Machine learning model for the prediction of postoperative acute kidney injury (AKI) in neonates undergoing digestive...

The aim of our study is to determine the main predictors of postoperative AKI in neonates using machine learning models compared with the logistic regression model.

Demographic and pre-operative data of 742 neonates who underwent digestive surgery were collected during this retrospective study. Data were randomly split into training and testing datasets at a ratio of 7:3. The training dataset was balanced. Key features with predictive value for postoperative AKI in neonates were selected by univariate analysis. Then, those features were fed into the four different machine learning prediction models. Each model was evaluated, and the best-performing model was selected for interpreting the importance of each variable.

A total of 742 neonates who underwent digestive surgery were analyzed. The incidence of neonatal AKI after digestive surgery is 7.8%. During the machine learning analysis, we found that the logistic regression model outperformed other models, and five features were identified (IVF, Apgar score at 5 min, mean body temperature during surgery, pre-operative SCr, and postoperative UN). These five candidates were used to develop a nomogram, with the AUC in the training and the testing cohorts being 0.76 (0.69-0.83) and 0.77 (0.66-0.88), respectively. Calibration curve, decision curve analysis (DCA), and clinical impact curve (CIC) indicated that the logistic regression model also had good predictive performance and clinical application value.

The established machine learning model can predict the risk of postoperative AKI in neonates, and provide a reliable interpretation for high-risk factors identification, thus further assisting clinical decision-making.