AI can predict and prevent high-risk pregnancy complications

Artificial intelligence (AI) is emerging as a powerful ally in addressing one of the most persistent global health challenges: adverse pregnancy outcomes (APOs). These complications, ranging from preterm birth and gestational diabetes to hypertensive disorders and fetal growth restrictions, remain leading causes of maternal and neonatal morbidity and mortality worldwide.

A study titled "Artificial Intelligence’s Role in Improving Adverse Pregnancy Outcomes: A Scoping Review and Consideration of Ethical Issues", published in the Journal of Clinical Medicine, offers a sweeping scoping review of how AI has been applied across obstetrics and maternal health to mitigate APOs. The authors catalogued 131 primary studies covering multiple continents, revealing that AI-driven approaches have successfully improved diagnostic accuracy, risk prediction, treatment planning, and care delivery for pregnant individuals at risk.

AI models, particularly those built on machine learning and deep learning techniques, are being used to process complex data such as maternal vital signs, biochemical markers, imaging results, and electronic health records. By detecting patterns that elude traditional statistical tools, these systems can identify high-risk pregnancies earlier and with greater precision, allowing for timely interventions that can dramatically improve outcomes for both mother and fetus.

Where has AI been most effective and where do gaps remain?

The study finds that AI systems have been most effectively applied in identifying risks for preeclampsia, gestational diabetes, preterm labor, and stillbirth. Many of the reviewed models demonstrated excellent predictive performance, with area under the curve (AUC) values consistently above 0.80 - indicative of high diagnostic accuracy.

Ultrasound analysis and imaging-based diagnostics were among the most frequent use cases. Convolutional neural networks (CNNs) have enhanced fetal anomaly detection and gestational age estimation, enabling clinicians to make better-informed decisions. Moreover, AI has improved real-time fetal monitoring through cardiotocography analysis, detecting abnormal patterns that signal fetal distress.

Despite these successes, the study highlights considerable disparities in research distribution. High-income countries dominate the landscape of AI in maternal health, while low- and middle-income regions, where the burden of APOs is often greatest, remain underrepresented. This geographic skew limits the generalizability of existing models and perpetuates inequities in maternal care.

Another critical gap lies in external validation. Many AI tools reviewed in the study have been trained and tested within single-site datasets, making them vulnerable to performance drops when deployed in diverse clinical settings. The lack of real-world implementation studies further underscores the need for rigorous translational research before AI tools can be widely adopted in maternal care settings.

What are the ethical challenges of AI in obstetrics?

While the clinical potential of AI in reducing APOs is compelling, the study also explores the ethical tensions introduced by this emerging technology. One of the most pressing concerns is data privacy. AI tools require access to large volumes of sensitive health data, including genetic, biometric, and behavioral information, which raises questions about consent, security, and surveillance.

Bias in data and model training is another concern. If AI systems are trained on datasets that reflect existing disparities, such as unequal access to care or socioeconomic determinants of health, they may amplify those biases in clinical decision-making. This could lead to misdiagnosis, overtreatment, or neglect of already vulnerable populations.

The issue of explainability also looms large. Many AI models, especially deep learning systems, operate as “black boxes,” offering little transparency into how decisions are made. In maternal health, where trust between patient and provider is paramount, opaque AI-generated recommendations may erode clinician confidence and patient autonomy.

To address these challenges, the study urges the development of transparent, fair, and accountable AI systems. It advocates for regulatory oversight, stakeholder engagement, and interdisciplinary collaboration to ensure that ethical standards are embedded at every stage of AI development and deployment.

Moreover, the authors emphasize the importance of involving pregnant individuals in discussions about how their data is used and how AI-based decisions are communicated. This participatory approach aligns with broader trends in patient-centered care and offers a path toward more equitable and ethically sound AI integration.