ClinDet-Bench: Beyond Abstention, Evaluating Judgment Determinability of LLMs in Clinical Decision-Making

TL;DR

ClinDet-Bench is a new benchmark for assessing whether large language models can recognize when they have enough information to make clinical decisions, addressing a critical safety aspect in high-stakes medical applications.

Contribution

This paper introduces ClinDet-Bench, a benchmark that evaluates LLMs' ability to determine information sufficiency in clinical decision-making scenarios.

Findings

LLMs often fail to identify determinability, leading to premature judgments or unnecessary abstention.

Existing benchmarks do not adequately assess LLM safety in clinical contexts.

ClinDet-Bench offers a framework for evaluating and improving LLM judgment reliability.

Abstract

Clinical decisions are often required under incomplete information. Clinical experts must identify whether available information is sufficient for judgment, as both premature conclusion and unnecessary abstention can compromise patient safety. To evaluate this capability of large language models (LLMs), we developed ClinDet-Bench, a benchmark based on clinical scoring systems that decomposes incomplete-information scenarios into determinable and undeterminable conditions. Identifying determinability requires considering all hypotheses about missing information, including unlikely ones, and verifying whether the conclusion holds across them. We find that recent LLMs fail to identify determinability under incomplete information, producing both premature judgments and excessive abstention, despite correctly explaining the underlying scoring knowledge and performing well under complete information. These findings suggest that existing benchmarks are insufficient to evaluate the safety of LLMs in clinical settings. ClinDet-Bench provides a framework for evaluating determinability recognition, leading to appropriate abstention, with potential applicability to medicine and other high-stakes domains, and is publicly available.