A new approach to predict graft outcome from histology in children undergoing kidney transplantation using machine learning methods

摘要 尽管在过去的几十年里，移植肾的总体存活率有了显著的提高， 几乎一半的儿科患者在10年内失去了移植。流行病学、免疫学 和组织学都是了解移植后并发症的关键因素， 预测临床结果，但准确预测移植结果仍然存在 在儿童肾移植受者中具有挑战性和研究不足。机器学习 技术正在迅速发展，并已显示出显著的预测准确性 在几个生物医学应用中超越了经典的预测模型。这个目标 一项研究建议是开发机器学习方法，以提高移植预测 通过组织学分析，收集足够的数据来描述 不同类型的排斥反应和儿科肾移植活检的组织学结果， 评估它们对移植结果的影响。 为了实现这些目标，我们将组建一个由700多名儿童组成的国际多中心队列， 肾移植受者，并开发和验证机器学习模型， 临床、免疫学和组织学数据，用于预测短期儿科移植结果。 这项研究将使组织病理学评估准确、可重复和标准化。 在儿科移植受者中的发现，并提供了一种新的方法来预测移植结果。这 将为未来评估可行性和潜在影响的大型前瞻性研究铺平道路 机器学习方法在移植活检分析中的应用，目的是改善临床 通过识别移植物丢失高风险患者来管理和资源利用。最后， 本研究将为高效、有效地设计重点临床试验提供关键信息 改善儿童肾移植受者的预后，重点是 个性化治疗。

Abstract Although the overall renal graft survival has dramatically improved over the last several decades, almost half of pediatric patients lose their transplant within 10 years. Epidemiology, immunology and histology are all key elements in understanding post-transplant complications and in predicting clinical outcomes, however accurate prediction of transplant outcomes remains challenging and understudied among pediatric kidney transplant recipients. Machine learning technology is rapidly advancing and has demonstrated remarkable predictive accuracy surpassing classical predictive models in several biomedical applications. The goal of this research proposal is to develop machine learning methods to improve prediction of transplant outcomes through analysis of histology, assemble data sufficient to describe the incidence of different types of rejection and histological findings on pediatric kidney transplant biopsies, and to assess their impact on graft outcome. To achieve these goals, we will assemble an international multicenter cohort of over 700 pediatric kidney transplant recipients and develop and validate machine-learning models combining clinical, immunological and histological data for predicting short-term pediatric graft outcomes. This study will enable accurate, reproducible, and standardized assessment of histopathological findings in pediatric transplant recipients and provide a new method to predict graft outcome. This will pave the way to future large, prospective study assessing the feasibility and potential impact of machine-learning methods in transplant biopsy analysis, with the goal of improving clinical management and resource utilization by identifying patients with high risk of graft loss. Finally, this study will provide critical information for efficient and effective design of clinical trials focused on improving outcomes of pediatric kidney transplant recipients, with an emphasis on individualized treatment.