AI analysis reveals key risk factors for severe pregnancy outcomes

AI analysis reveals key risk factors for severe pregnancy outcomes | Image Credit: © Yakobchuk Olena - © Yakobchuk Olena - stock.adobe.com.

An artificial intelligence (AI)-based analysis has identified risk factor combinations associated with severe adverse pregnancy outcomes such as stillbirth, according to a recent study published in BMC Pregnancy and Childbirth.¹

Results indicated up to a 10-fold difference in risk among infants treated with current clinical guidelines. According to Nathan Blue, MD, senior author of the study, this is an important step toward more personalized risk assessment.

“AI models can essentially estimate a risk that is specific to a given person’s context, and they can do it transparently and reproducibly, which is what our brains can’t do,” said Blue. “This kind of ability would be transformational across our field.”

Investigators conducted the study to develop a probabilistic graphical model (PGM) for identifying probabilities of composite perinatal morbidity and mortality in fetal growth restriction (FGR), which is the leading risk factor for stillbirth.² Data was obtained from the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (nuMoM2b) observational cohort.

The cohort included over 10,000 nulliparous women recruited between 6- and 13-weeks’ gestation with no prior pregnancies reaching 20-weeks’ gestation. Patients aged under 13 years, with 3 or more prior pregnancy losses, with an already evident fatal fetal malformation at enrollment, using a donor oocyte, or with multifetal reduction were excluded.

Participants underwent 4 study visits to complete detailed interviews, questionnaires, research ultrasounds, maternal biometric measurements, and biospecimen collection. Relevant data collected included medical history, medication use, family medical history, and ultrasound estimated fetal weight (EFW).

Psychosocial domains were also assessed using validated instruments. Maternal interviews and medical record abstraction were performed to collect data about obstetric, maternal, and neonatal outcomes. All nuMoM2b participants with a delivery at 20-weeks’ gestation or later were included in the current analysis.

Composite perinatal morbidity and mortality was reported as the primary outcome. Morbidity included stillbirth, neonatal death, respiratory distress syndrome, needing mechanical ventilation, confirmed sepsis, grade 3 or 4 intraventricular hemorrhage, necrotizing enterocolitis, neonatal seizures, and neonatal intensive care unit
admission over 1 week.

There were 9,558 participants included in the final analysis. Maternal variables selected for the PGM included pre-existing diabetes, progesterone use, hypertensive disorders of pregnancy, gestational age at birth, urgent cesarean, and preterm premature rupture of membranes.

Neonatal variables selected include congenital anomaly presence, sex, 5-minute Apgar score under 7, and EFW percentile at visit 3. Low missingness was reported for all variables. Most variables were linked to an increased risk of composite perinatal morbidity, except for term birth which was linked to a decreased risk.

The receiver-operating characteristics area under the curve (AUC) for the PGM was 0.83 when applied to the validation cohort, indicating good performance. This was similar to the logistic regression (LR) model with an AUC of 0.82.

The PGM can evaluate the complete joint probability distribution of the graphical model, giving them an advantage over the LR model. This allowed the PGM to identify FGR scenarios with varying risk relationships. Most scenarios highlighted a link between the EFW percentile category and a common pattern of risk distribution.

These results indicated strong performance form the AI model in providing context-specific risk estimates across multiple FGR scenarios. Investigators concluded “this represents an important proof of concept and demonstration of the potential for PGMs to refine risk estimation in obstetrics.”

References

1. AI-based pregnancy analysis discovers previously unknown warning signs for stillbirth and newborn complications. University of Utah Health. January 29, 2025. Accessed February 11, 2025. https://www.eurekalert.org/news-releases/1071815
2. Zimmerman RM, Hernandez EJ, Yandell M, et al. AI-based analysis of fetal growth restriction in a prospective obstetric cohort quantifies compound risks for perinatal morbidity and mortality and identifies previously unrecognized high risk clinical scenarios. BMC Pregnancy and Childbirth. 2025;25(80). doi:10.1186/s12884-024-07095-6