Black women in America face pregnancy-related mortality rates more than twice the national average, dying at a rate of 49.5 per 100,000 live births compared to 22.3 deaths overall. A new study analyzing data from nearly 191,000 Medicaid-enrolled pregnant women reveals how incorporating social factors into predictive models could help address these persistent inequities.
Researchers at the University of Pennsylvania developed machine learning algorithms to identify high-risk pregnancies an average of 55 days before traditional clinical warning signs appear. The initial model, using only demographic and clinical data, demonstrated troubling bias. It correctly identified 73.0% of high-risk White patients but only 71.5% of Black patients, perpetuating the same disparities embedded in the healthcare system.
When researchers added social determinants of health—including healthcare provider availability, distance to care, and community infrastructure—the racial gap disappeared entirely. Both Black and White patients were identified with 81% sensitivity, while maintaining high accuracy overall.
The implications are significant. Simulations suggest that improving modifiable social factors like maternal healthcare workforce availability could reduce adverse pregnancy outcomes by nearly 32%. Black women would see the greatest absolute benefit, with projections showing 3,673 fewer adverse events.
The findings underscore how structural inequities, rather than individual clinical risk factors, drive maternal health disparities. As the researchers note, racial outcome differences “reflect societal inequities rather than biological differences.” More than 80% of pregnancy-related deaths are classified as preventable, with Medicaid recipients—who represent nearly half of all US births—bearing a disproportionate burden.
See: “Early detection of high risk pregnancies using clinical and social data to improve health outcomes” (January 21, 2026)Â
