A machine learning-based phenotype for long COVID in children: an EHR-based study from the RECOVER program.

Area of Science:

• Pediatric Health Informatics
• Machine Learning in Clinical Research
• Epidemiology of Post-Acute Sequelae of SARS CoV-2 (PASC)

Background:

• Clinical understanding and definitions of pediatric Post-Acute Sequelae of SARS CoV-2 (PASC) are evolving.
• Reliable identification of PASC patients within health systems data is crucial for research and clinical care.
• Distinguishing PASC from Multisystem Inflammatory Syndrome in Children (MIS-C) is important for accurate classification.

Approach:

• Developed and validated a machine learning algorithm using the PEDSnet Electronic Health Record (EHR) network.
• Selected patient features from conditions, procedures, diagnostics, and medications using a tree-based scan statistic.
• Employed an XGBoost model with hyperparameter tuning via cross-validated grid search and evaluated using 5-fold cross-validation.
• Utilized Shapley Additive exPlanations (SHAP) for model prediction and feature importance analysis.

Key Points:

• The algorithm effectively classifies pediatric patients with PASC from electronic health records.
• Feature importance analysis using SHAP values provides insights into characteristics associated with PASC.
• The model demonstrated robust performance through rigorous validation methods.

Conclusions:

• The developed machine learning model serves as a valuable tool for identifying and characterizing PASC in pediatric populations.
• The model's flexibility allows for precise patient identification for studies or broad screening for clinical trials.
• This approach enhances the utility of health systems data for PASC research, especially where diagnostic codes are unreliable.