Real-time Prediction of Adverse Outcomes After Surgery

这个K23应用程序的目标是为比沙拉博士提供必要的研究经验和 是时候将自己确立为主要研究人员，专注于设计和实现机器学习 (ML)和围手术期的人工智能。此申请表中的职业发展活动 包括早期强化ML课程和侧重于改进模型开发和因果关系的统计 推理技术。然后课程重点是临床试验培训，勇气，负责任的行为 研究，并最终在一门研究实现科学和算法人-机器人交互的课程中达到顶峰。 对此培训的补充是与专家一起提供的特定于项目的教程，以改进本 应用于围手术期急性肾损伤(AKI)的实时预测和风险描述 AKI围手术期景观。为了实现这些目标，比沙拉博士组建了一个专家团队， 数据科学、AKI、ML和统计学领域的导师。他的主要导师阿图尔·巴特博士是 数据科学和ML，培养了近百名博士后、本科生和研究生， 和员工。刘慧琳博士是AKI领域的思想领袖，有一项专注于AKI的积极研究计划 和重症监护临床试验。她曾指导过许多初级教职员工，包括之前的NIH K23获奖者。 Romain Pirracchio博士是生物统计学和急性护理ML方面的专家。他与伯克利分校有合作 FDA和该领域的100多家出版物。这三位导师和令人印象深刻的顾问团队将 指导比沙拉博士完成下面描述的项目，并成长为一名独立的调查员。 最近出版的关于ML在医学上的文献激增，研究表明 当ML增强了提供者的专业知识时，患者护理就会得到改善。不幸的是，实现已发布的ML 为临床护理提供信息的模型并不是微不足道的，因为存在许多障碍。此应用程序侧重于探索和 通过在围手术期环境中实施特定的模式来克服这些障碍。比沙拉医生已经 开发了新的ML可视化技术，允许改善提供商和 模型，向这些供应商提供预测和建议。这项技术还允许 改进了对模型的定期监测和解释，以确保持续的准确性和可靠性。他会的 在他开发的模型的基础上，应用这项新技术来实时预测围手术期AKI。 术后AKI是一个主要的公共健康问题，影响高达47%的患者，并一直与 不良后果，包括重大不良心血管事件(MACE)，增加医疗成本， 和死亡。随机对照试验表明，实施肾脏保护策略可预防急性肾损伤 适用于高危患者。有证据表明，这些策略没有得到充分利用，因为AKI的风险往往被低估 在围手术期的环境中。比沙拉博士假设，他的ML模型将识别那些最有可能 在可以启动预防策略的时候，及时进行肾脏保护干预，从中受益。

The goal of this K23 application is to provide Dr. Bishara with the necessary research experience and time to establish himself as a primary investigator focused on designing and implementing machine learning (ML) and artificial intelligence in the perioperative setting. The career development activities in this application include early intensive course work in ML and statistics focused on improving model development and causal inference techniques. Then coursework focuses on clinical trial training, grantsmanship, responsible conduct of research, and culminates in a course studying implementation science and algorithmic human-robot interaction. Augmenting this training are project-specific tutorials with experts to improve the models proposed in this application with a focus on real-time prediction of perioperative acute kidney injury (AKI) and describing the risk landscape of perioperative AKI. To achieve these goals, Dr. Bishara has assembled a team of experts and mentors in the areas of data science, AKI, ML, and statistics. Dr. Atul Butte, his primary mentor, is an expert in data science and ML and has trained nearly 100 post-doctoral fellows, undergraduate and graduate students, and staff. Dr. Kathleen Liu is a thought leader in the field of AKI with an active research program focused on AKI and critical care clinical trials. She has mentored numerous junior faculty, including previous NIH K23 awardees. Dr. Romain Pirracchio is an expert in biostatistics and ML in acute care. He has collaborations with Berkeley and the FDA and over 100 publications in the realm. These three mentors and the impressive team of advisors will guide Dr. Bishara to complete the project described below and to grow into an independent investigator. There has been a recent surge in the published literature on ML in medicine, and studies have shown patient care improves when provider expertise is augmented by ML. Unfortunately, implementing published ML models to inform clinical care is not trivial, as many obstacles exist. This application focuses on exploring and overcoming those obstacles by implementing specific models in the perioperative setting. Dr. Bishara has developed novel ML visualization technology that allows for improved interactions between providers and models, which provide predictions and recommendations to those providers. This technology also allows for improved regular monitoring and interpretation of the model to assure sustained accuracy and reliability. He will apply this new technology to predict perioperative AKI in real-time, building upon models he has developed. Postoperative AKI is a major public health problem affecting up to 47% of patients and is consistently associated with adverse outcomes, including, major adverse cardiovascular events (MACE), increased healthcare costs, and death. Randomized controlled trials show that implementation of kidney-protective strategies prevents AKI for high-risk patients. Evidence suggests these strategies are underutilized as risk of AKI is often underestimated in the perioperative setting. Dr. Bishara hypothesizes his ML models will identify those patients who will most benefit from timely kidney protective interventions at the time when preventative strategies can be initiated.