A data science approach to identify and manage Multisystem Inflammatory Syndrome in Children (MIS-C) associated with SARS-CoV-2 infection and Kawasaki disease in pediatric patients

摘要 – 自 SARS-CoV-2 大流行开始以来，出现了相关的新型多系统 儿童炎症综合征（MIS-C）已有报道。有趣的是，MIS-C 患者遵循 表现、治疗和临床过程与川崎病患者有些相似 疾病（KD）。目前，临床特征和治疗方面出现这种重叠的原因尚不清楚，并且 这种重叠是否是部分共享病因学或病理生理学的结果，是激烈争论的主题。 辩论。重叠程度意味着我们开发的一些临床预测工具 KD 的过去可以重新利用，以加速 MIS-C 临床支持决策工具的开发。在 在本研究中，我们将首先（R61 组件）系统地解决 KD 和 MIS-C 之间的重叠问题，并创建 基于机器学习的显着预测模型，用于诊断/识别（目标#1）、管理（目标#2）、 基于我们之前开发的预测模型的 MIS-C 的短期和长期结果（目标#3） KD 的过程类似于迁移学习。其次（R33组件），我们将验证和评估 这些模型在诊断预测临床决策支持系统中的性能和临床实用性 具有 MIS-C 或 KD 特征的儿科患者的管理。在这项研究中我们 将包括 3 组患者：1) 患有 SARS-CoV-2 感染且 MIS-C（CDC 标准）的患者，无论其病情如何 他们是否有重叠的 KD 症状，2) 接受调查的 SARS-CoV-2 感染患者，但 最终未诊断为 MIS-C，以及 3) 患有 KD 但未感染 SARS-CoV-2 的患者。目标数据 将从入组患者中收集（900 名用于训练，450 名用于验证）进行深度表型分析和 生物标志物测量。医生对算法生成的预测的反馈将用于 建立临床实用性。模型训练所需的数据将在活动的前两年中累积（R61 资助期限）；算法的开发及其内部验证将同时进行。在 2年后（R33资助期），我们将进行外部验证，建立临床效用，添加真实的 将流行病学监测数据输入模型并最终打包，并验证算法以供将来使用 部署并集成到电子健康记录中。该项目将与 国际川崎病登记 (IKDR) 联盟。 IKDR 联盟拥有活跃的 KD 和 儿科新冠肺炎登记处遍布全球 35 个地点，目前登记地点数量已扩大至 60 个。 IKDR 中心已识别出 600 多名 MIS-C 患者，这使得该项目明显可行 并完美定位 IKDR 来开展这项研究。我们坚信新兴数据科学的使用 方法以及我们之前在 KD 背景下开发的算法，而不是专注于 MIS-C 单独的患者，将增进我们对 MIS-C 和 KD 的病因学和病理生理学的理解，并将 更快地导致针对 MIS-C 患者的数据驱动管理方案的出现。

Summary – Since the SARS-CoV-2 pandemic began, the emergence of an associated novel multisystem inflammatory syndrome in children (MIS-C) has been reported. Interestingly, patients with MIS-C follow a presentation, management and clinical course that are somewhat similar to that of patients with Kawasaki disease (KD). Currently, the reason for such an overlap in clinical features and management is unclear and whether this overlap is the result of a partially shared etiology or pathophysiology is the subject of fierce debates. The degree of overlap implies that some of the clinical prediction tools that we have developed in the past for KD could be repurposed to accelerate the development of clinical support decision tools for MIS-C. In this study, we will first (R61 component) systematically address the overlap between KD and MIS-C and create salient machine-learning based prediction models for diagnosis/identification (Aim #1), management (Aim #2), and short- and long-term outcomes (Aim #3) of MIS-C based on our previously developed predictive models for KD in a process akin to transfer learning. Secondly (R33 component), we will validate and evaluate the performance and clinical utility of these models in a predictive clinical decision support system for the diagnosis and management of pediatric patients presenting with features indicative of either MIS-C or KD. In this study we will include 3 groups of patients: 1) patients with SARS-CoV-2 infection with MIS-C (CDC criteria) regardless of whether they have overlapping signs of KD, 2) patients with SARS-CoV-2 infection investigated for but eventually not diagnosed with MIS-C, and 3) patients with KD but without SARS-CoV-2 infection. Targeted data will be collected from enrolled patients (900 for training and 450 for validation) for deep phenotyping and biomarker measurements. Physician feedback on the predictions generated by the algorithm will be used to establish clinical utility. Data required for model training will be accrued in the first two years of activity (R61 period of the grant); the development of algorithms and their internal validation will occur concurrently. In the following 2 years (R33 period of the grant), we will perform external validation, establish clinical utility, add real- time epidemiological surveillance data to the models and finally package, and certify the algorithms for future deployment and for the integration in electronic health records. This project will be a collaboration with the International Kawasaki Disease Registry (IKDR) Consortium. The IKDR Consortium has an active KD and pediatric COVID registry in 35 sites across the world and the number of sites is currently expanding to 60+ sites. More than 600 MIS-C patients have already been identified at IKDR centers, making this project clearly feasible and perfectly positioning IKDR to perform this study. We strongly believe that the use of emerging data science methods and of our previously developed algorithms in the context of KD, as opposed to focusing on MIS-C patients alone, will boost our understanding of the etiology and pathophysiology of both MIS-C and KD and will more rapidly lead to the emergence of data-driven management protocols for patients with MIS-C.