Gynecologic Oncology 133 (2014) 105–110
Contents lists available at ScienceDirect
Gynecologic Oncology journal homepage: www.elsevier.com/locate/ygyno
Independent radiologic review: Bevacizumab in combination with gemcitabine and carboplatin in recurrent ovarian cancer☆ Carol Aghajanian a,f,⁎, Barbara Goff b, Lawrence R. Nycum c, Yan Wang d, Amreen Husain d, Stephanie Blank e a
Memorial Sloan-Kettering Cancer Center, Gynecologic Medical Oncology Service, 300 East 66th Street, New York, NY 10065, USA University of Washington School of Medicine, Department of Obstetrics and Gynecology, Box 356460, Seattle, WA 98195, USA Novant Health Forsyth Medical Center, Division of Gynecologic Oncology, 3333 Silas Creek Pkwy, Winston-Salem, NC 27103, USA d Genentech, Inc., Product Development, 1 DNA Way, South San Francisco, CA 94080, USA e New York University School of Medicine, Department of Obstetrics and Gynecology, 160 East 34th Street, New York, NY 10016, USA f Weill Cornell Medical College, 445 East 69th Street, New York, NY 10021, USA b c
H I G H L I G H T S • Tumor assessment by an independent review committee confirmed improved PFS and ORR for GC + BEV vs. GC + placebo. • Radiologic imaging in conjunction with RECIST provides a reliable means of assessing objective response and PD status in recurrent OC.
a r t i c l e
i n f o
Article history: Received 19 November 2013 Accepted 2 February 2014 Available online 6 February 2014 Keywords: Recurrent ovarian cancer Bevacizumab Phase 3 IRC Independent review
a b s t r a c t Objective. OCEANS, a randomized, placebo-controlled, phase III trial, found that adding bevacizumab to gemcitabine–carboplatin (GC) significantly improved investigator-determined progression-free survival (PFS) and objective response rate (ORR) in platinum-sensitive, recurrent ovarian cancer. To evaluate the reliability of assessment of progression and objective response per RECIST, radiologic and clinical data were assessed by an independent review committee (IRC). Methods. Radiologic images and clinical data were provided prospectively to the IRC for all randomized patients (N = 484). Data were reviewed in a blinded fashion per RECIST (modified v1.0). PFS and ORR were analyzed based on the IRC assessment. Concordance between investigator- and IRC-assessed progression and objective response was assessed. Results. The IRC analysis demonstrated a statistically significant increase in PFS (hazard ratio [HR] = 0.451; 95% confidence interval [CI] = 0.351 to 0.580, p b 0.0001) consistent with the benefit reported by investigators (HR = 0.484; 95% CI = 0.388 to 0.605, p b 0.0001). The concordance rate, defined by agreement on progression status, was 74.2% overall, and comparable between treatment arms (bevacizumab, 75.2% vs. placebo, 73.1%). IRCassessed ORR was significantly improved with bevacizumab (bevacizumab, 74.8% vs. placebo, 53.7%; p b 0.0001), consistent with the investigator-assessed results. The concordance rate for objective response was 79.8% overall, and comparable between treatment arms (bevacizumab, 78.9% vs. placebo, 80.6%). Conclusions. IRC-determined results were highly consistent with those determined by investigators, demonstrating that bevacizumab plus GC provides a significant improvement in PFS and ORR. These results suggest that investigators can reliably assess disease progression and objective response in recurrent ovarian cancer using RECIST, without the necessity of a full IRC review. © 2014 Elsevier Inc. All rights reserved.
