| アブストラクト | BACKGROUND: Antibody-drug conjugates (ADCs) are increasingly used across breast cancer subtypes, but their postmarketing safety-reporting patterns differ across agents and are difficult to interpret across spontaneous-reporting and clinical settings. We integrated breast-cancer-restricted FDA Adverse Event Reporting System (FAERS) analyses, temporal machine-learning prioritization, and an independent institutional cohort to characterize drug-specific safety patterns across complementary evidence layers. METHODS: FAERS quarterly files from 2013Q1 through 2025Q4 were deduplicated and analyzed at the report level. The primary dataset comprised 13,529 breast-cancer primary-suspect reports involving trastuzumab emtansine (T-DM1), trastuzumab deruxtecan (T-DXd), or sacituzumab govitecan; 20,526 all-indication reports were retained for supportive sensitivity analysis. Analyses included clinically reviewed preferred-term disproportionality, 11 prespecified adverse events of special interest (AESIs), intraclass adjusted reporting odds, reported time to onset, and temporally validated exploratory machine-learning models for report-level AESI prioritization. An independent retrospective cohort of 105 patients was analyzed descriptively for clinical contextualization. RESULTS: The primary dataset included 3,962 T-DM1, 6,452 T-DXd, and 3,115 sacituzumab govitecan reports. Compared with T-DM1, T-DXd showed higher adjusted reporting odds for interstitial lung disease (ILD)/pneumonitis (adjusted reporting odds ratio [aROR], 3.23; 95% CI, 2.55-4.09) and gastrointestinal toxicity (aROR, 2.45; 95% CI, 2.07-2.90), whereas sacituzumab govitecan showed higher reporting odds for hematologic (aROR, 1.95; 95% CI, 1.67-2.28) and gastrointestinal toxicity (aROR, 2.91; 95% CI, 2.44-3.46). T-DM1 showed higher hepatobiliary reporting odds than both comparators. Temporal-validation AUROC values ranged from 0.593 to 0.772. In the institutional cohort, hepatobiliary toxicity was most frequent with T-DM1 (32.3%), ILD/pneumonitis with T-DXd (12.0%), and hematologic toxicity with sacituzumab govitecan (66.7%). CONCLUSION: The three ADCs showed distinct breast-cancer FAERS safety-reporting patterns. Machine learning provided exploratory report-level prioritization, while institutional observations provided clinical context. These findings support pharmacovigilance signal prioritization but should not be interpreted as incidence, comparative clinical risk, causal effects, or patient-level toxicity prediction. |
|---|
| 投稿者 | Tang, Yuhan; Liu, Shaochun; Zhu, Yingze; Fan, Linlin; Han, Xiaoxi; Ma, Wenjie; Zhang, Haolu; Sui, Zhiqi; Pang, Hui; Zhao, Wenhui |
|---|
| 組織名 | Department of Oncology, Harbin Medical University Cancer Hospital, Harbin, China.;Department of General Surgery, West China Ya'an Hospital, Sichuan University,;Ya'an, China. |
|---|
