Evaluation of Regorafenib in Newly Diagnosed and Recurrent Glioblastoma: GBM AGILE Phase II/III Bayesian Randomized Platform Trial

Patrick Y. Wen, MD1 ; Donald A. Berry, PhD2 ; Meredith B. Buxton, PhD3 ; Howard Colman, MD, PhD, FAAN4 ; John de Groot, MD5 ; Michael Lim, MD6 ; Ingo Mellinghoff, MD7 ; James R. Perry, MD, FRCPC, FAAN8 ; Michael Weller, MD9 ; Nicholas A. Blondin, MD10; Omar H. Butt, MD, PhD, MS11 ; Denise M. Damek, MD12; Macarena I. de la Fuente, MD13 ; Jan Drappatz, MD14; Erin Dunbar, MD15; Pierre Giglio, MD16; Emma Viktoria Hyddmark, PhD3 ; Fabio Iwamoto, MD17; Kurt A. Jaeckle, MD18 ; Lyndon Kim, MD19; Heather M. Kling, PhD3 ; Eudocia Q. Lee, MD, MPH20 ; Megan Mantica, MD14 ; Tom Mikkelsen, MD21; Burt Nabors, MD22 ; Herbert B. Newton, MD23 ; Jeffrey J. Olson, MD24 ; David Schiff, MD25 ; Tobias Walbert, MD26 ; Shiao-Pei Weathers, MD27 ; Timothy Cloughesy, MD28 ; and Andrew B. Lassman, MD29 ; for the GBM AGILE Regorafenib Study Group DOI https://doi.org/10.1200/JCO-25-01137

ABSTRACT

PURPOSE GBM AGILE (ClinicalTrials.gov identifier: NCT03970447) is a phase II/III Bayesian adaptive platform registration trial testing multiple arms against a common control; the primary end point is overall survival (OS). Regorafenib, a multikinase inhibitor, showed OS benefit in recurrent (RD) glioblastoma in the phase II REGOMA trial and entered GBM AGILE as the first investigational arm.

METHODS Patient subtypes included in the regorafenib arm of GBM AGILE were newly diagnosed unmethylated (NDU) and RD glioblastoma. Prospective defined sets of subtypes, or arm signatures, were NDU, RD, and all (NDU 1 RD). As the first investigational arm in GBM AGILE, regorafenib was equally randomized to the control arm. Treatment in the control arm is temozolomide 1 radiotherapy (in newly diagnosed) or lomustine (in RD). Efficacy was assessed by OS hazard ratio (HR), arm/control, and demonstrated when the Bayesian probability of benefit (HR <1.00) was ≥98%. Analysis was performed monthly for limited efficacy, which occurs when the Bayesian predictive power is <25% for all signatures, and determines stopping enrollment. Follow-up continued for 12 months after accrual stopped.

RESULTS When the predictive power was <25% in all predefined signatures for regor- afenib, accrual stopped for limited efficacy. The final analysis did not dem- onstrate OS improvement in the regorafenib arm in RD nor NDU glioblastoma. Median HRs were 1.05 (NDU), 1.07 (RD), and 1.07 (all) with final probabilities of benefit (HR <1.00) of 0.421 (NDU), 0.312 (RD), and 0.296 (all). Regorafenib was associated with increased toxicity relative to control.

CONCLUSION GBM AGILE did not show superiority of regorafenib over control in RD (lomustine) or NDU (temozolomide 1 radiotherapy) glioblastoma, yet caused increased toxicities. Regorafenib has been removed from National Compre- hensive Cancer Network guidelines as a treatment option for RD.

PURPOSE

·  GBM AGILE: 2/3상 Bayesian 적응형 플랫폼 임상시험

·  주요 평가변수: 전체생존기간(OS)

·  Regorafenib: 다중 키나아제 억제제

·  REGOMA 2상: 재발성 GBM에서 OS 개선 효과 확인

·  GBM AGILE: Regorafenib이 첫 번째 시험적 치료군으로 참여

METHODS

·  대상 환자: NDU: 새롭게 진단된 비메틸화 GBM RD: 재발성 GBM

·  사전 정의 평가군: NDU / RD / All(NDU+RD)

·  무작위 배정: Regorafenib군과 대조군 동일 비율

·  대조군: NDU → Temozolomide + 방사선치료 RD → Lomustine

·  유효성 평가: OS Hazard Ratio(HR)

·  효과 판정 기준: HR <1.00일 Bayesian probability of benefit ≥98%

·  Limited efficacy 기준: 모든 평가군에서 Bayesian predictive power <25%

·  등록 중단: Limited efficacy 확인 시

·  추적관찰: 등록 중단 후 12개월

RESULTS

·  Predictive power <25%: NDU / RD / All 전체 평가군

·  Limited efficacy: 확인

·  환자 등록: 중단

·  OS 개선: NDU 및 RD 모두 확인되지 않음

·  Median HR: NDU → 1.05 RD → 1.07 All → 1.07

·  Probability of benefit(HR<1.00): NDU → 42.1% RD → 31.2% All → 29.6%

·  독성: 대조군 대비 증가

CONCLUSION

·  RD: Regorafenib의 Lomustine 대비 우월성 없음

·  NDU: Regorafenib의 Temozolomide + 방사선치료 대비 우월성 없음

·  OS 개선: 입증 실패

·  독성: 증가

NCCN 가이드라인: 재발성 GBM 치료 옵션에서 Regorafenib 삭제

FIG 3. (A) NDU 및 (B) RD 교모세포종 환자 중 레고라페닙군 또는 대조군에 무작위 배정된 대상자의 최종 전체생존기간(OS)을 각 하위 유형에 따라 제시함. (C) Kaplan-Meier 방법으로 산출한 전체생존기간(OS) 결과를 개월 단위로 제시하였으며, 중앙값(median)은 Bayesian 모델링에 기반한 값으로 관련 표에 제시함. (주요 평가변수인 OS에 대한 프로토콜 분석 결과)  OS 중앙값은 Kaplan-Meier 곡선이 아닌 Bayesian 모델에 기반하여 산출함.