Background Late-onset sepsis is one of the main causes of adverse events and mortality in premature infants, and its early evaluation and prediction are essential.

Objective To explore the risk factors for late-onset sepsis in premature infants and construct a nomogram model to predict the risk of late-onset sepsis in premature infants.

Methods Retrospective analysis was perfomed in preterm infants who were born at the Seventh Medical Center of Chinese PLA General Hospital from January 2019 to June 2023 and transferred to the neonatal intensive care unit for hospitalization immediately after birth, and they were divided into late-stage sepsis group and non-sepsis group according to the diagnosis. The pregnancy data of mothers, basic data of the infants, and their clinical data were collected. Taking December 2022 as the cut-off point, they were divided into modeling group and validation group, and the modeling group was screened for variables using the LASSO regression model, and the statistically significant factors were constructed into a column-line graph prediction model. Clinical efficacy of the predictive model was assessed for differentiation and accuracy using subject operating characteristic (ROC) curves, area under the curve (AUC), C-statistic, and calibration curves. For the validation group, the area under the curve (AUC) of the subject's work characteristic curve (ROC) was used to assess the discrimination and calculate the sensitivity and specificity of the model. And 300 times bootstrap resampling was applied for internal validation of the model.

Results A total of 734 preterm infants (105 cases of late-onset sepsis in preterm infants) were included in the modeling group and 133 cases (13 cases of late-onset sepsis in preterm infants) were included in the validation group. LASSO regression analysis found that the child's birth weight, gestational age, Apgar score (10 min), history of bronchopulmonary dysplasia, necrotizing small intestinal colitis, respiratory distress syndrome, and endotracheal intubation were the risk factors for the development of late-onset sepsis in preterm infants in the modeling group. A column-line graphical model was constructed, and by plotting the ROC curve, the AUC of the predictive model was 0.748 (95% CI: 0.694-0.802) and the C-index value was 0.761, suggesting that the model had a good degree of differentiation and precision. The C-index value obtained from the resampling internal validation was 0.744, suggesting that the model had good stability, and in the validation population, the external validation of the column-line graph prediction model had a ROC curve AUC of 0.843 (95% CI: 0.742-0.944), suggesting that the model had good predictive performance.

Conclusion The prediction model constructed based on birth weight, gestational age, Apgar score (10 min), bronchopulmonary dysplasia history, necrotizing enterocolitis, respiratory distress syndrome and endotracheal intubation has good predictive performance, but further external validation is needed.