Critical Care Informatics

摘要 这是重症监护信息学补助金（NIBIB R 01 EB 017205）的续期申请，该补助金授予 计算生理学实验室（LCP）这笔赠款支持了 重症监护医学信息集市（MIMIC）研究数据库，这是一个去识别数据库， 世界各地的研究和健康数据科学教育。我们的目标是解决机器中最重要的问题 学习今天的医疗保健，专注于健康差异，算法偏见，并了解 有效和公平地实施算法模型。我们的建议将丰富MIMIC的新数据 类型和来源，包括公开的人口健康数据集，推进联合 重症监护数据集，并添加一个新的模块与COVID-19特定的本体和代码。我们的研究将建立 先前的研究结果显示，数据来源中隐藏的社会人口统计学偏见普遍存在，包括临床 数据、医学图像和叙述性患者文档。医疗保健数据科学最终存在于 目的是改善人类健康。然而，在研究论文中发表的模型很少有影响 临床护理由于实施的挑战。通俗地说，这意味着获得有关测试的知识 和治疗导致最好的结果，为每一个病人在重症监护室（ICU），无论 人口统计学。我们将进行严格的定性研究，以更好地了解 关键利益相关者-临床医生和数据科学家-在开发和实施以公平为中心的 人工智能这些信息将用于制定与执行科学相结合的准则 框架，以支持在临床环境中有效实施公平的AI。基于这些目标，MIMIC 将显著扩大这一研究资源的相关性，以更广泛的研究人员，包括 那些在社会和行为科学和公共卫生，并继续成为临床研究的资源 和日益复杂的模型开发，推进我们对公平性关键问题的理解 以及医疗保健、数据科学和更广泛社会的公平性。

Abstract This is a renewal application for the Critical Care Informatics grant (NIBIB R01 EB017205) that was awarded to the Laboratory for Computational Physiology (LCP) in 2014. This grant has supported the development of the Medical Information Mart for Intensive Care (MIMIC) research database, which is a de-identified database for research and health data science education around the world. We aim to address the foremost issues in machine learning in healthcare today, focusing on health disparities, algorithmic bias, and understanding the barriers to effective and equitable implementation of algorithmic models. Our proposal will enrich MIMIC with new data types and sources, including publicly available population health datasets, advance progress on a federated critical care dataset, and add a new module with COVID-19 specific ontology and codes. Our research will build on prior findings showing the pervasiveness of hidden socio-demographic bias in data sources including clinical data, medical images, and narrative patient documentation. Health care data science ultimately exists for the purpose of improving human health. Yet, extremely few models published in research papers have impacted clinical care due to challenges in implementation. In layman’s terms, this means knowledge gained about tests and treatments leads to the best possible outcome for every patient in the intensive care unit (ICU) regardless of demographic. We will conduct a rigorous qualitative research study to better understand the barriers faced by key stakeholders - clinicians and data scientists - in the development and implementation of equity-centered artificial intelligence. This information will be used to develop guidelines to integrate with implementation science frameworks to support the effective implementation of equitable AI in clinical settings. With these aims, MIMIC will significantly expand the relevance of this research resource to a greater diversity of investigators including those in the social and behavioral sciences and public health and continue to be a resource for clinical research and increasingly sophisticated model development, advancing our understanding of critical issues of fairness and equity in healthcare, data science, and the broader society.