Development and internal validation of machine learning models for identifying concurrent bloodstream infection in children with pneumococcal pneumonia: a retrospective case-control study
Year: 2026
Type: article
Study design: Observational Study
Abstract: Background Streptococcus pneumoniae (SP) is a leading cause of invasive infections in children. Early detection of pneumococcal pneumonia complicated by bloodstream infection currently relies on time-consuming blood culture, which may delay clinical assessment and appropriate management. Therefore, rapid and accurate methods are needed to facilitate timely diagnosis and treatment among children. This study sought to evaluate the feasibility of machine learning (ML)-based models for early identification of pneumococcal pneumonia complicated by bloodstream infection in children, which may inform the development of objective, efficient auxiliary diagnostic tools. Methods This retrospective, single-center internal-validation study included children with confirmed pneumococcal pneumonia who were admitted to The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University between May 2018 and August 2025. Demographic, clinical, laboratory, and imaging data from the first 24 h of admission were collected. These data were obtained before blood culture results were available. The dataset was randomly divided into training (70%) and validation (30%) sets using a fixed seed. Least Absolute Shrinkage and Selection Operator (LASSO)-based feature selection was performed exclusively within the training set. Six ML models were developed, including logistic regression, random forest, support vector machine (SVM), XGBoost, LightGBM, and Naïve Bayes. Model performance was assessed using discrimination, calibration, and decision curve analysis. Positive predictive value (PPV), negative predictive value (NPV), and net benefit were additionally evaluated under external prevalence assumptions of 1%–5%. Bootstrap optimism correction and nested cross-validation were used to assess model stability, and SHapley Additive exPlanations (SHAP) analysis was applied for model interpretation. Antimicrobial susceptibility variables were excluded from the main analysis because they were unavailable at admission. The corresponding sensitivity analysis is provided in the Supplementary Materials. Results A total of 284 children with pneumococcal pneumonia were included (71 with concurrent bloodstream infection). In the validation set, the six ML models achieved AUC values ranging from 0.900 to 0.986. The SVM model achieved an AUC of 0.980 (95% CI: 0.950–1.000, DeLong method). Bootstrap optimism-corrected AUCs ranged from 0.983 to 0.998, with estimated optimism of 0.001–0.017. These findings suggested relative stability under internal validation, though nested cross-validation revealed greater variability for logistic regression. XGBoost was selected for SHAP interpretation based on a multi-criteria framework considering discrimination, calibration, stability, interpretability, and simplicity. Conclusions In this exploratory internal validation study, ML models demonstrate good performance for identifying concurrent bloodstream infection in children with pneumococcal pneumonia. When applied to early admission data, such models have potential as diagnostic support tools for detecting concurrent bloodstream infection in these children, though external validation is still needed before clinical use. Clinical trial number Not applicable.
Source: BMC Infectious Diseases
Authors: Xiaojiao Xia, Xiaoguang Hu, Dongshi Liang, Tingting Weng, Shixiong Wu, Yimei Jin
Language: English
Subfield: Epidemiology
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,090
OpenAlex ID: W7204463310
FWCI:
Cites:
Cited by:
Related to: