AUTHOR=Li Yan , Zhang Zhihui , Mo Yan , Wei Qiufen , Jing Lianfang , Li Wei , Luo Mengmeng , Zou Linxia , Liu Xin , Meng Danhua , Shi Yuan TITLE=A prediction model for short-term neurodevelopmental impairment in preterm infants with gestational age less than 32 weeks JOURNAL=Frontiers in Neuroscience VOLUME=17 YEAR=2023 URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2023.1166800 DOI=10.3389/fnins.2023.1166800 ISSN=1662-453X ABSTRACT=Introduction

Early identification and intervention of neurodevelopmental impairment in preterm infants may significantly improve their outcomes. This study aimed to build a prediction model for short-term neurodevelopmental impairment in preterm infants using machine learning method.

Methods

Preterm infants with gestational age  < 32 weeks who were hospitalized in The Maternal and Child Health Hospital of Guangxi Zhuang Autonomous Region, and were followed-up to 18 months corrected age were included to build the prediction model. The training set and test set are divided according to 8:2 randomly by Microsoft Excel. We firstly established a logistic regression model to screen out the indicators that have a significant effect on predicting neurodevelopmental impairment. The normalized weights of each indicator were obtained by building a Support Vector Machine, in order to measure the importance of each predictor, then the dimension of the indicators was further reduced by principal component analysis methods. Both discrimination and calibration were assessed with a bootstrap of 505 resamples.

Results

In total, 387 eligible cases were collected, 78 were randomly selected for external validation. Multivariate logistic regression demonstrated that gestational age(p = 0.0004), extrauterine growth restriction (p = 0.0367), vaginal delivery (p = 0.0009), and hyperbilirubinemia (0.0015) were more important to predict the occurrence of neurodevelopmental impairment in preterm infants. The Support Vector Machine had an area under the curve of 0.9800 on the training set. The results of the model were exported based on 10-fold cross-validation. In addition, the area under the curve on the test set is 0.70. The external validation proves the reliability of the prediction model.

Conclusion

A support vector machine based on perinatal factors was developed to predict the occurrence of neurodevelopmental impairment in preterm infants with gestational age  < 32 weeks. The prediction model provides clinicians with an accurate and effective tool for the prevention and early intervention of neurodevelopmental impairment in this population.