Comparative safety signal prioritization of CDK4/6 inhibitors in breast cancer: a FAERS pharmacovigilance, machine learning, and clinical contextualization study.

Background: Cyclin-dependent kinase 4 and 6 (CDK4/6) inhibitors are standard endocrine-based treatments for hormone receptor-positive (HR-positive), HER2-negative breast cancer, but drug-specific adverse event reporting patterns differ among agents. This study prioritized adverse events of special interest (AESIs) for palbociclib, abemaciclib, and ribociclib using the Food and Drug Administration Adverse Event Reporting System (FAERS), exploratory machine learning (ML), and institutional clinical contextualization. Methods: Primary suspect FAERS reports from 2015Q1 to 2025Q4 were analyzed at the report level. Reports were mapped to eight AESI domains and assessed using disproportionality measures, intraclass-adjusted reporting odds, reported time to onset, and exploratory ML signal prioritization. ML models were developed using data from 2015Q1 to 2022Q4 and temporally validated using data from 2023Q1 to 2025Q4. Consecutive patients treated at a Chinese tertiary cancer center between 6 January 2023 and 29 December 2024 were reviewed. Results: FAERS included 115,971 class-wide primary suspect reports and 67,314 reports with breast cancer recorded as the indication. The three-agent ML dataset included 115,771 reports, with 76,581 for model development and 39,190 for temporal validation. Palbociclib was characterized mainly by hematologic reporting; abemaciclib by gastrointestinal, thrombotic, pulmonary, and renal- or creatinine-related prioritization; and ribociclib by QT interval and hepatobiliary prioritization. The institutional cohort included 82 patients and contextualized these AESIs through treatment-emergent events, grade 3 or higher toxicity, dose modification, and temporal clinical consequence markers. Conclusion: The three agents showed distinct report-level safety reporting profiles. These findings prioritize drug-specific AESIs for pharmacovigilance review and safety hypothesis generation; however, incidence, causal effects, and patient-level risk estimation require denominator-based clinical datasets.