arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2510.14286cs.LG
收藏

医疗时间序列数据中不良事件的稳定预测

Stable Prediction of Adverse Events in Medical Time-Series Data

University of Pennsylvania(宾夕法尼亚大学)

Mayank Keoliya, Seewon Choi, Rajeev Alur, Mayur Naik, Eric Wong

AI总结:

针对现有早期事件预测基准忽略风险评分稳定性且仅用表格数据评估的问题,提出CAREBench多模态基准与基于局部Lipschitz常数的稳定性度量,发现现有方法尤其LLMs难以兼顾准确性与稳定性。

摘要:

早期事件预测(EEP)系统持续估计患者的即时风险,以支持临床决策。为了获得床旁信任,风险轨迹必须准确且时间上稳定,仅随新的相关证据而改变。然而,当前的基准测试(a)忽略了风险评分的稳定性,并且(b)主要基于表格输入进行评估,导致轨迹行为未得到测试。为解决这一空白,我们引入了CAREBench,一个EEP基准测试,它使用多模态输入——表格电子健康记录(EHR)、心电图(ECG)波形和临床文本——来评估可部署性,并同时评估时间稳定性与预测准确性。我们提出了一种稳定性度量,该度量量化了每位患者风险的短期变异性,并基于局部Lipschitz常数对突然的振荡进行惩罚。CAREBench涵盖六项预测任务,如脓毒症发作,并比较了经典学习器、深度序列模型和零样本大语言模型(LLMs)。在所有任务中,现有方法,尤其是LLMs,难以同时优化准确性和稳定性,在高精度操作点上的召回率明显较差。这些结果凸显了需要产生与证据一致、稳定的轨迹的模型,以在连续监测环境中赢得临床医生的信任。(代码:https://github.com/SeewonChoi/CAREBench。)

英文摘要:

Early event prediction (EEP) systems continuously estimate a patient's imminent risk to support clinical decision-making. For bedside trust, risk trajectories must be accurate and temporally stable, shifting only with new, relevant evidence. However, current benchmarks (a) ignore stability of risk scores and (b) evaluate mainly on tabular inputs, leaving trajectory behavior untested. To address this gap, we introduce CAREBench, an EEP benchmark that evaluates deployability using multi-modal inputs-tabular EHR, ECG waveforms, and clinical text-and assesses temporal stability alongside predictive accuracy. We propose a stability metric that quantifies short-term variability in per-patient risk and penalizes abrupt oscillations based on local-Lipschitz constants. CAREBench spans six prediction tasks such as sepsis onset and compares classical learners, deep sequence models, and zero-shot LLMs. Across tasks, existing methods, especially LLMs, struggle to jointly optimize accuracy and stability, with notably poor recall at high-precision operating points. These results highlight the need for models that produce evidence-aligned, stable trajectories to earn clinician trust in continuous monitoring settings. (Code: https://github.com/SeewonChoi/CAREBench.)

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