FAI: An Interpretable AI Framework for Care of Critically Ill Patients Involving Matching and Decision Trees

这个项目介绍了一个可解释的，以患者为中心的因果推理和住院患者护理策略设计的框架。这一框架源于一个具有挑战性的问题，即如何治疗有癫痫(亚临床癫痫)风险的危重患者，这种癫痫可能会严重损害患者的大脑。在这一高风险的人工智能应用中，数据是复杂的，包括嘈杂的时间序列、病史和人口统计信息。其目标是产生可解释的因果估计和政策决策，使医生能够准确地了解数据是如何组合的，从而能够更好地进行故障排除、不确定性量化，并最终实现信任。该项目框架的核心由新颖而复杂的匹配技术组成，该技术将数据集中每个接受治疗的患者与其他(类似)未接受治疗的患者进行匹配。配对模拟随机对照试验，允许根据匹配组的结果评估每个患者的治疗效果。该框架的第二个重要元素涉及可解释的策略设计，其中将使用稀疏决策树来确定应接受类似治疗的可解释的个体子组。该项目中开发的匹配技术将属于“几乎完全匹配”(AME)技术的新家族。AME技术在训练集上使用机器学习来确定如何构建高质量的匹配组。在应用AME技术分析危重患者的癫痫风险和治疗时，该项目解决了两个主要挑战：如何纳入药物吸收的机制模型，以及如何进行不确定性量化。重要的是，该项目还解决了非专家使用的几种格式的AME代码的发布问题。该项目的策略设计方面涉及稀疏决策树的优化。该项目涉及机器学习、因果推理、数据库和神经学方面的专家之间的密切合作，目的是在无法进行实验的高风险医院环境中改善患者护理，而评估因果影响的唯一方法是通过分析观察数据。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project introduces a framework for interpretable, patient-centered causal inference and policy design for in-hospital patient care. This framework arose from a challenging problem, which is how to treat critically ill patients who are at risk for seizures (subclinical seizures) that can severely damage a patient's brain. In this high-stakes application of artificial intelligence, the data are complex, including noisy time-series, medical history, and demographic information. The goal is to produce interpretable causal estimates and policy decisions, allowing doctors to understand exactly how data were combined, permitting better troubleshooting, uncertainty quantification, and ultimately, trust. The core of the project's framework consists of novel and sophisticated matching techniques, which match each treated patient in the dataset with other (similar) patients who were not treated. Matching emulates a randomized controlled trial, allowing the effect of the treatment to be estimated for each patient, based on the outcomes from their matched group. A second important element of the framework involves interpretable policy design, where sparse decision trees will be used to identify interpretable subgroups of individuals who should receive similar treatments.The matching techniques developed in this project will be within the new family of "almost-matching-exactly" (AME) techniques. AME techniques use machine learning on a training set to determine how to construct high-quality matched groups. In applying AME techniques to analyze seizure risk and treatment for critically ill patients, there are two major challenges that this project addresses: how to incorporate mechanistic models for drug absorption, and how to perform uncertainty quantification. Importantly, the project also addresses the release of AME code in several formats to be used by non-experts. The policy design aspect of the project involves the optimization of sparse decision trees. This project involves a close collaboration between experts in machine learning, causal inference, databases, and neurology, with the goal to improve patient care in high-stakes hospital settings where experiments cannot be conducted, and the only way to assess causal effects is through the analysis of observational data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.