Risk factors for mortality in neonates with bloodstream infections: Development and validation of a nomogram

Background: Neonatal bloodstream infection is one of the major causes of neonatal death. A prediction model for adverse outcomes may support risk stratification in neonates with bloodstream infections; however, its clinical utility requires appropriate validation. This study aimed to develop and temporally validate a nomogram for predicting a study-defined composite outcome in neonates with bloodstream infections.

Methods: This single-center retrospective study evaluated predictors of a composite outcome comprising in-hospital death attributed to bloodstream infection or discharge against medical advice after clinical deterioration despite maximal therapy. Death after discharge was not confirmed. A nomogram was developed to estimate the probability of the study-defined composite outcome, and its predictive performance was evaluated. The clinical data of 181 neonates with bloodstream infections admitted to Tianjin Central Hospital of Gynecology Obstetrics from September 2018 to September 2023 were used as the modeling cohort, whereas patients admitted from October 2023 to December 2024 at the same hospital constituted the temporal validation cohort. We analyzed the species and proportions of bacteria detected in blood cultures. The data were analyzed using R version 4.4.2. Least absolute shrinkage and selection operator (LASSO) regression and multivariate logistic regression were employed to identify predictors associated with the study-defined outcome. We then constructed a nomogram, and its performance was evaluated using the calibration curve, receiver operating characteristic curve and area under the curve (AUC).

Results: The final nomogram incorporated six variables: gestational age, Apgar score, respiratory distress, hyperbilirubinemia, procalcitonin (PCT), and carbapenem exposure. Gestational age was retained after LASSO selection because of its established clinical relevance, although it did not reach conventional statistical significance in the multivariable model. Graphical calibration in the modeling cohort showed deviations between predicted probabilities and observed outcome proportions. In the modeling cohort, the nomogram had an AUC of 0.822, whereas the AUC was 0.863 in the temporal validation cohort, indicating good discriminative performance.

Conclusions: An exploratory six-variable nomogram was developed to predict the study-defined outcome and showed good discrimination in the modeling and same-center temporal validation cohorts. Graphical calibration showed deviations from ideal agreement in the modeling cohort. Further independent external validation and prospective evaluation of its clinical utility are required before clinical implementation.