Artificial intelligence in breast cancer pharmacotherapy: Treatment optimization, medication safety, and the evolving role of oncology pharmacists

Objective To critically review current evidence on artificial intelligence (AI) in breast cancer pharmacotherapy, emphasizing treatment optimization, treatment-response prediction, toxicity, medication safety, pharmacovigilance, clinical decision support, and the evolving role of oncology pharmacists. Data Sources A structured narrative search used PubMed/MEDLINE as the principal database, supplemented by Google Scholar, citation tracking, and reference-list screening. The final PubMed search was executed on September 20, 2026, using a publication cutoff of September 14, 2026, corresponding to the date of the editorial decision, and identified 764 records. The six-day interval reflects the timing of the search rerun after the publication cutoff; publications after September 14 were not included. Evidence relevant to AI-supported breast cancer pharmacotherapy and oncology pharmacy practice was prioritized. Data Summary AI applications include treatment-response prediction, clinical decision support, treatment stratification, toxicity prediction, and pharmacovigilance. Machine-learning models have addressed severe treatment-related toxicity, febrile neutropenia, and cardiotoxicity, while AI-based pharmacovigilance may support detection of adverse-event and polypharmacy patterns. However, much of the evidence remains retrospective or focused on technical performance rather than demonstrated improvement in patient-centered outcomes. Direct prospective evidence evaluating AI-integrated oncology pharmacist interventions remains limited. Conclusions AI applications in breast cancer pharmacotherapy show promising technical and, in selected settings, clinical evidence, but evidence maturity varies substantially across applications. Direct prospective evidence demonstrating additional clinical benefit from integrated AI–oncology pharmacist interventions remains limited. Pharmacist-in-the-loop approaches should therefore be regarded as conceptual and testable models rather than established standards of care.