Development and validation of a federated learning framework for detection of subphenotypes of multisystem inflammatory syndrome in children

BackgroundMultisystem inflammatory syndrome in children (MIS-C) is a severe post-acute sequela of SARS-CoV-2 infection. The highly diverse clinical features of MIS-C necessities characterizing its features by subphenotypes for improved recognition and treatment. However, jointly identifying subphenotypes in multi-site settings can be challenging. We propose a distributed multi-site latent class analysis (dMLCA) approach to jointly learn MIS-C subphenotypes using data across multiple institutions. MethodsWe used data from the electronic health records (EHR) systems across nine U.S. childrens hospitals. Among the 3,549,894 patients, we extracted 864 patients < 21 years of age who had received a diagnosis of MIS-C during an inpatient stay or up to one day before admission. Using MIS-C conditions, laboratory results, and procedure information as input features for the patients, we applied our dMLCA algorithm and identified three MIS-C subphenotypes. As validation, we characterized and compared more granular features across subphenotypes. To evaluate the specificity of the identified subphenotypes, we further compared them with the general subphenotypes identified in the COVID-19 infected patients. FindingsSubphenotype 1 (46.1%) represents patients with a mild manifestation of MIS-C not requiring intensive care, with minimal cardiac involvement. Subphenotype 2 (25.3%) is associated with a high risk of shock, cardiac and renal involvement, and an intermediate risk of respiratory symptoms. Subphenotype 3 (28.6%) represents patients requiring intensive care, with a high risk of shock and cardiac involvement, accompanied by a high risk of >4 organ system being impacted. Importantly, for hospital-specific clinical decision-making, our algorithm also revealed a substantial heterogeneity in relative proportions of these three subtypes across hospitals. Properly accounting for such heterogeneity can lead to accurate characterization of the subphenotypes at the patient-level. InterpretationOur identified three MIS-C subphenotypes have profound implications for personalized treatment strategies, potentially influencing clinical outcomes. Further, the proposed algorithm facilitates federated subphenotyping while accounting for the heterogeneity across hospitals. Research in context panelO_ST_ABSEvidence before this studyC_ST_ABSBefore undertaking this study, we searched PubMed and preprint articles from in early 2022 for studies published in English that investigated the clinical subphenotypes of MIS-C using the terms "multi-system inflammatory syndrome in children" or "pediatric inflammatory multisystem syndrome", and "phenotypes". One study in 2020 divided 63 patients into Kawasaki and non-Kawasaki disease subphenotypes. Another CDC study in 2020 evaluated 3 subclasses of MIS-C in 570 children, with one class representing the highest number of organ systems, a second class with predominant respiratory system involvement, and a third class with features overlapping with Kawasaki Disease. However, both studies were conducted during the early phase of the pandemic when misclassification of cases as Kawasaki disease or acute COVID-19 may have occurred. Therefore, the subphenotypes of MIS-C needs further investigation. In addition, we searched research articles for studies published in English on algorithms for distributed multi-site latent class analysis with the terms "distributed latent class analysis" or "multi-site latent class analysis". Most of the existing literatures for distributed learning have focused on supervised learning. Literatures discuss latent class analysis for disease sub phenotyping in a multi-site setting where data are distributed across different sites are lacking. Added value of this studyWe developed a new algorithm to jointly identify subphenotypes of MIS-C using data across multiple institutions. Our algorithm does not require individual-level data sharing across the institutions while achieves the same result as when the data are pooled. Besides, our algorithm properly accounts for the heterogeneity across sites, and it can lead to accurate characterization of the subphenotypes at the patient-level. We then applied our algorithm to PEDSnet data for identifying the subphenotypes of MIS-C. PEDSnet provides one of the largest MIS-C cohorts described so far, providing sufficient power for detailed analyses on MIS-C subphenotypes. We identified three subphenotypes that can be characterized as mild with minimal cardiac involvement (46.1%), severe requiring intensive care with >4 organ being impacted, and the one with intermediate risk of respiratory symptoms, and high risk of shock, cardiac and renal involvement (25.3%). For hospital-specific clinical decision-making, our algorithm revealed a substantial heterogeneity in relative proportions of these three subtypes across hospitals. Implications of all the available evidenceOur algorithm provides an effective distributed learning framework for disease subphenotyping using multi-site data based on aggregated data only. It facilitates high accuracy while properly accounts for the between-site heterogeneity. The results provide an update to the subphenotypes of MIS-C with larger and more recent data, aid in the understanding of the various disease patterns of MIS-C, and may improve the evaluation and intervention of MIS-C.