Comparison of Multivariable Logistic Regression and Other Machine Learning Algorithms for Prognostic Prediction...

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Biostatistics

Background:

• Pregnancy care predictions are complex due to multifactorial interactions.
• Single predictors or methods often fail to accurately forecast pregnancy outcomes.

Purpose of the Study:

• To systematically review and compare the predictive performance of logistic regression (LR) against other machine learning (ML) algorithms in pregnancy care.
• To evaluate ML models for developing or validating multivariable prognostic prediction models to aid clinical decision-making.

Main Methods:

• A systematic review and meta-analysis of research articles from major databases (MEDLINE, Scopus, Web of Science, Google Scholar).
• Studies were assessed for risk of bias (ROB) and synthesized following PRISMA guidelines.
• Compared non-LR ML models against LR models using logit area under the receiver operating characteristic curve (AUROC) differences.

Main Results:

• 142 studies were included in the review, with 62 in the meta-analysis. Most models used LR (64.8%).
• Only 16.9% of studies had low ROB.
• Random forest and gradient boosting algorithms significantly outperformed LR in specific predictions like preterm delivery, pre-eclampsia, cesarean section, and gestational diabetes.

Conclusions:

• Prediction models utilizing random forest and gradient boosting demonstrated strong performance, not exclusively LR models.
• Reanalysis of existing LR models is recommended, comparing them against ML algorithms adhering to standard guidelines for improved pregnancy outcome prediction.