Predicting macrolide resistance in pediatric Mycoplasma pneumoniae pneumonia: A machine learning modeling study

Area of Science:

• Pediatric Infectious Diseases
• Machine Learning in Medicine
• Clinical Prediction Modeling

Background:

• Macrolide-resistant Mycoplasma pneumoniae pneumonia (MRMPP) poses a significant challenge in pediatric care.
• Early identification of MRMPP is crucial for effective treatment and preventing antimicrobial resistance.
• Current diagnostic methods may not always facilitate timely identification of resistant cases.

Purpose of the Study:

• To develop and validate a machine learning-based clinical prediction model for identifying pediatric MRMPP.
• To facilitate early detection of resistant cases and guide targeted therapeutic interventions.
• To provide a data-driven tool for rational antibiotic use in pediatric pneumonia.

Main Methods:

• A retrospective, single-center study utilizing demographic, laboratory, and inflammatory data from pediatric patients with Mycoplasma pneumoniae pneumonia (MPP).
• Development of a stacking ensemble prediction model with feature selection to identify key predictors.
• Validation using internal cross-validation and an independent external temporal cohort; SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The stacking ensemble model demonstrated strong performance with an AUC of 0.857 (internal validation) and 0.812 (external validation).
• Key predictors identified include IL-17A, IFN-γ, CRP, A/G ratio, prior macrolide use, and pre-hospital course.
• The model is accessible via a web tool for rapid assessment of resistance risk.

Conclusions:

• The developed machine learning model effectively identifies children at high risk for MRMPP.
• This tool serves as a valuable decision-making aid for clinicians regarding antibiotic use.
• The model shows significant clinical translational value in managing pediatric pneumonia.