Early identification of sepsis in critically ill children using inflammatory and nutritional markers: development and internal validation of a prediction model with a maternal–neonatal subgroup analysis

Objective To construct and validate an early prediction model for pediatric ICU sepsis based on easily accessible inflammatory and Nutritional markers, to provide an objective quantitative tool for ICU clinicians to rapidly screen sepsis in critically ill children. Methods 456 critically ill children (Including 55 newborns) admitted to the ICU of A certain children's hospital from January 2024 to December 2025 were retrospectively enrolled, and randomly divided into a training set and a validation set at a ratio of 7:3. Taking whether the child was diagnosed with sepsis as the primary outcome, inflammatory and related biochemical markers including age, gender ratio, Lactate, WBC, PCT, Plt, N%, L%, RDW, CRP, Tbil, Cre, ALB, ALT and IL-6 and etc. were included for univariate and multivariate logistic regression analysis to screen independent predictive factors for sepsis. The “rms” package in R software was used to construct a visual nomogram model. The discrimination, calibration, and clinical net benefit of the model were verified through ROC curve, calibration curve, and decision curve analysis, respectively. Results The occurrence of neonatal sepsis is related to prenatal chorioamnionitis( n = 9), urogenital tract infections( n = 6), and colonization of Group B streptococcus (GBS) in the mother( n = 5). Eight independent predictive factors were finally included to construct the nomogram. The prediction model is abbreviated as Log(P/1-P)=-1.63 + 0.34 LACT+0.03WBC-2.75L%+0.23RDW + 0.01CRP-0.11ALB + 0.01ALT + 0.01IL-6. The area under the ROC curve (AUC) of the training and validation set were 0.87, and 0.81, respectively. The calibration curves were close to the 45° reference line. Decision curve analysis showed that the training set could obtain a higher net benefit than the “intervene all” strategy within the risk threshold range of 0.07–0.90, and the validation set could still maintain stable clinical utility within the range of 0.07–0.82, indicating that the model has good generalization ability. Conclusion The nomogram prediction model for pediatric sepsis has good discrimination and calibration, with a wide range of clinical application scenarios. It can assist pediatric ICU physicians in early identification of children at high risk of sepsis and provide evidence-based support for clinical diagnosis and treatment decisions.