Study Design

GOG-0218 is an international, multicenter, double-blind, placebo-controlled, phase III trial in which two experimental arms were compared with a control arm that consisted of six 21-day cycles of intravenous carboplatin (area under the concentration v time curve 6) plus paclitaxel (175 mg/m² body surface area) followed by 16 21-day cycles of placebo. To limit wound healing complications, bevacizumab or placebo was initiated with cycle 2. The bevacizumab-concurrent arm consisted of intravenous chemotherapy as per the control arm with bevacizumab (15 mg/kg body weight) added to cycles 2 through 6. The bevacizumab-concurrent plus maintenance arm substituted bevacizumab (15 mg/kg) for placebo in cycles 7 to 22. Treatment was discontinued upon completion of all 22 cycles or with documented progression, unacceptable toxicity, or voluntary patient withdrawal.

Disease was assessed using Response Evaluation Criteria in Solid Tumors (RECIST).¹⁹ Patients underwent computed tomography or magnetic resonance imaging of the abdomen and pelvis before cycle 1, and for those without evidence of progression, imaging was repeated after cycles 3, 6, 10, 14, 18, and 22. Serum cancer antigen 125 was obtained before each cycle for cycles 1 through 6 and at the beginning of alternate cycles for cycles 7 to 22. Radiologic evaluation was repeated every 3 months for 2 years, then every 6 months for 3 years, and then annually. The quality-of-life instruments and analysis of patient-reported outcomes are provided in the Data Supplement.

No dose reductions of bevacizumab or placebo occurred. Dose was modified only when weight changed by more than 10%. Delay and/or discontinuation of bevacizumab was predicated on the occurrence, duration, and severity of uncontrolled hypertension (defined as systolic blood pressure greater than 150 mm Hg or diastolic blood pressure greater than 90 mm Hg) and/or proteinuria (urine protein-creatinine ratio greater than 3.5). Intestinal wall disruption during cycle 2 or later, arterial thrombosis, venous thrombosis, and/or the development of reversible posterior leukoencephalopathy syndrome also led to bevacizumab discontinuation. Criteria for chemotherapy dose modifications and administration of myeloid growth factor are provided in the Data Supplement.

Statistical Analysis

The statistical analysis plan from the master protocol is available in the Data Supplement. Before random assignment, according to a 1:1:1 dynamic allocation procedure, patients were stratified by surgical debulking outcome and stage and by initial GOG performance status score (0 v 1 or 2).²⁰ Early adverse events were graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (version 3) and reported until 30 days after administration of the last study treatment. Long-term adverse events of interest were submitted for up to 5 years after treatment allocation discovery.

OS was initially specified as the primary end point and defined as the observed length of life from study entry to death, regardless of cause. The study sought to determine whether the addition of five concurrent cycles of bevacizumab with or without maintenance bevacizumab increases the duration of OS compared with chemotherapy alone. To allow unblinding at progression, the primary end point was changed to PFS and reported in 2010.

The targeted accrual was 1,800 patients, with approximately 600 per arm. This would provide 90% power to detect a 23% reduction in the hazard for progression or death with either bevacizumab-containing regimen or control while limiting the overall one-sided type I error rate to 2.5%. If both experimental regimens were found to significantly increase OS relative to the control regimen, then the two investigational arms would be compared. Median survival for stage III with residual disease greater than 1 cm or stage IV was estimated to be 31 months, whereas that of stage III with macroscopic residual disease of 1 cm or less was 42 months. The final analysis of OS required at least 375 deaths in the control arm. Differences in OS among the treatment arms were assessed by a log-rank test.²¹ Cox proportional hazards regression models adjusted for GOG performance status and stage were used to estimate relative hazards and provided two-sided P values.²²

Efficacy

For the protocol-specified intention-to-treat analysis of OS, the database was locked on January 17, 2018, at a median follow-up of 102.9 months and 493 deaths in the control arm. Relative to control treatment (median OS, 41.1 months), there were no significant differences in OS among patients treated in the bevacizumab-concurrent plus maintenance arm (median OS, 43.4 months; HR, 0.96; 95% CI, 0.85 to 1.09; P = .53) or in the bevacizumab-concurrent arm (median OS, 40.8 months; HR 1.06; 95% CI, 0.94 to 1.20; P = .34; Fig 2A).

Open in a separate window

FIG 2.

