Development of A Machine Learning Algorithm to Classify Drugs Of Unknown Fetal Effect

Area of Science:

• Pharmacology
• Chemoinformatics
• Reproductive Toxicology

Background:

• Many commonly prescribed drugs lack fetal safety recommendations (FDA 'category C').
• Existing data on the fetal effects of these medications is insufficient.
• Accurate classification is vital for maternal and infant health.

Purpose of the Study:

• To classify FDA 'category C' drugs into harmful and safe categories regarding fetal outcomes.
• To assess risks for fetal loss and congenital anomalies.
• To leverage chemoinformatics and electronic health records for drug safety assessment.

Main Methods:

• Utilized chemoinformatics (pharmacological similarity) and empirical data from Electronic Health Records (EHRs).
• Trained a random forest model on large cohorts for fetal loss (14,922 affected, 33,043 unaffected) and congenital anomalies (5,658 affected, 31,240 unaffected).
• Classified drugs based on predicted risk for two distinct adverse pregnancy outcomes.

Main Results:

• Random forest models achieved high accuracy: 91% for fetal loss and 87% for congenital anomalies.
• Identified 57 'category C' drugs as harmful for fetal loss and 11 for congenital anomalies.
• Confirmed known harmful effects (e.g., naproxen, ibuprofen) and identified novel risks (e.g., haloperidol) for fetal loss.

Conclusions:

• The chemoinformatics and EHR-based approach effectively classifies the fetal safety of 'category C' drugs.
• This classification provides essential, actionable information for clinicians and patients.
• The study highlights the need for ongoing evaluation of drug safety during pregnancy.