| アブストラクト | BACKGROUND: Renal adverse-event reports during immune checkpoint inhibitor (ICI)-VEGFR tyrosine kinase inhibitor (TKI) therapy may combine immune-interstitial, proteinuric, and nonspecific acute kidney injury (AKI) features. We compared their reporting patterns with those of nivolumab plus ipilimumab in renal cell carcinoma (RCC). METHODS: We analyzed FAERS/AERS (2004Q1-2026Q2) and the JADER 202607 release (through 2026Q1), applying latest-version handling, strict indication-drug linkage, author-defined Tier-A reporting labels, regimen-specific common support, and year-stratified Mantel-Haenszel reporting odds ratios (MH-RORs). JADER supplied cross-database directional corroboration. Secondary classification of the rule-derived labels used non-anchor features, with and without regimen indicators, and included temporal evaluation in 2024-2025. Additional analyses assessed sparse cells, terminal periods, standard renal queries, and model ablations. RESULTS: Among 84,369 strict-RCC FAERS reports, 2,307 met Tier-A definitions: 219 immune-interstitial, 1,076 VEGF-axis, 932 nonspecific AKI, and 80 mixed. For pembrolizumab plus axitinib, pembrolizumab plus lenvatinib, and nivolumab plus cabozantinib, VEGF-axis MH-RORs were 5.31 (95% CI 2.43-11.59), 19.48 (9.03-42.01), and 4.59 (2.07-10.19). Proteinuria dominated this signal. Exclusive Tier-A AKI estimates were 1.05, 0.93, and 0.78, with intervals crossing one; this finding was definition-dependent, as a separate narrow acute-renal-failure query yielded 1.56 (1.20-2.02) for pembrolizumab plus lenvatinib. Excluding 2026 retained the primary VEGF-axis direction. JADER estimates were 6.28, 19.27, and 6.16. In a secondary consistency analysis, adding regimen increased temporal macro-AUPRC from 0.606 to 0.638, but immune-interstitial class average precision remained 0.253. CONCLUSIONS: The reproducible finding is a proteinuria-dominant VEGF-axis reporting pattern, not a validated renal mechanism or comparative incidence estimate. Differential surveillance, sparse comparator events, phenotype misclassification, and shared reporting ecology constrain interpretation. The classifier describes internal label consistency and has no established diagnostic use. |
|---|
