The early prediction of neonatal necrotizing enterocolitis in high-risk newborns based on two medical center clinical...

Background: Early identification and timely preventive interventions play an essential role for improving the prognosis of newborns with necrotizing enterocolitis (NEC). Thus, establishing a novel and simple prediction model is of great clinical significance. Methods: The clinical data of 143 NEC neonates in the Zhujiang Hospital of Southern Medical University from October 2010 to October 2022 were collected, whereas 429 non-NEC patients in the same period were allocated to the control group by random sampling. Afterward, all participants were randomly divided into a training group (70%) and a testing group (30%). Then, five machine learning (ML) algorithms and classical logistic regression models were established, combining relevant clinical features and laboratory results. The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity of various models were compared to evaluate the performance of each model. Ten-folds cross-validation was used to find the best hyperparameters for each model. Decision curve analysis (DCA) was further used to evaluate the performance of the established models for clinical applications and create a column-line graph, ranking the feature importance in models by SHapely Additive exPlanation (SHAP). The column plots were calibrated using calibration curves. Additionally, the established model was validated in time series analysis and another medical center. Results: Six important features were included for modeling: days of age (odds ratio [OR] = 1.16; 95% confidence interval [CI]: 1.08-1.25; p = 0.001), gestational age (OR = 0.77; 95% CI: 0.62-0.96; p = 0.018), eosinophil count (EOS) (OR = 3.78; 95% CI: 1.74-8.19; p < 0.001), hemoglobin (HB) (OR = 0.98; 95% CI: 0.97-1.00; p = 0.008), platelet distribution width (PDW) (OR = 1.18; 95% CI: 1.05-1.33; p = 0.004), and high-sensitivity C-reactive protein (HSCRP) (OR = 1.03; 95% CI: 1.01-1.06; p = 0.013). While the logistic regression model achieved an AUC of 0.904, accuracy of 0.865, sensitivity of 0.786, F1-score of 0.742, and a Brier score of 0.1009 in the training group, the AUCs for the five ML models ranged from 0.806 to 0.960. Among these models, the LightGBM model performed the best, providing an AUC of 0.960, accuracy of 0.858, sensitivity of 0.970, F1-score of 0.775, and a Brier score of 0.071. Conclusion: The LightGBM ML model can effectively identify neonatal patients at higher risk of NEC based on the day of age, gestational age, EOS, and HB, PDW, and HSCRP levels. Thus, this model is useful for assisting clinical decision-making.