Kaplan-Meier curves that compare overall survival (OS) of the various treatment arms studied in GOG-0218. Database lock was January 17, 2018. (A) Final protocol-specified OS according to treatment arm. Overall median follow-up was 102.9 months (95% CI, 100.7 to 105.9 months). In the 1,873 patients who comprised the entire study population, 1,491 died, 204 were alive with a progression event, and 178 were alive without progression. Median follow-up for the control cohort (carboplatin and paclitaxel plus placebo followed by placebo [CT + P → P]) was 103.4 months (95% CI, 100.4 to 107.9 months). In 625 patients, 565 experienced a progression event, and 493 died; 72 were alive with a progression event, and 60 were alive without progression. Median follow-up for the bevacizumab-concurrent cohort (carboplatin and paclitaxel plus bevacizumab followed by placebo [CT + B → P]) was 102.3 months (95% CI, 98.8 to 106.5 months). In 625 patients, 563 experienced a progression event, and 506 died; 57 were alive with a progression event, and 62 were alive without progression. The hazard ratio (HR) for bevacizumab-concurrent v control stratified by stage and Gynecologic Oncology Group (GOG) performance status was 1.06 (95% CI, 0.94 to 1.20; P = .34). Median follow-up for the bevacizumab-concurrent plus maintenance cohort (carboplatin and paclitaxel plus bevacizumab followed by bevacizumab [CT + B → B]) was 101.9 months (95% CI, 98.7 to 107.4 months). In 623 patients, 567 experienced a progression event, and 492 died; 75 were alive with a progression event, and 56 were alive without progression. The HR for death for bevacizumab-concurrent plus maintenance v control stratified by stage and GOG performance status was 0.96 (95% CI, 0.85 to 1.09; P = .53). (B) Disease-specific OS according to treatment arm. Patients who died as a result of causes other than disease or treatment-related causes were censored. Overall, the number of deaths reduced from 1,491 to 1,387 (change of 104; 7.0%). In the control arm, the number of deaths reduced from 493 to 462 (change of 31; 6.3%). In the bevacizumab-concurrent arm, the number of deaths reduced from 506 to 474 (change of 32; 6.3%). In the bevacizumab-concurrent plus maintenance arm, the number of deaths reduced from 492 to 451 (change of 41; 8.3%). The HR for bevacizumab-concurrent v control stratified by stage and GOG performance status was 1.06 (95% CI, 0.93 to 1.20; P = .39). The HR for death for bevacizumab-concurrent plus maintenance v control stratified by stage and GOG performance status was 0.94 (95% CI, 0.82 to 1.07; P = .33). (C) Final OS according to treatment arm after censoring those who received bevacizumab as a second line of therapy. Overall, 184 patients received bevacizumab as a second-line therapy after progression. When measuring OS since the time patients initiated treatment (n = 205), the HR for bevacizumab-concurrent v control stratified by stage and GOG performance status and after censoring those who received bevacizumab postprogression was 1.05 (95% CI, 0.92 to 1.20; P = .49). The HR for death for bevacizumab-concurrent plus maintenance v control stratified by stage and GOG performance status and after censoring those who received bevacizumab postprogression was 0.96 (95% CI, 0.84 to 1.09; P = .52). (D) Final OS according to treatment arm and stratified by International Federation of Gynecology and Obstetrics stage III. Survival of patients with stage III disease was similar in all three treatment arms. Relative to control, the HR for death for bevacizumab-concurrent was 1.08 (95% CI, 0.93 to 1.25), and that of bevacizumab-concurrent plus maintenance was 1.05 (95% CI, 0.91 to 1.22). (E) Final OS according to treatment arm and stratified by International Federation of Gynecology and Obstetrics stage IV. Relative to control, the HR for death of bevacizumab-concurrent was 0.99 (95% CI, 0.78 to 1.26), whereas that of bevacizumab-concurrent plus maintenance was 0.75 (95% CI, 0.59 to 0.95). Note that in this exploratory analysis, the survival of patients with stage IV disease treated in the bevacizumab-concurrent plus maintenance arm (box) approximated that observed for patients with stage III disease (as in panel D).

