Construction and validation of neonatal sepsis diagnostic nomograms: a single-site retrospective study.

Background: Neonatal sepsis remains a major cause of morbidity and mortality, while blood culture, although considered the diagnostic reference standard, is limited by delayed reporting and reduced sensitivity in routine practice. This study aimed to develop and internally validate practical diagnostic nomograms for neonatal sepsis using routinely available clinical and laboratory data. Methods: We conducted a single-center retrospective study using medical record and laboratory data from the Children's Hospital of Soochow University between October 13, 2019 and October 13, 2025. After screening 523 records, 483 neonates with sepsis were included in the case group, and controls were sampled from hospitalized neonates with non-septic infection or no infection during the same period. Candidate predictors were screened using univariable logistic regression and further reduced using least absolute shrinkage and selection operator regression, with final variable selection informed by clinical relevance and published evidence. Two logistic regression models were constructed and presented as nomograms: model A included maternal and infant clinical variables, and model B additionally incorporated routine laboratory indicators. Internal validation was performed using a 7:3 training-test split. Model performance was assessed by discrimination and calibration. Results: The training and test sets included 338 and 145 sepsis cases and 140 and 60 controls, respectively. Model A incorporated maternal residence, threatened abortion history, breastfeeding status, prenatal antibiotic exposure, infant temperature, weak mental response, primitive reflex status, antibiotics before visit, and weight difference between visit and birth. Model B added abnormal white blood cell count and elevated C-reactive protein. In the training and test sets, the area under the receiver operating characteristic curve was 0.839 and 0.852 for model A, and 0.905 and 0.923 for model B, respectively. Sensitivity and specificity were 0.654 and 0.907 for model A in the training set and 0.710 and 0.867 in the test set; corresponding values for model B were 0.864 and 0.800 in the training set and 0.910 and 0.750 in the test set. Calibration was acceptable overall, although model B showed weaker calibration in the training set. Conclusion: Two nomogram-based diagnostic models for neonatal sepsis were developed and internally validated in a single-center retrospective case-control sample. Both models showed useful discriminatory ability, and the model combining clinical and laboratory variables performed better. Because the nomogram probabilities are sample-based estimates rather than absolute population risks, external validation and prevalence-based recalibration are needed before broader clinical application.