Screening of early predictive serum biomarkers and construction of a combined predictive model for refractory Mycoplasma pneumoniae infection in children

TL;DR

This study identifies serum biomarkers and builds a predictive model to distinguish severe Mycoplasma pneumoniae infections in children early on.

Contribution

A novel multi-marker random forest model is developed for early prediction of refractory Mycoplasma pneumoniae infection in children.

Findings

RMPP patients showed higher rates of severe symptoms and longer hospital stays compared to GMPP.

Elevated CRP, LDH, IgA, IL-6, IL-10, and IFN-γ were identified as significant serum biomarkers for RMPP.

The random forest model achieved high accuracy (AUC = 0.978) in distinguishing RMPP from GMPP.

Abstract

Mycoplasma pneumoniae pneumonia (MPP) is a prevalent respiratory infection. Refractory MPP (RMPP) presents more severe symptoms and requires more intensive treatment compared to general MPP (GMPP). This study aimed to identify distinguishing clinical, laboratory, and radiological characteristics between RMPP and GMPP and develop an early predictive model for RMPP risk stratification.

A total of 568 patients, including 130 RMPP cases and 438 GMPP cases, were enrolled. Clinical information, laboratory tests, and radiological features were compared. Univariate and multivariate logistic regression analyses identified serum biomarkers associated with RMPP. A combined predictive model using random forest approach was developed and externally validated.

RMPP patients showed significantly higher rates of older age, fever, tachypnea, chest tightness, wheezing, chills, extrapulmonary complications, decreased unilateral lung sounds, longer fever duration, hospital stay, antibiotic therapy, oxygenotherapy use, and Intensive Care Unit (ICU) admission (all P < 0.05). Laboratory findings revealed elevated neutrophil percentage, C-reactive protein (CRP), lactate dehydrogenase (LDH), immunoglobulin A (IgA), interleukin (IL)-6, IL-10, and interferon-gamma (IFN-γ), but lower prealbumin (PAB) concentrations in RMPP. Radiologically, RMPP exhibited more severe manifestations such as large lesions, pleural effusion, lobar atelectasis, pulmonary consolidation, and pleural thickening. Using the eight independently associated serum biomarkers, we developed a multi-factor random forest model that showed excellent discrimination between RMPP and GMPP (AUC = 0.978 in the development cohort), which was confirmed in an external validation cohort (AUC = 0.957).

Significant differences in clinical, laboratory, and radiological characteristics were observed between RMPP and GMPP. The combined multi-marker model shows strong potential for early risk identification of RMPP and may support timely clinical decision-making; however, prospective validation is needed before routine implementation.

Figures