A Prognostic Model to Predict Survival in Children with Ebola Virus Disease

Abstract

Repeated outbreaks of Ebola Virus Disease (EVD) in low-resource settings emphasize the importance of evidence-based guidelines to direct treatment. Previous research has shown that EVD causes high case fatality rates (CFRs) in young children, yet there are limited data focusing on pediatric patients. Here we present a prognostic model to predict mortality in children who are Ebola-positive using information available during the first 48 hours after admission to the treatment center. A logistic regression model was trained on triage data from the Ebola Data Platform, a repository of retrospective patient data compiled from actors that responded to the West African EVD outbreak from 2014-2016. Patients <18 years of age were included in the analysis (N=579) and the CFR was 40%. Overall 13% of data were missing, and multiple imputation was used to estimate missing values. Variable selection using elastic net regularization selected age, CT value, bleeding, breathlessness, bone or muscle pain, anorexia, swallowing problems, and diarrhea as predictors. Bootstrap validation yielded an optimism-corrected area under the curve (AUC) of 0.75 (95% CI: 0.71-0.79). The model was externally validated using data from the current EVD outbreak in the Democratic Republic of the Congo (DRC). While the model’s discriminative ability on the DRC data was similar (AUC=0.75, 95% CI: 0.63-0.87) to the training data, calibration was poor. We recalibrated the model by re-estimating the intercept and slope, and further improved model performance by including aspartate aminotransferase (AST) as a biomarker. The updated model with AST as an added predictor has an AUC of 0.90 (95% CI: 0.77-1). These preliminary results are encouraging but should be interpreted with caution because of limited availability of AST values in the validation data (n=25). The prognostic model described here has promising potential for use in a clinical setting and will continue to be validated as more data becomes available. Future efforts will focus on integrating the validated model into mHealth tools to aid clinicians in making informed, data-driven decisions about patient care.