Could AI Help to Spot and Reduce Pregnancy Risk?
Researchers at Mount Sinai – one of the oldest and largest teaching hospitals in the US – are spotlighting two critical areas where artificial intelligence (AI) could reshape pregnancy risk assessment.
Congenital heart defects and placenta accreta spectrum both carry high morbidity rates, substantial costs and significant resource demands.
Through machine learning models trained on extensive electronic medical record and imaging datasets, the Mount Sinai team aims to identify risk earlier in the care pathway, potentially before conception for placenta accreta and during routine mid-trimester ultrasounds for foetal cardiac abnormalities.
Specialists from the institution presented an AI-supported screening tool for severe congenital heart defects and machine learning algorithms that predict placenta accreta spectrum using pre-pregnancy EMR data at the 2026 SMFM Annual Pregnancy Meeting in February 2026. The research forms part of a wider body of work covering social vulnerability, gun violence exposure and labour management.
Predicting placenta accreta before conception
Placenta accreta spectrum represents a dangerous complication in which the placenta embeds excessively into the uterine wall, creating high-risk delivery scenarios that demand intensive resources.
Mount Sinai's case-control analysis examined 118,890 deliveries spanning 2013 to 2023, finding PAS in 0.23% of cases. Despite its rarity, the condition presents considerable maternal morbidity and mortality risks.
A notable discovery for digital health and population management providers centres on pre-pregnancy anaemia, which emerged as a previously unidentified PAS risk factor.
The model's SHAP (SHapley Additive exPlanations) analysis identified anaemia alongside established factors including advanced maternal age, previous caesarean delivery, dilation and curettage procedures, gynaecological surgery and obstetric complications as leading predictors.
Given that anaemia can be modified through intervention, this finding offers a tangible action point. Preconception screening initiatives could identify patients whose EMR profiles indicate elevated PAS risk and direct them towards nutritional support, haematology services or preconception counselling before pregnancy occurs.
Such pathways could reduce high-cost emergency deliveries through planned care at tertiary facilities equipped with appropriate surgical and transfusion resources. The research team suggests that early identification enables proactive management strategies that improve both maternal outcomes and healthcare system efficiency.
Machine learning models and performance
Mount Sinai's research team trained several machine learning algorithms on pre-pregnancy EMR information, encompassing demographics, obstetric and surgical histories, vital signs, laboratory results, billing codes and provider documentation.
An XGBoost model achieved an area under the ROC (Receiver Operating Characteristic) curve of 0.86, surpassing logistic regression's 0.76 performance.
Random forest models delivered 91% sensitivity, whilst logistic regression reached 91% specificity, demonstrating the balance health systems must strike when optimising models for recall against false positive rates.
The algorithms process thousands of data points from patient records to generate risk scores that can inform clinical decision-making. This computational approach identifies patterns that may not be immediately apparent through traditional risk assessment methods.
Enhancing foetal cardiac screening
Mount Sinai West has implemented FDA (Food and Drug Administration)-cleared AI technology from BrightHeart to strengthen foetal ultrasound screening for congenital heart defects across multiple centres.
A study involving 200 anonymised second-trimester ultrasounds from 11 medical centres across two countries showed that AI-assisted workflows improved major congenital heart defect detection to above 97%, whilst reducing reading time by 18% and increasing confidence scores by 19%.
Fourteen clinicians, comprising seven obstetrician-gynaecologists and seven maternal-foetal medicine specialists, assessed each examination with and without AI support.
The technology is currently undergoing evaluation within a live prenatal diagnostic centre environment, where the AI-supported system highlights concerning findings for severe congenital heart defects as part of routine foetal cardiac screening protocols.
