Objective:

To synthesize current evidence on AI applications for predicting adverse pregnancy outcomes and examine sources of algorithmic bias.

Approach:

Key Findings:

• AI shows potential for improving early detection and management of adverse pregnancy outcomes, including preeclampsia, preterm birth, gestational diabetes, and fetal growth restriction, with predictive performance (AUROC values from 0.73 to 0.97).
• Eight key sources of algorithmic bias identified: sampling bias, measurement bias, algorithmic bias, temporal bias, selection bias, labelling bias, deployment context bias, and access bias.
• Limited external validation across diverse populations and absence of prospective clinical impact trials constrain current evidence.

Interpretation:

AI applications in maternal healthcare must address algorithmic bias and ensure equitable implementation.

Limitations:

• Heterogeneity in predictive performance across studies due to differences in datasets and model architectures.
• Lack of rigorous multisite validation and cost-effectiveness analyses.
• Evolving regulatory frameworks governing AI accountability.

Conclusion:

Achieving equitable implementation of AI in maternal health requires transparency, accountability, and health equity throughout the AI development and deployment lifecycle.

Attribution Notice

This content is an AI-generated, fully rewritten summary based on a published scholarly article. It does not reproduce the original text and is not a substitute for the original publication. Readers are encouraged to consult the source for full context, data, and methodology.