa Hassoon, Ahmed Dredze, Mark Computers and Society Artificial Intelligence LLM-as-a-Judge frameworks are increasingly trusted to automate evaluation in place of human experts, yet their reliability in high-stakes medical contexts remains unproven. We stress-test this assumption for detecting incomplete patient-facing medical responses, evaluating three rubric granularities (General-Likert, Analytical-Rubric, Dynamic-Checklist) and three backbone models across two clinician-annotated datasets, including HealthBench, the largest publicly available benchmark for medical response evaluation. LLM Judges discriminate complete from incomplete responses at and slightly above near chance (AUC $0.49$--$0.66$); at the threshold required to recall $90\%$ of incomplete responses, clinicians must still review the vast majority of the dataset, offering no triage utility. Even when model and clinician verdicts agree, they rarely cite the same explanation; and when they diverge, false positives stem from over-flagging non-essential gaps while false negatives reflect outright detection failures. These results reveal that LLM Judges and clinicians apply fundamentally different completeness standards; a finding that undermines their use as autonomous evaluators or triage filters in clinical settings. |
| title | Same Verdict, Different Reasons: LLM-as-a-Judge and Clinician Disagreement on Medical Chatbot Completeness |
| topic | Computers and Society Artificial Intelligence |
| url | https://arxiv.org/abs/2604.16383 |