Today, almost 80% of women take at least one medication during pregnancy, but often these treatment decisions are made without the quantity of safety data we are accustomed to having. Researchers are now increasingly turning to AI and advanced data analytics to help fill that gap.

Rather than relying on traditional clinical trials, scientists are analysing large healthcare datasets, prescription records, and real-world patient outcomes to identify and evaluate patterns between drug exposure and maternal or foetal health outcomes. The goal is not to replace clinical research, but to generate evidence where none previously existed.

Projects such as BOOST-HP and BIONIC are combining machine learning with causal inference, a statistical method used to distinguish genuine causality from correlation. However, it is important to note that while AI can identify patterns, it cannot reliably determine causality solely from data. Understanding which medicine causes a particular outcome still requires a much deeper understanding of epidemiology and clinical context.

Researchers are therefore approaching this work with caution. Rather than deploying opaque “black box” models, many in the field favour more interpretable systems that allow scientists to understand exactly how conclusions are reached. Almut Winterstein, a lead investigator on the BOOST-HP project, argues that AI must work alongside established scientific principles rather than replace them. Without transparency and causal reasoning, there is a risk that these promising AI approaches could generate misleading conclusions and undermine trust in the evidence they produce. If left unchecked, this could have serious health implications for both mothers and babies.

The main challenge remains data quality. Effective AI systems require large, diverse datasets that reflect different populations, healthcare contexts, and social determinants of health. Without them, models risk introducing new biases while attempting to solve existing ones.

For the drug safety industry, this innovation has the potential to address longstanding evidence gaps that traditional clinical methods have struggled to fill. If successful, these approaches could ultimately lead to safer prescribing decisions, more informed clinical care, and better outcomes for mothers and babies alike.

“It is important for people who are very strongly on the AI side to collaborate with people who come from the causal inference epidemiology side and put those two together“. Professor Almut Winterstein, Principal Investigator at BOOST-HP