Real time risk prognostication via scalable hazard trees and forests

通过可扩展的危险树和森林进行实时风险预测

• 批准号: 10655749
• 负责人: Hemant Ishwaran
• 金额: $ 55.74万
• 依托单位: UNIVERSITY OF MIAMI SCHOOL OF MEDICINE
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-04-01 至 2027-03-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10655749
• 关键词: Affect; American; Benchmarking; Cessation of life; Code; Computer software; Data; Devices; Dialysis procedure; Disease; Electronic Health Record; Ensure; Event; Goals; Hazard Models; Heart; Heart Transplantation; Heart failure; Individual; Information Technology; Kidney Failure; Learning; Libraries; Machine Learning; Measurement; Methodology; Methods; Modeling; Modernization; Multicenter Studies; Organ failure; Patients; Performance; Procedures; Prognosis; Risk; Risk Assessment; Survival Analysis; System; Time; Trees; Update; Writing; adverse outcome; comorbidity; cost; data complexity; demographics; flexibility; forest; hazard; health data; hemodynamics; implantation; improved; innovation; left ventricular assist device; machine learning method; mechanical circulatory support; mortality; novel strategies; open source; personalized risk prediction; platform-independent; population health; portability; predictive modeling; prognostication; risk prediction; risk prediction model; statistics; success; tool; user friendly software; user-friendly; web page

Project Summary/Abstract Wearable sensing devices and Electronic Health Records (EHRs) are some examples of emerging information technologies expected to generate huge volumes of data recording individual’s health data over time. If properly utilized, these data provide a treasure trove of information for building real-time warning systems for adverse outcomes and to construct individualized risk prediction. To model the dynamic changes of covariate effects, time-varying survival models have emerged as a powerful approach. To deal with the size and complexity of data, with potential interactions among large number of variables, and interactions with time itself, we propose a state of the art machine learning approach using hazard trees and forests for estimating flexible hazard models with time-dependent covariates. Scalable and user friendly open source software implementing the methodology will be developed and made publicly available. The software will be applied to a rich, multicenter study of heart failure patients listed for heart transplantation to develop a state of the heart hazard risk prediction model.

项目总结/摘要 可穿戴传感设备和电子健康记录（EHR）是新兴的一些例子。 信息技术预计将产生大量的数据记录个人的健康数据， 时间如果利用得当，这些数据将为构建实时预警提供宝贵的信息 系统的不利后果，并建立个性化的风险预测。为了模拟动态 随着协变量效应的变化，时变生存模型已成为一种强有力的方法。处理 数据的规模和复杂性，以及大量变量之间的潜在相互作用， 随着时间的推移，我们提出了一种最先进的机器学习方法，使用危险树和森林， 估计具有时间依赖协变量的灵活风险模型。可扩展且用户友好的开源 将开发并公布实施该方法的软件。该软件将 应用于一项针对拟进行心脏移植的心力衰竭患者的丰富的多中心研究， 心脏病风险预测模型