A machine learning-based early in-hospital risk stratification model for severe mycoplasma pneumoniae pneumonia in children

Background: Mycoplasma pneumoniae pneumonia (MPP) is a common cause of community-acquired pneumonia in children. Although most children recover after routine treatment, some develop severe MPP (SMPP), which is associated with greater complications and treatment burden. Early identification of children at high risk of SMPP remains clinically important. This study aimed to develop an early in-hospital prediction model for SMPP using routine clinical data.

Methods: This single-center retrospective study included 98 hospitalized children with MPP at Chuzhou Hospital Affiliated to Anhui Medical University between February 2022 and July 2023, including 49 with SMPP and 49 with non-severe MPP. SMPP status during hospitalization was used as the study outcome. Candidate predictors were obtained from routinely available early in-hospital clinical data, and five predictors were finally retained for model development: basophil absolute count (BASO), platelet count (PLT), atelectasis, pleural effusion, and extrapulmonary complications (EXCP). The dataset was randomly split into training and validation sets at a ratio of 7:3, and seven machine learning models were developed and compared.

Results: Five variables were selected: BASO, platelet count, atelectasis, pleural effusion, and EXCP. Based on overall performance, the multilayer perceptron model was selected for further optimization. The optimized model showed an area under the curve of 0.971 in the training set and 0.933 in the validation set, which should be interpreted as exploratory internal estimates because of the small dataset and the absence of external validation. SHapley Additive exPlanations (SHAP) analysis indicated that higher PLT and the presence of atelectasis, pleural effusion, and EXCP were associated with higher predicted risk, whereas higher BASO was associated with lower predicted risk.

Conclusions: This study developed an exploratory multivariable machine learning model for early in-hospital severity risk stratification of children with MPP. Because of the limited sample size, small validation cohort, and clinical proximity between several retained predictors and the SMPP severity framework, the model should be interpreted as a structured auxiliary tool for in-hospital risk stratification rather than a stable prognostic model or a replacement for routine clinical judgment. Further external validation and clinical utility assessment in larger multicenter cohorts are required before broader clinical application.