Pregnant women have been systematically locked out of the clinical trials that determine whether drugs are safe for them to take. For nearly 50 years, this exclusion has created a dangerous blind spot in medicine, leaving doctors with minimal evidence to guide medication decisions for one of the most vulnerable populations. Now, two machine learning projects are using artificial intelligence to fill that gap by analyzing vast medication datasets and uncovering patterns that human researchers alone could miss.
The problem runs deep. Only 4% of clinical trials conducted over the last decade have included pregnant women as participants—a legacy that traces back to 1977, when the US Food and Drug Administration discouraged pharmaceutical companies from enrolling pregnant women or women capable of becoming pregnant in early-stage drug testing. The reasoning was protective: avoid potential harm to a developing fetus. But the result was the opposite. By excluding pregnant women from the evidence-gathering process, medicine created uncertainty precisely where clarity was most needed. Decades of efforts to determine medication safety for pregnant and breastfeeding women have fallen short in practice, leaving an evidence gap that touches millions of women making real decisions about their health during pregnancy.
The BOOST-HP and BIONIC projects represent a promising new approach. BOOST-HP uses a tree-based approach to data mining, systematically sorting through medication exposure records and health outcomes to identify potential connections. BIONIC combines causal inference with machine learning, allowing researchers to go beyond correlation and probe the actual causes linking medications to outcomes. In both cases, machine learning does the computational heavy lifting—analyzing datasets so large that human researchers would be overwhelmed—and then flagging possible safety signals for further investigation.
Yet this technological solution carries its own ethical weight. Cristina Longo, the principal investigator of the BIONIC study, emphasizes that more data alone won't solve the problem. The researchers also need what she calls "a healthy dose of caution." That's where transparency becomes crucial. Almut G. Winterstein, a principal researcher on BOOST-HP, and her team deliberately chose machine learning models whose decision-making pathways can be traced and understood by humans. They explicitly rejected "black box" systems—the kind of AI that operates invisibly, with internal workings that are opaque or obfuscated. Using such a model would risk missing crucial epidemiological errors, errors that could have real consequences for real patients.
The path forward lies in marrying better technology with better data. The promise is substantial: more thoughtful design of machine learning models, combined with larger and more comprehensive datasets drawn from actual pregnancy experiences, holds genuine potential to finally close the evidence gap that medicine has ignored for half a century. For pregnant women, that shift from uncertainty to evidence-based care could change everything.