Development and evaluation of machine learning training strategies for neonatal mortality prediction using...

Area of Science:

• Global Health
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Neonatal mortality remains a significant global health issue, especially in low- and middle-income countries.
• Technological advancements, including machine learning (ML), offer potential solutions for predicting and preventing neonatal mortality.
• Improving neonatal care through predictive analytics is crucial for reducing infant mortality rates.

Purpose of the Study:

• To evaluate the effectiveness of machine learning (ML) models in predicting neonatal mortality risk.
• To compare the performance of generalized, country-specific, and single-country dataset-derived ML models.
• To identify the optimal ML approach for improving neonatal health outcomes globally.

Main Methods:

• Utilized the Maternal and Neonatal Health Registry (MNHR) dataset from the National Institutes of Health (NIH).
• Included data from 575,664 pregnancies (2010-2016 for training, 2017-2019 for validation).
• Assessed five ML algorithms using World Health Organization-recommended neonatal health indicators.

Main Results:

• The generalized ML model achieved the highest predictive performance with an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.816.
• This indicates that leveraging diverse, multicentric neonatal data enhances predictive accuracy.
• The generalized model outperformed country-specific and single-country dataset models.

Conclusions:

• Generalized machine learning models demonstrate superior predictive capability for neonatal mortality risk.
• Integrating diverse datasets into ML models is vital for improving global neonatal health outcomes.
• Findings support the adoption of generalized ML models in healthcare strategies to reduce neonatal mortality.