PBMC gene expression profiles differentiate refractory Mycoplasma pneumoniae pneumonia in children

ObjectiveRefractory Mycoplasma pneumoniae pneumonia (RMPP) presents a major clinical challenge in children, largely due to the absence of reliable early diagnostic markers, which contributes to delayed intervention and an increased risk of severe complications. This study aimed to identify early diagnostic biomarkers based on peripheral blood mononuclear cell (PBMC) gene expression profiles and to develop and validate a model capable of distinguishing RMPP from general MPP (GMPP) and healthy controls (Normal).MethodsA total of 349 children (117 Normal, 123 GMPP, and 109 RMPP) were chronologically divided into a prospective training cohort (n=295) for model development and a prospective validation cohort (n=54) for external validation. Single-cell RNA sequencing (scRNA-seq) was performed on PBMCs from a discovery cohort (n=8) randomly selected from the training cohort. Differentially expressed genes that were specifically and significantly upregulated in RMPP groups were screened as candidate early diagnostic biomarkers. After primer validation, expressions of these candidate genes were subsequently measured using RT-qPCR in the entire study population. A multinomial logistic regression model with backward selection was developed on the training set, externally validated in the validation set, and its internal validation was further assessed via 1000 bootstrap resamples of the full dataset.ResultscRNA-seq identified eight specifically upregulated genes in the RMPP group. Subsequent RT-qPCR validation in the training cohort confirmed four genes—IGHM, NEAT1, IL32, and ACTG1—as early diagnostic biomarker capable of differentiating among the three groups. A combined four-gene three-category logistic regression model (Normal/GMPP/RMPP) demonstrated strong performance, with macro-average area under the curve values of 0.968 and 0.987 in the training and external validation, respectively. The final model, refit on the full dataset, attained an overall diagnostic accuracy of 88.8% for three-category classification, which was further confirmed by bootstrap resampling (macro-average AUC = 0.969).ConclusionWe established a robust PBMC-based four-gene signature diagnostic model that accurately discriminates among Normal, GMPP, and RMPP statuses at an early disease stage. This model provides a clinically accessible and precise tool to facilitate early intervention and improve patient management.