Auditing Multimodal LLM Raters: Central Tendency Bias in Clinical Ordinal Scoring

Jiaqing Zhang, Sandeep Elluri, Bhanu Cherukuvada, Yonah Joffe, Jessica Sena, Miguel Contreras, Scott Siegel, Subhash Nerella, Catherine Price, Parisa Rashidi

arXiv:2605.16386 · 2026-05-19 공개 · arXiv · PDF

llm-as-a-judge calibration vision-transformers multimodal-llms clinical-ai clinical-evaluation central-tendency-bias clock-drawing-test

Abstract

Multimodal large language models (LLMs) are increasingly explored as automated evaluators in clinical settings, yet their scoring behavior on ordinal clinical scales remains poorly understood. We benchmark three frontier LLM families against supervised deep learning models for scoring Clock Drawing Test (CDT) images on two public datasets using the Shulman rubric. While fully fine-tuned Vision Transformers achieve the best calibration (MAE 0.52, within-1 accuracy 91%), zero-shot LLMs remain competitive on tolerance-based agreement (GPT-5 MAE 0.67, within-1 accuracy 92%) despite higher absolute error. However, per-score analysis reveals that all three LLM families exhibit a pronounced central tendency effect (systematic endpoint compression): predictions are systematically compressed toward the middle of the scale, with over-prediction at the low end (score 0 to 1) and under-prediction at the high end (score 5 to 4). This effect disproportionately affects the clinically critical extremes where accurate scoring most impacts screening decisions for cognitive impairment. Targeted ablations show that neither few-shot exemplars spanning the full score range nor removing clinical terminology from the prompt eliminates the effect. Our findings extend the LLM-as-a-judge bias literature from NLP evaluation to clinical assessment, and highlight the need for calibration-aware evaluation and post-hoc calibration before deploying LLM-based raters in high-stakes screening workflows.

한국어 요약

📋 한 줄 요약

**[멀티모달 LLM / 임상 평가]** 시계 그리기 검사(CDT) 이미지에 대한 정수 등급 점수에서 프론티어 멀티모달 LLM들이 일관된 중심 경향(central tendency) 편향을 보임을 정량 분석.

🎯 핵심 기여도

💡 핵심 아이디어

LLM-as-a-judge 편향 연구는 NLP 평가에 집중되어 있으나 임상 정수 등급에서도 별개의 강한 편향(central tendency)이 작동하며, 이는 정확한 채점이 가장 중요한 끝점에서 가장 큰 오류를 만든다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 임상 평가 워크플로에 LLM 채점자를 배치할 때 calibration-aware 평가와 post-hoc calibration이 필수임을 보임. **한계**: 단일 임상 척도(CDT, Shulman) 기반 연구로 다른 정수 등급 척도로의 일반화는 추가 검증 필요.

🚀 실용적 활용