Machine Learning Approach for Preterm Birth Prediction Using Health Records: Systematic Review

Area of Science:

• Medical Informatics
• Computational Biology
• Public Health

Background:

• Preterm birth (PTB) is a major global health issue, causing millions of deaths and affecting millions of children annually.
• Current PTB prediction methods are unreliable, failing to detect over 50% of cases.
• Machine learning (ML) offers a potential complementary approach for PTB prediction using health records (HRs).

Purpose of the Study:

• To systematically review existing literature on ML-based PTB prediction using HR data.
• To assess the characteristics and performance of ML models applied to PTB prediction.

Main Methods:

• Systematic literature review adhering to PRISMA guidelines.
• Comprehensive search across 7 bibliographic databases until May 15, 2021.
• Quality assessment and extraction of descriptive data on datasets, ML processes, and model performance.

Main Results:

• 13 studies met inclusion criteria from 732 screened papers.
• Included studies utilized datasets ranging from 274 to 1.4 million individuals, with data spanning 1 to 11 years (1988-2018).
• Model performance was reported using metrics like accuracy, sensitivity, specificity, and AUC, but methodological details were often insufficient.

Conclusions:

• ML models demonstrate potential for PTB prediction using HR data.
• Lack of justification for evaluation metrics, software, data characteristics, feature selection, and data management compromises model reliability and validity.
• Future studies should compare ML models with conventional methods on identical datasets to clarify ML's utility.