Building a diagnostic prediction model for severe Mycoplasma pneumoniae pneumonia in children using machine learning.

OBJECTIVE

Mycoplasma pneumoniae is the leading pathogen of community-acquired pneumonia in children. In recent years, M. pneumoniae pneumonia (MPP) has shown a global pandemic trend. The increasing incidence of severe MPP (SMPP) leads to complications and even deaths, severely impacting prognosis and quality of life. Our study aimed to use machine learning to construct an early diagnostic model for severe MPP in children. It supports early prediction, prevention, and individualized precise treatment of SMPP.

METHODS

We collected medical records from 372 MPP cases. We compared case characteristics between groups with and without SMPP and used a random forest to screen key factors. We then constructed a multivariate logistic prediction model. We evaluated the model with ROC curves, calibration curves, and DCA. Five-fold cross-validation tested prediction stability.

RESULTS

We identified ESR, PCT, IL-6, and lung auscultation as key factors to construct the prediction model. The model's ROC was 0.964 (95% CI: 0.945-0.983). Calibration curves and DCA confirmed model accuracy. Five-fold cross-validation validated internal stability.

CONCLUSION

Our study developed a prediction model with good efficacy for early SMPP risk assessment. Our research provides a basis for clinical early prediction and prevention of SMPP, reducing its risk and offering a foundation for individualized treatment and improved long-term outcomes in affected children.