![[2602.22771] ClinDet-Bench: Beyond Abstention, Evaluating Judgment Determinability of LLMs in Clinical Decision-Making](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22771] ClinDet-Bench: Beyond Abstention, Evaluating Judgment Determinability of LLMs in Clinical Decision-Making
Summary
The paper presents ClinDet-Bench, a benchmark for evaluating the judgment determinability of large language models (LLMs) in clinical decision-making, highlighting their limitations in handling incomplete information.
Why It Matters
As LLMs are increasingly used in healthcare, understanding their ability to make safe and accurate decisions under uncertainty is crucial. ClinDet-Bench addresses the gap in existing benchmarks, emphasizing the need for reliable evaluation tools to ensure patient safety in clinical settings.
Key Takeaways
- ClinDet-Bench evaluates LLMs' ability to determine judgment under incomplete information.
- Current LLMs struggle with identifying determinability, leading to premature conclusions or excessive abstention.
- The benchmark is essential for assessing LLM safety in high-stakes environments like medicine.
Computer Science > Artificial Intelligence arXiv:2602.22771 (cs) [Submitted on 26 Feb 2026] Title:ClinDet-Bench: Beyond Abstention, Evaluating Judgment Determinability of LLMs in Clinical Decision-Making Authors:Yusuke Watanabe, Yohei Kobashi, Takeshi Kojima, Yusuke Iwasawa, Yasushi Okuno, Yutaka Matsuo View a PDF of the paper titled ClinDet-Bench: Beyond Abstention, Evaluating Judgment Determinability of LLMs in Clinical Decision-Making, by Yusuke Watanabe and 5 other authors View PDF HTML (experimental) 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 e...
