构建中晚期早产儿严重脑室内出血的预测模型

Development of a Prediction Model for Severe Intraventricular Hemorrhage in Moderate-to-Late Preterm Infants

摘要: 目的：严重脑室内出血(Severe Intraventricular Hemorrhage, sIVH)是早产儿死亡及神经发育障碍的重要原因。现有预测模型多针对胎龄32周以下的极早产儿，对中晚期早产儿存在研究空白。本研究旨在构建机器学习模型，以实现对该群体sIVH的精准早期预测。方法：共纳入1831例中晚期早产儿，收集其出生前及出生后48小时内的临床数据。采用多种算法进行特征选择，比较了16种预测模型的性能，评估并优化最佳模型。结果：在纳入的患儿中，共确诊32例sIVH (1.7%)。最佳模型为朴素贝叶斯模型。高C反应蛋白水平、入院后48小时之内应用儿茶酚胺类药物以及生后立即接受气管插管是sIVH的独立危险因素。结论：本研究构建了一个机器学习预测模型，帮助临床医生在中晚期早产儿生命早期识别sIVH高风险个体，改善预后，降低死亡率。

Abstract: Objective: Severe Intraventricular Hemorrhage (sIVH) is a major cause of death and neurodevelopmental impairment in preterm infants. Existing prediction models primarily focus on very preterm infants under 32 weeks of gestation, leaving a research gap for moderate-to-late preterm infants. This study aims to utilize early multimodal data to construct a machine learning model for accurate and early prediction of sIVH in this specific population. Methods: A total of 1,831 moderate-to-late preterm infants were included in this study. Clinical data collected from the prenatal period and the first 48 hours after birth were analyzed. Multiple algorithms were employed for feature selection, and the predictive performance of 16 different models was compared to identify and optimize the best-performing model. Results: Among the included pediatric patients, a total of 32 cases of sIVH (1.7%) were diagnosed. The optimal model was the Naive Bayes model. High C-reactive protein levels, administration of catecholamine drugs within 48 hours of admission, and receiving tracheal intubation immediately after birth were independent risk factors for sIVH. Conclusions: This study successfully developed an effective machine learning prediction model. It can assist clinicians in early identification of moderate-to-late preterm infants at high risk for sIVH, thereby improving prognosis and reducing mortality.