Analysis and comparative study on the influencing factors of bronchopulmonary dysplasia in premature infants with gestational age ≤32 weeks based on logistic regression and decision tree models.

Background: Bronchopulmonary dysplasia (BPD), as a common and serious complication in premature infants, has a high mortality rate, and the prognosis of most premature infants with BPD is poor. Early prediction and intervention treatment can improve the prognosis of children to a certain extent. This study aimed to explore the risk factors affecting BPD in premature infants with gestational age ≤32 weeks, provide scientific basis for the prevention and control of BPD. Methods: We retrospectively collected clinical data from 478 premature infants with gestational age ≤32 weeks admitted to the Neonatology Department of Tongji University Affiliated Obstetrics and Gynecology Hospital from January 2023 to December 2024. The collected data were divided into a training set and a validation set in an 8:2 ratio, logistic regression and classification decision tree models were used to study the influencing factors of BPD in premature infants with gestational age ≤32 weeks. Receiver operating characteristic (ROC) curves and calibration curves [area under the curve (AUC)] were used to evaluate the effectiveness of the two prediction models. Results: A total of 75 cases (15.69%) of premature infants with gestational age ≤32 weeks developed BPD. Both models showed that ventilator assisted breathing, birth weight, and neonatal respiratory distress syndrome (NRDS) were risk factors for developing BPD in premature infants with gestational age ≤32 weeks. In the logistic regression model, neonatal asphyxia was also included as one of the influencing factors. Ventilator assisted breathing was the main influencing factor for the occurrence of BPD in premature infants aged ≤32 weeks. The comparison of the analysis results of the two models showed that, in the training set, the AUC of the binary logistic regression model was 0.901 [95% confidence interval (CI): 0.868-0.945], with an accuracy of 90.48%, sensitivity of 62.26%, specificity of 95.76%, and F1 score of 0.67. The AUC of the classification decision tree model based on Chi-squared Automatic Interaction Detection (CHAID) algorithm was 0.809 (95% CI: 0.870-0.910), with an accuracy of 88.10%, sensitivity of 43.39%, specificity of 96.47%, and F1 score of 0.53. The area under the ROC curve of the logistic regression analysis model was larger than that of the decision tree model (Z=9.568, P<0.001). In the validation set, the AUC of the logistic regression model was 0.912, and the AUC of the decision tree model was 0.871. Conclusions: Both logistic regression and decision tree models have certain classification prediction value, among which the logistic regression model has better predictive ability than the decision tree model. Birth weight, total duration of mechanical ventilation, asphyxia, and NRDS are independent risk factors affecting the occurrence of BPD in premature infants ≤32 weeks. Clinical medical staff can develop predictive plans based on the predicted results, conduct individualized risk assessments for infants in the early stages, achieve early warning and precise intervention, and improve the overall quality of life of premature infants with gestational age ≤32 weeks.