In an exploratory analysis that censored patients who died as a result of causes other than ovarian cancer or ovarian cancer treatment (n = 104), there was no survival advantage relative to the control treatment (median OS, 42.8 months) among patients randomly assigned to bevacizumab-concurrent plus maintenance (median OS, 45.6 months; HR, 0.94; 95% CI, 0.82 to 1.07; P = .33) or bevacizumab-concurrent (median OS, 42.3 months; HR, 1.06; 95% CI, 0.93 to 1.20; P = .39; Fig 2B). When censoring patients who had received bevacizumab at crossover before progression (Data Supplement) or as a second-line therapy postprogression (as a single agent or in combination with chemotherapy; Fig 2C), no significant survival differences relative to control were observed with either experimental arm.

In an exploratory subset analysis in which treatment was stratified by stage, median OS among patients with stage III disease was relatively similar at approximately 44.2 months (control), 42.9 months (bevacizumab-concurrent; HR, 1.08); and 44.3 months (bevacizumab-concurrent plus maintenance; HR, 1.05; Fig 2D). For those with stage IV disease, the control and bevacizumab-concurrent arms were associated with a median OS of 32.6 and 34.5 months, respectively (HR, 0.99). For patients with stage IV disease who received bevacizumab-concurrent plus maintenance, the median OS was 42.8 months (HR, 0.75; 95% CI, 0.59 to 0.95; Fig 2E).

Safety

Adverse events of interest were updated with long-term toxicity data (Table 2). During extended follow-up, five additional hypertensive events occurred, four of which in the bevacizumab-concurrent plus maintenance cohort. Two additional cases of proteinuria that involved two different patients treated in the bevacizumab-concurrent plus maintenance arm were also identified. Five venous thromboembolic events were reported, including one stroke in a patient treated with bevacizumab-concurrent plus maintenance, and three deep venous thromboses and one pulmonary embolism in three patients in the bevacizumab-concurrent arm. Eleven additional GI events were identified, six of which occurred in four patients treated with bevacizumab-concurrent plus maintenance, and included diarrhea, diverticulitis, enterocutaneous fistula, intestinal wall disruption, and two small bowel obstructions. There was one patient with myelodysplasia in the bevacizumab-concurrent arm and one with acute myeloid leukemia in the control arm.

TABLE 2.

Long-Term Toxicology Involving Adverse Events of Interest According to Treatment Group

Open in a separate window

Potential Prognostic and Predictive Factors

In an analysis of known clinical and pathologic prognostic markers, including age, GOG performance status, stage III, suboptimal stage III plus stage IV, presence or absence of ascites, histology, and grade, treatment with either of the bevacizumab-containing arms did not correlate with OS (Fig 3A). No prognostic or predictive signals were observed among the 1,438 patients included in the CD31 analyses (Fig 3B; Data Supplement). Analyses of genetic categories among 1,195 blood and/or tumor samples demonstrated that relative to wild type, the HR for death as a result of BRCA1/2 mutated carcinomas was 0.62 (95% CI, 0.52 to 0.73), and that for non-BRCA1/2 HRR mutated carcinomas was 0.65 (95% CI, 0.51 to 0.85; Fig 4A). Although clearly prognostic, these mutations were not predictive of bevacizumab activity (Fig 4B).

Open in a separate window

FIG 3.

Analysis of prognostic factors. (A) Forest plot of clinical and pathologic prognostic factors (N = 1,873). (B) Forest plot of molecular prognostic biomarkers. The sample size for the genetic category analysis was 1,195 blood and tumor samples. Patients were categorized into three genetic categories: BRCA1/2, homologous recombination repair mutations, or wild type (no mutation). For the forest plots, BRCA1/2 and homologous recombination repair mutations were depicted as separate categories. The sample size for the CD31 analysis was 1,438 blood and tumor samples. HR, hazard ratio; HRD, homologous repair deficiency; Q3, quartile 3.

Open in a separate window

FIG 4.

