Prediction of Health Outcomes and Adverse Events in Pediatric Organ Transplantation in Florida

摘要 先前的研究已经提供了初步证据，表明机器学习算法可以用于预测 儿科器官移植的术后转归。这一领域的当前预测建模提供了 预测准确性不令人满意，并且没有检查患者和家庭风险的纵向影响 影响移植后结果的因素。R21探索性/发展性研究补助金的目标 方案建议是将高级预测建模的使用整合到儿科疾病的预测中 移植后的健康结果，以改善对患者和移植物存活率的预测。在协作中 在佛罗里达州立大学(FSU)、佛罗里达大学(UF)和迈阿密大学(UM)之间 拟议的R21项目将支持通过高级技术预测移植后健康结果的研究 儿科器官移植患者的预测性建模。这个项目的总体目标和新颖性是 使用患者电子健康记录(EHR)数据、特定中心器官共享联合网络(UNOS) 来自佛罗里达州两个最大的具有机器学习(ML)的移植中心的数据和文本临床数据， 深度学习(DL)和自然语言处理(NLP)，以开发多种预测模型 儿童移植后的结局。我们建议分析多个数据集，以更好地了解风险因素 影响儿童移植后结局的因素，包括人口、家庭、医疗、健康和其他方面 移植后的特点。移植后的结果包括晚期急性排斥反应，需要再次移植， 和死亡率。我们的中心假设是移植后的长期结果将更有效 由心理社会和医学风险因素组合通过使用高级ML、DL和 NLP分析方法。我们的长期目标是提高儿科移植团队预测 出现移植后不良结果，识别高危患者，减少健康差距，并促进 这些患者的健康结果和生活质量。结果将为临床决策的发展提供信息- 为移植医生和团队制作工具，允许更高效和及时地识别和 对移植后预后不良风险最大的儿童和家庭进行适当的干预。

Abstract Prior research has provided initial evidence that machine learning algorithms can be used to predict posttransplant outcomes in pediatric organ transplantation. Current prediction modeling in this area offers unsatisfactory predictive accuracy and has not examined the longitudinal effects of patient and familial risk factors on posttransplant outcomes. The objective of this R21 Exploratory/Developmental Research Grant Program proposal is to integrate the use of advanced predictive modeling into the prediction of pediatric posttransplant health outcomes in order to improve prediction of patient and graft survival. In a collaboration between Florida State University (FSU), the University of Florida (UF), and the University of Miami (UM), the proposed R21 project will support research predicting posttransplant health outcomes through advanced predictive modeling in pediatric organ transplant patients. The overall objective and novelty of this project is to use patient electronic health record (EHR) data, center-specific United Network for Organ Sharing (UNOS) data, and textual clinical data from the two largest transplant centers in Florida with machine learning (ML), deep learning (DL), and natural language processing (NLP) to develop multiple predictive models of posttransplant outcomes in children. We propose to analyze multiple datasets to better understand risk factors that affect posttransplant outcomes in children, including demographic, familial, medical, health, and other posttransplant characteristics. Posttransplant outcomes include late acute rejection, need for retransplantation, and mortality. Our central hypothesis is that long-term posttransplant outcomes will be more effectively predicted by a combination of psychosocial and medical risk factors through the use of advanced ML, DL, and NLP analytic approaches. Our long-term goal is to improve the ability of pediatric transplant teams to predict emerging poor posttransplant outcomes, identify high-risk patients, reduce health disparities, and promote health outcomes and quality of life in these patients. Results will inform the development of a clinical decision- making tool for transplant physicians and teams, allowing more efficient and timely identification and appropriate interventions with children and families at most risk for poor posttransplant outcomes.