This website requires JavaScript.

Causal Graph Recovery for Sepsis-Associated Derangements via Interpretable Hawkes Networks

Song WeiYao XieChristopher S. JosefRishikesan Kamaleswaran
Jun 2021
摘要
Continuous, automated surveillance systems that incorporate machine learningmodels are becoming increasingly common in healthcare environments. Thesemodels can capture temporally dependent changes across multiple patientvariables and can enhance a clinician's situational awareness by providing anearly warning alarm of an impending adverse event such as sepsis. However, mostcommonly used methods, e.g., XGBoost, fail to provide an interpretablemechanism for understanding why a model produced a sepsis alarm at a giventime. The ``black box'' nature of many models is a severe limitation as itprevents clinicians from independently corroborating those physiologic featuresthat have contributed to the sepsis alarm. To overcome this limitation, wepropose a generalized linear model (GLM) approach to fit a Granger causal graphbased on the physiology of several major sepsis-associated derangements (SADs).We adopt a recently developed stochastic monotone variational inequality(VI)-based estimator coupled with forwarding feature selection to learn thegraph structure from both continuous and discrete-valued as well as regularlyand irregularly sampled time series. Theoretically, we develop a non-asymptoticupper bound on the estimation error for any monotone link function in the GLM.Using synthetic and real-data examples, we demonstrate that the proposed methodenjoys result interpretability while achieving comparable performance topopular methods such as XGBoost.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答