A Machine Learning-Based Triage Tool for Children With Acute Infection in a Low Resource Setting

Area of Science:

• Pediatric critical care
• Machine learning in healthcare
• Global child health

Background:

• Hospital mortality in children with acute infections remains high in low- and middle-income countries (LMICs).
• Predictive models are crucial for early intervention and resource allocation in pediatric care.

Purpose of the Study:

• To develop and validate a machine learning model for predicting hospital mortality in children with acute infections in LMICs.
• To identify key clinical variables at admission for accurate mortality prediction.

Main Methods:

• A post hoc analysis of a prospective feasibility trial involving 949 children admitted with acute infections in rural Rwanda.
• Random forests, a machine learning algorithm, were employed to build predictive models using variables like age, vital signs, and mental state.
• Five models were tested, comparing different combinations of variables and optimization criteria.

Main Results:

• The overall in-hospital mortality rate was 1.5%.
• All five machine learning models demonstrated good predictive performance, with Area Under the Curve (AUC) ranging from 0.69 to 0.8.
• The optimal model, incorporating age, respiratory rate, capillary refill time, and altered mental state, achieved an AUC of 0.8.

Conclusions:

• Machine learning, utilizing readily available admission data, can reliably predict hospital mortality in pediatric populations in Sub-Saharan Africa.
• The developed model offers a promising tool for improving clinical decision-making and patient outcomes in resource-limited settings.
• Further validation in larger, diverse pediatric cohorts is recommended to strengthen the algorithm's generalizability.