开放权重 LLM 的临床风险分类错位与成本不对称性研究

View PDF HTML (experimental)

Abstract:How large language models (LLMs) integrate patient risk with clinical cost tradeoffs remains poorly understood. We investigated how four open-weight LLMs (Qwen-2.5-7B/32B and Llama-3.1-8B/70B) internally represent cost tradeoffs, how these representations relate to clinical predictions, and whether decisions shift as predicted by the specified cost direction and magnitude. Using a public diabetes dataset, we varied 11 false-negative (FN) to false-positive (FP) cost ratios across three phrasings and examined representations and behavioral outputs. Patient risk was linearly recoverable on par with conventional classifiers (AUC $\approx 0.83$), and cost direction was recoverable in every model. However, representational shifts in cost direction tracked output changes only in the two larger models, and responses to cost magnitude were predominantly direction-agnostic. Only 2 of 12 model-phrasings showed both opposing responses to increasing FN versus FP costs and cost-correct ordering. Representationally, a direction fitted on one cost side did not invert when transferred to the other, as expected under mirror-symmetric encoding. These findings suggest that LLMs encode risk and cost information but do not reliably integrate them into cost-correct decisions. Clinical evaluations should therefore include tradeoff tests, phrasing sensitivity, and default operating points alongside predictive performance.

Comments:	Submitted to ML4H 2026
Subjects:	Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2609.23999 [cs.LG]
	(or arXiv:2609.23999v3 [cs.LG] for this version)
	https://doi.org/10.48550/arXiv.2609.23999 arXiv-issued DOI via DataCite

Submission history

From: Star S.D. Liu [view email]
[v1] Mon, 21 Sep 2026 02:02:10 UTC (4,937 KB)
[v2] Mon, 28 Sep 2026 03:06:45 UTC (4,937 KB)
[v3] Tue, 13 Oct 2026 04:26:41 UTC (5,622 KB)