Analyses of BRCA1/2 and homologous recombination deficiency (HRD) according to final overall survival. (A) Kaplan-Meier survival curves demonstrate the prognostic impact of mutations in genes involved in homologous recombination. Of the 1,873 patients studied in GOG-0218, 1,195 provided blood or tumor samples to be sequenced and were able to be categorized into three genetic categories: breast cancer susceptibility gene 1/2 (BRCA1/2), homologous recombination repair (HRR) mutations, or wild type (no mutation). Using wild type as the reference, the hazard ratio (HR) for death for BRCA1/2 was 0.62 (95% CI, 0.52 to 0.73) and for death for HRR mutations, 0.65 (95% CI, 0.51 to 0.85). (B) Kaplan-Meier survival curves that demonstrate the effect of treatment allocation on overall survival stratified by genetic category. Although BRCA1/2 and HRR mutations were associated with improved prognosis, their occurrence did not predict bevacizumab activity. For graphic purposes, BRCA1/2 and HRR were combined into one category. For the patients who were BRCA/HRR negative (ie, wild type), the HR for death in those treated with bevacizumab-concurrent plus maintenance (carboplatin and paclitaxel plus bevacizumab followed by bevacizumab [CT + B → B]) relative to control (carboplatin and paclitaxel plus placebo followed by placebo [CT + P → P]) and stratified by stage and Gynecologic Oncology Group (GOG) performance status was 0.89 (95% CI, 0.74 to 1.07); the HR for death in those treated with bevacizumab-concurrent (carboplatin and paclitaxel plus bevacizumab followed by placebo [CT + B → P]) relative to control and stratified by stage and GOG performance status was 0.96 (95% CI, 0.80 to 1.15). For the patients who had BRCA/HRR mutations, the HR for death in those treated with bevacizumab-concurrent plus maintenance relative to control and stratified by stage and GOG performance status was 1.18 (95% CI, 0.86 to 1.63); the HR for death in those treated with bevacizumab-concurrent relative to control and stratified by stage and GOG performance status was 1.44 (95% CI, 1.02 to 2.02). (*) Germline wild type: BRCA1/2; ataxia-telangiectasia mutated (ATM); ataxia telangiectasia and Rad3 related (ATR); BRCA1-associated RING domain 1 (BARD1); Bloom DNA helicase (BLM); BRCA1 interacting protein C-terminal helicase 1 (BRIP1); checkpoint kinase 2 (CHEK2); MRE11 meiotic recombination 11 homolog A (MRE11A; Saccharomyces cerevisiae); nibrin (NBN); partner and localizer of BRCA2 (PALB2); RAD51 homolog C (RAD51C; S cerevisiae); RAD51 homolog D (RAD51D); RB binding protein 8, endonuclease (RBBP8); structure-specific endonuclease subunit homolog (SLX4; S cerevisiae); and x-ray repair cross complementing 2 (XRCC2).

We thank all participants and their significant others as well as the teams of physicians, nurses, and research coordinators for their dedication. We also thank the National Cancer Institute and Genentech for their generous support. The following institutions participated in this study: Abington Memorial Hospital; Abramson Cancer Center at the University of Pennsylvania; Aurora Women’s Pavilion of West Allis Memorial Hospital; Cancer Trials Support Unit; Cleveland Clinic Foundation; Community Clinical Oncology Program; Cooper Hospital University Medical Center; Duke University Medical Center; Fox Chase Cancer Center; Fred Hutchinson Cancer Research Center; Georgia Core; Gynecologic Oncology Network; Gynecologic Oncology of West Michigan; Indiana University Medical Center; Magee Women’s Hospital, University of Pittsburgh Medical Center; Mayo Clinic Rochester; MD Anderson Cancer Center; Memorial Sloan Kettering Cancer Center; Moffitt Cancer Center and Research Institute; Mount Sinai Medical Center; New York University Medical Center; Northwestern University; The Ohio State University Medical Center; Penn State Milton S. Hershey Medical Center; Roswell Park Cancer Institute; Rush University Medical Center; Saitama Medical University International/GOG Japan; Seoul National University Hospital/Korean GOG; State University of New York Downstate Medical Center; Stony Brook University Medical Center; The Hospital of Central Connecticut; University Hospitals–Ireland Cancer Center; University of Alabama at Birmingham; University of California, Los Angeles; University of California Medical Center at Irvine, Orange Campus; The University of Chicago; University of Cincinnati; University of Colorado Cancer Center, Anschutz Cancer Pavilion; University of Iowa Hospitals and Clinics; University of Kentucky; University of Massachusetts Medical School; University of Minnesota Medical Center–Fairview; University of Mississippi Medical Center; University of New Mexico; University of North Carolina; University of Oklahoma Health Sciences Center; The University of Texas Medical Branch; The University of Texas Southwestern Medical Center; University of Virginia; University of Wisconsin Hospitals and Clinics; Wake Forest University Health Sciences; Walter Reed Army Medical Center; Washington University School of Medicine; Wayne State University; Women and Infants’ Hospital; Women’s Cancer Center of Nevada; and Yale University.