Prediction for late-onset sepsis in preterm infants based on data from East China

AIM: To construct a prediction model based on the data of premature infants and to apply the data in our study as external validation to the prediction model proposed by Yuejun Huang et al. to evaluate the predictive ability of both models. METHODS: In total, 397 premature infants were randomly divided into the training set (n = 278) and the testing set (n = 119). Univariate and multivariate logistic analyses were applied to identify potential predictors, and the prediction model was constructed based on the predictors. The area under the curve (AUC) value, the receiver operator characteristic (ROC) curves, and the calibration curves were used to evaluate the predictive performances of prediction models. The data in our study were used in the prediction model proposed by Yuejun Huang et al. as external validation. RESULTS: In the current study, endotracheal intubation [odds ratio (OR) = 10.553, 95% confidence interval (CI): 4.959–22.458], mechanical ventilation (OR = 10.243, 95% CI: 4.811–21.806), asphyxia (OR = 2.614, 95% CI: 1.536–4.447), and antibiotics use (OR = 3.362, 95% CI: 1.454–7.775) were risk factors for late-onset sepsis in preterm infants. The higher birth weight of infants (OR = 0.312, 95% CI: 0.165–0.588) and gestational age were protective factors for late-onset sepsis in preterm infants. The training set was applied for the construction of the models, and the testing set was used to test the diagnostic efficiency of the model. The AUC values of the prediction model were 0.760 in the training set and 0.796 in the testing set. CONCLUSION: The prediction model showed a good predictive ability for late-onset sepsis in preterm infants.