Machine learning-enabled prediction of ART pregnancy outcomes: a systematic review and meta-analysis

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Biostatistics

Background:

• Assisted reproductive technology (ART) success is crucial for infertile couples.
• Predictive models can optimize ART outcomes, but their accuracy needs evaluation.
• Machine learning (ML) offers potential for improved prediction accuracy.

Purpose of the Study:

• To systematically evaluate the diagnostic accuracy of ML models for predicting pregnancy outcomes post-ART.
• To assess the methodological quality of studies developing or validating these ML models.
• To identify factors influencing the performance of ML prediction models in ART.

Main Methods:

• A comprehensive systematic search of multiple databases (PubMed, Embase, etc.) was conducted up to July 2026.
• Diagnostic meta-analysis was performed on eligible studies using random-effects models to pool sensitivity, specificity, and DOR.
• Risk of bias was assessed using the PROBAST tool.

Main Results:

• Twenty studies were included, with 14 contributing to the meta-analysis.
• Pooled sensitivity was 0.737 and specificity was 0.789, indicating moderate diagnostic accuracy.
• Substantial heterogeneity was observed, and 95% of studies had high or unclear risk of bias.

Conclusions:

• ML models demonstrate moderate diagnostic accuracy for ART pregnancy outcomes.
• The current evidence is limited by significant heterogeneity and methodological quality concerns.
• Future research should adhere to reporting standards and prioritize external validation for clinical implementation.