Development and Validation of Multivariable Machine-Learning Models for the Prediction of Multisystemic Inflammatory...

Area of Science:

• Pediatric critical care medicine
• Machine learning applications in healthcare
• Infectious disease epidemiology

Background:

• Multisystem inflammatory syndrome in children (MIS-C) is a serious condition requiring prompt identification of severe cases.
• Predicting severe outcomes in MIS-C is crucial for timely intervention and resource allocation.

Purpose of the Study:

• To develop and validate machine learning (ML) algorithms for predicting severe outcomes in children diagnosed with MIS-C.
• To identify key predictors of severe MIS-C using ML techniques.

Main Methods:

• An observational ambispective cohort study included 1303 children with MIS-C from 84 hospitals in the REKAMLATINA network.
• Multiple ML models were developed to predict pediatric intensive care unit (PICU) admission, inotrope use, mechanical ventilation, and death.
• Model performance was evaluated using accuracy and AUROC, with variable importance assessed via SHAP values.

Main Results:

• The ML model for PICU admission prediction achieved an accuracy of 0.80.
• Models for inotrope use and mechanical ventilation showed accuracies of 0.86 and 0.84, respectively.
• The model predicting death demonstrated an AUROC of 0.85.

Conclusions:

• Validated ML models can effectively predict severe outcomes in children with MIS-C.
• These predictive tools can assist clinicians in risk stratification, enabling earlier identification of high-risk patients.
• The developed models offer potential for improved management of MIS-C cases.