Machine Learning Prediction Models for Neurodevelopmental Outcome After Preterm Birth: A Scoping Review and New...

Area of Science:

• Neonatal Medicine
• Artificial Intelligence
• Developmental Neuroscience

Background:

• Predicting neurodevelopmental outcomes in preterm infants is crucial for neonatal care.
• Conventional statistical models have limitations in accuracy for outcome prediction.
• Machine learning (ML) offers potential for complex outcome prediction in this population.

Purpose of the Study:

• To review current ML applications for predicting neurodevelopmental outcomes in preterm infants.
• To assess the quality of existing ML models for this purpose.
• To provide guidance for future ML model development in this field.

Main Methods:

• A systematic literature search was conducted using PubMed.
• Studies included focused on ML techniques for predicting neurodevelopmental outcomes in preterm infants using neonatal predictors.
• Data extraction and quality assessment were performed by two independent reviewers.

Main Results:

• Fourteen studies were included, primarily focusing on very preterm infants and outcomes before age 3.
• MRI-based predictors and techniques like linear regression and neural networks were common.
• No studies fully met all quality criteria, but those with less inflated performance showed promising predictive values (AUC up to 0.86, R2 up to 91%).

Conclusions:

• Studies demonstrating less inflated prediction results are the most encouraging.
• An evaluation framework is proposed to enhance the quality of future ML models.
• Further research is needed to improve the reliability and generalizability of ML models for preterm infant neurodevelopmental outcome prediction.