Automatic Evaluation of Healthcare LLMs Beyond Question-Answering

2025-02-10

AI Summary

Medical large language models (LLMs) have long suffered from a fragmented evaluation paradigm—open-ended assessment emphasizes paraphrasing and lacks verifiability, while closed-ended assessment prioritizes factual accuracy but neglects generative fluency; their correlation and coverage gaps remain poorly characterized. To address this, we propose a multidimensional medical LLM evaluation framework: (1) a novel open-generation metric, Relaxed Perplexity, which relaxes strict token-order constraints of conventional perplexity; (2) CareQA, a hybrid benchmark jointly optimizing factual correctness and expressive richness; and (3) the first empirical characterization of coupling relationships and coverage deficiencies across multi-axis evaluation metrics in medical LLM assessment. Experiments demonstrate that Relaxed Perplexity achieves a 23.6% improvement in inter-physician rating consistency over baseline metrics, and CareQA significantly enhances the joint evaluation of factual grounding and generative quality.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Current Large Language Models (LLMs) benchmarks are often based on open-ended or close-ended QA evaluations, avoiding the requirement of human labor. Close-ended measurements evaluate the factuality of responses but lack expressiveness. Open-ended capture the model's capacity to produce discourse responses but are harder to assess for correctness. These two approaches are commonly used, either independently or together, though their relationship remains poorly understood. This work is focused on the healthcare domain, where both factuality and discourse matter greatly. It introduces a comprehensive, multi-axis suite for healthcare LLM evaluation, exploring correlations between open and close benchmarks and metrics. Findings include blind spots and overlaps in current methodologies. As an updated sanity check, we release a new medical benchmark--CareQA--, with both open and closed variants. Finally, we propose a novel metric for open-ended evaluations --Relaxed Perplexity-- to mitigate the identified limitations.

Problem

• Evaluate healthcare LLMs beyond QA benchmarks
• Identify gaps in current LLM evaluation methods
• Propose new metric for open-ended assessment

Innovation

• Multi-axis evaluation suite
• CareQA medical benchmark
• Relaxed Perplexity metric