Construction of a risk prediction model for severe community-acquired pneumonia in children in Qinghai region.

objectiveThis study aims to identify risk factors associated with the progression of community-acquired pneumonia (CAP) to severe CAP in children from Qinghai Province and to develop a risk prediction model, thereby providing a basis for early identification of severe CAP in children and guiding clinical decision-making.

methodsThis study retrospectively enrolled paediatric patients (aged ≥29 days to ≤14 years) diagnosed with CAP and hospitalised in the Department of Pediatrics at Qinghai University Affiliated Hospital between June 2023 and September 2024. The patients were randomly allocated into a training cohort (n=345) and a validation cohort (n=149) at a 7:3 ratio. Predictors were selected via least absolute shrinkage and selection operator (LASSO) regression, and a multivariate logistic regression model (LR) was developed. Discrimination and calibration were assessed using the area under the receiver operating characteristic curve (AUROC) and calibration curves, while clinical utility was evaluated through decision curve analysis (DCA) and clinical impact curves (CIC).

resultsLASSO and LR analyses identified several significant predictors of severe paediatric CAP: duration of cough, CT consolidation patterns, lymphocyte percentage, procalcitonin, C-reactive protein (CRP), alanine aminotransferase (ALT), aspartate transaminase (AST), lactate dehydrogenase (LDH), interleukin (IL)-6 and IL-8. The R software-generated nomogram demonstrated strong predictive performance, with training set AUROC=0.863 (95% CI 0.811 to 0.916; specificity=0.827, sensitivity=0.809) and validation set AUROC=0.829 (95% CI 0.732 to 0.927; specificity=0.893, sensitivity=0.714). Bootstrap validation (1000 iterations) and calibration curves confirmed model robustness. Hosmer-Lemeshow tests showed excellent fit for both training (p=0.416) and validation sets (p=0.775). DCA revealed superior net benefit across clinical thresholds versus 'treat-all' or 'treat-none' strategies. CIC demonstrated close agreement between predicted and observed high-risk cases at thresholds >0.5 (training) and >0.4 (validation), confirming clinical utility.

conclusionIn this study, we successfully constructed a model for predicting the occurrence of severe CAP in children in Qinghai based on 10 clinical indicators: number of days of coughing, site of lung solid involvement, percentage of lymphocytes, procalcitonin, CRP, ALT, AST, IL-6, IL-8 and LDH, which had a good predictive efficacy and was useful for the intuitive clinical assessment. The model has good predictive efficacy. This study developed a nomogram model for severe CAP by integrating multidimensional clinical, radiological and laboratory indicators. The model not only provides a reliable tool for early identification of severe CAP in paediatric populations from the Qinghai region but also demonstrates promising clinical application potential with its high sensitivity and specificity. Future multicentre collaborative studies are warranted to further optimise the model and explore its potential value in dynamic monitoring and precision therapy, thereby advancing personalised and precision management of paediatric CAP.