Related Experiment Videos

Machine learning-based prediction of shigellosis in children under five: development and internal validation of a

Md Fuad Al Fidah1, Md Ridwan Islam1, Asg Faruque1

  • 1NRD, ICDDRB, Dhaka, Bangladesh.

Insights

Machine learning models can predict Shigella infections in children, aiding diagnosis in low-resource settings. This research developed a prototype tool to support risk stratification for shigellosis.

Area of Science:

  • Infectious Diseases
  • Computational Biology
  • Public Health

Background:

  • Shigella causes significant childhood diarrhea and mortality in low- and middle-income countries.
  • Laboratory confirmation of Shigella is often unavailable, and clinical management lacks sensitivity.
  • This highlights the need for accessible diagnostic tools.

Purpose of the Study:

  • Develop and internally evaluate machine learning models to predict microbiologically confirmed Shigella infection.
  • Assess the feasibility of creating a web-based decision-support application using the best model.
  • Prioritize minimizing false negatives in predictions.

Main Methods:

  • Analyzed data from 3356 children with diarrhea from the Global Enteric Multicenter Study.
  • Trained multiple machine learning algorithms using 32 predictors and 10-fold cross-validation.
  • Evaluated models using AUC, recall, and Brier score, prioritizing recall. Simplified the best model to 10 predictors.

Main Results:

  • A Support Vector Machine (SVM) model achieved the highest recall (0.67) and AUC (0.77).
  • The simplified SVM model demonstrated good performance with a Brier score of 0.16.
  • A prototype web-based application was developed for research purposes.

Conclusions:

  • The developed model shows moderate internal performance for risk stratification of shigellosis.
  • The tool has potential in settings with limited diagnostic capacity.
  • External validation and impact evaluation are necessary before clinical use.
Abstract

Related Concept Videos