A retrospective study using machine learning to develop predictive model to identify rotavirus-associated acute...

Area of Science:

• Pediatric infectious diseases
• Computational epidemiology
• Clinical informatics

Background:

• Rotavirus is a primary cause of severe dehydrating diarrhea in young children globally.
• Limited access to laboratory diagnostics in hospitals necessitates alternative diagnostic approaches.
• Machine learning (ML) shows promise for symptom-based disease diagnosis in resource-constrained environments.

Purpose of the Study:

• To develop an ML predictive model for rotavirus infection using clinical parameters.
• To avoid reliance on laboratory tests for diagnosis.
• To support timely and accessible rotavirus diagnosis.

Main Methods:

• Collected clinical data from 509 children, including symptoms like diarrhea, vomiting, fever, and dehydration.
• Performed correlation and feature selection (ANOVA F test) to identify important clinical indicators.
• Trained and compared seven supervised ML models: SVM, KNN, NB, Log_R, RF, DT, and XGBoost.
• Evaluated model performance using accuracy, precision, recall, specificity, F1, F2, macro F1, and AUC.

Main Results:

• The Random Forest (RF) model demonstrated superior performance among the seven ML models.
• RF achieved an accuracy of 81.4%, an F1 score of 86.9%, a macro F1-score of 77.3%, an F2 score of 86.5%, and an AUC of 89%.

Conclusions:

• ML models can aid in the symptom-based diagnosis of rotavirus-associated acute gastroenteritis in children.
• These models are particularly valuable in resource-limited settings.
• Further validation with larger datasets is recommended to optimize sensitivity and specificity for pediatric diarrheal diseases.