Development and internal validation of a nomogram for predicting sepsis risk in neonates with pneumonia

Abstract

To develop and internally validate an individualized model for predicting sepsis risk in neonates with pneumonia and to support early risk stratification at admission. Clinical records of 817 neonates with pneumonia admitted to the Department of Neonatology, Jiangsu Province (Suqian) Hospital, between January 2021 and March 2026 were retrospectively reviewed. The outcome was the occurrence of sepsis. Patients were divided at an approximate 7:3 ratio into a training cohort ( n = 571) and a test cohort ( n = 246). Demographic characteristics, age at admission, gestational age (GA), birth weight (BW), Apgar scores, and routine laboratory findings were collected. Candidate predictors were selected using least absolute shrinkage and selection operator (LASSO) regression, and a nomogram was then constructed using multivariable logistic regression. Receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA) were used to evaluate discrimination, calibration, and clinical net benefit. Sepsis occurred in 109 of the 817 neonates, corresponding to an incidence of 13.3%. The incidence was 13.3% in the training cohort and 13.4% in the test cohort, with no significant difference between cohorts ( P = 0.968). LASSO regression identified eight preliminary variables: sex, GA, procalcitonin (PCT), albumin (ALB), urea (UREA), hematocrit (HCT), white blood cell count (WBC), and C-reactive protein (CRP). After multivariable logistic regression, six variables were retained in the final model: sex, GA, PCT, HCT, WBC, and CRP. Male sex, lower GA, higher PCT, lower HCT, increased WBC, and higher CRP were all associated with a higher risk of sepsis (all P < 0.05). The combined model achieved an area under the receiver operating characteristic curve (AUC) of 0.902 (95% confidence interval [CI]: 0.862–0.943) in the training cohort and 0.861 (95% CI: 0.768–0.954) in the test cohort. Calibration plots showed acceptable agreement between predicted and observed risks, and DCA demonstrated a positive net benefit across a defined range of threshold probabilities. A nomogram incorporating sex, GA, PCT, HCT, WBC, and CRP showed favorable predictive performance for sepsis risk stratification in neonates with pneumonia in this single-center retrospective cohort. These findings should be considered hypothesis-generating, and external validation is required to assess the generalizability and clinical utility of the model.