AI-Driven Prediction of Drug Safety in Pregnancy: A Machine Learning Approach

Evaluating drug safety during pregnancy remains an ongoing clinical and pharmacological challenge due to ethical, practical, and regulatory barriers, resulting in scarce human clinical trial data. Consequently, healthcare providers must frequently rely on limited observational data and incomplete safety profiles when prescribing medications, especially psychiatric and neurological drugs, whose discontinuation could lead to significant maternal health risks. This research addresses these critical gaps by developing an advanced, machine learning (ML)-based predictive model specifically aimed at assessing and classifying the safety of psychiatric and neurological medications during pregnancy. Leveraging extensive, synthesized, and publicly available datasets including the FDA Adverse Event Reporting System (FAERS) and various pregnancy registries, the study utilized a robust methodological pipeline encompassing data preprocessing, exploratory analysis, feature engineering, model training (Random Forest), rigorous model evaluation (including confusion matrices), and visualization-driven insights. The resulting predictive model categorizes medications into three distinct classes: Safe, Potentially Harmful, or Contraindicated. The performance evaluation demonstrated high predictive accuracy across these classifications, with critical influencing features identified as trimester of medication use, drug class (particularly antidepressants), maternal age, and molecular weight. The model’s high interpretability facilitates informed clinical decision-making, significantly enhancing maternal-fetal safety outcomes. This ML-driven predictive tool represents an important advancement in personalized medicine and clinical pharmacology, offering healthcare professionals and regulatory bodies an evidence-based framework for better risk assessment and drug prescribing practices in pregnancy. Future developments include incorporating deep learning techniques for analysing unstructured clinical data, broadening the drug categories studied, and integrating the model into clinical decision-support systems.