Pregnant women take some medications whose effects on the fetus are largely unknown. For these "category C" drugs—which account for about 38 percent of the drugs taken by pregnant women—animal studies are either non-existent or inconclusive regarding the effect on the developing fetus. The authors sought to design a machine-learning method that uses big data housed in clinical records, coupled with information on the chemical properties of the drug, to determine if these drugs were likely to be harmful or safe to the fetus.

To classify the drugs, they employed a machine learning algorithm (i.e., "random forest") that uses information from drugs with known fetal effects, either harmful or safe, to make comparisons with drugs of unknown fetal effect (the category C drugs).

Fifty-seven category C medications were classified as harmful for fetal loss, and 11 were deemed harmful for congenital anomalies. Overall, medications that were predicted to be unsafe included many drugs with documented harmful effects, such as naproxen, ibuprofen and rubella live vaccine. These medications were still listed as category C at the time when the study began because the FDA determined that not enough research had been done to confirm their fetal effect. The authors also identified several novel drugs, such as the anti-psychotic haloperidol, that increased the risk of fetal loss.

This study is crucial, as no FDA recommendation currently exists for category C drugs with regard to fetal safety. The team’s approach shows the probability that a drug is either harmful or safe for the fetus, given chemo-informatics methods and direct observation from the clinical records. Some findings confirm reports by other researchers, while some require further studies to validate them in other datasets.