Machine learning can help close pregnancy drug safety gaps

JMIR Publications today released a report on developments in the evidence gap in drug safety during pregnancy in its News and Perspectives section. In "How Machine Learning Can Help Close Evidence Gaps for Drug Safety in Pregnant Women", health writer Michelle Falci interviews the principal investigators of two projects which use machine learning to analyze large datasets of medication exposure and outcomes, then identify and evaluate possible links.

Pregnant participants excluded from clinical trials

Medical research has a serious problem with underrepresentation, Falci reports; only 4% of clinical trials over the last decade included pregnant women as participants. This trend dates back to 1977, when the US Food and Drug Administration recommended not to include pregnant women, or women capable of becoming pregnant, in phase 1 and 2 clinical trials, resulting in a gap in evidence on drug safety for pregnant women (and contributing to a broader underrepresentation of female participants in research). Though efforts have been made to determine medication safety for pregnant and breastfeeding women, these have fallen short in practice.

Closing the gap with machine learning

Falci gets a closer look at two novel efforts to close this evidence gap: the BOOST-HP project, which uses a tree-based approach to data mining; and the BIONIC study, which combines causal inference and machine learning. Each approach uses machine learning to do the heavy lifting of analyzing large datasets, allowing the researchers to monitor and estimate the potential causal links. 

However, this kind of AI-assisted research will ideally benefit from more data, according to BIONIC study leader Cristina Longo-plus, a healthy dose of caution. Transparency is key, as Almut G. Winterstein, a principal researcher on the BOOST-HP project, notes: she and her team use an AI model which allows them to trace the decision pathways leading to the models' evaluations. If they were to use a 'black box' model-a system whose internal workings are opaque or obfuscated-they would run the risk of missing crucial epidemiological errors. Further thoughtful design of machine learning models, as well as a larger and more comprehensive dataset, nonetheless holds a great deal of promise for closing this evidence gap.