Introduction Progression-free survival (PFS) is commonly used as the primary endpoint in clinical trials evaluating interventional therapy for ovarian
☆ Presented at the 43rd Annual Meeting on Women's Cancer, Society of Gynecologic Oncology, Austin, Texas, March 24–27, 2012. ⁎ Corresponding author at: Memorial Sloan-Kettering Cancer Center, Gynecologic Medical Oncology Service, 300 East 66th St, New York, NY 10065, USA. Fax: +1 646 888 4267. E-mail address: [email protected] (C. Aghajanian). 0090-8258/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ygyno.2014.02.003
cancer [1–8]. Compared with overall survival (OS), PFS measures require less follow-up time and are not confounded by the effects of non-protocol therapies administered after progressive disease (PD). PFS has been endorsed by international gynecologic oncology groups as a valid and clinically relevant endpoint for assessing new treatments in ovarian cancer [9–11]. Objective response is frequently utilized as a secondary endpoint in ovarian cancer trials. Determination of a patient's objective response or progression status (for PFS) depends on the accurate evaluation of radiologic tumor assessments [12,13]. Definitive criteria, such as the Response Evaluation
106
C. Aghajanian et al. / Gynecologic Oncology 133 (2014) 105–110
Study Comparing Efficacy and Safety of Chemotherapy and AntiAngiogenic Therapy in Platinum-Sensitive Recurrent Disease (OCEANS) study. This randomized, phase III study compared bevacizumab with placebo in combination with gemcitabine and carboplatin (GC) for platinum-sensitive, recurrent ovarian cancer. In the previously published primary analysis, OCEANS demonstrated a significant improvement in the primary endpoint, investigator-determined PFS, with median values of 12.4 and 8.4 months in the GC plus bevacizumab and GC plus placebo arms, respectively (hazard ratio [HR] = 0.484; 95% confidence interval [CI] = 0.388 to 0.605, p b 0.0001) [29]. A blinded IRC assessment of radiologic images and clinical data from OCEANS was performed as a sensitivity analysis and to evaluate the reliability of RECIST criteria for determining PD and objective response. As previously reported, the IRC analysis confirmed the PFS results obtained by the investigators [29]. Here we report a more detailed evaluation of the concordance between investigator- and IRC-determined analyses of multiple categories, including PD, objective response, PFS, disease progression dates, and adjudication rates.
Criteria In Solid Tumors (RECIST), were developed to improve consistency in radiologic evaluations [14,15]. However, bias in estimating the treatment effects can be introduced by inter- or intra-observer variability in the choice of target lesions, the ability to identify new lesions, and the interpretation of nontarget or nonmeasurable lesions [16–20]. Variation between evaluators' assessments may be particularly germane in ovarian cancer, where diffuse peritoneal implantation and frequent nontarget lesions, including ascites, can complicate radiologic assessments. Regulatory authorities have recommended that phase III registration studies utilizing a PFS endpoint employ an independent review committee (IRC) to assess radiologic images and confirm PD status [12,21]. IRC assessment may reduce variability through uniform training of radiologists, use of test cases to standardize image interpretation, and use of common imaging analysis tools [22]. The value of IRC assessments has, however, been called into question [23]. Along with the substantial costs and logistical burdens that an IRC imposes, it may also introduce new biases into the estimation of treatment effects. Since radiologic assessment usually stops after PD, when a patient is judged by investigators to have progressed, PD may not be subsequently confirmed by the IRC. Censoring events not confirmed by the IRC results in informative censoring, which can lead to a biased estimation of treatment effect (underestimation or overestimation), especially if the informative censoring occurs more frequently in one arm [23,24]. In contrast to potential pitfalls, meta-analyses have found that IRC results tend to be consistent with those determined by investigators [24–26]. Given the consistency between investigator and IRC assessments, and the potential for IRCs to introduce bias, several authors have suggested that IRCs for large, blinded trials be replaced with an IRC audit examining only a random subset of patients [27]. The FDA's Oncologic Drugs Advisory Committee recently found that IRC audits represented a reasonable alternative to full IRC review [24,28]. The matter is undergoing regulatory consideration. Given these issues, additional research is needed to examine the usefulness of IRCs in determining PD and objective response. An opportunity to investigate such questions is provided by the Ovarian Cancer
Methods Patients and study design The methodology and design of the randomized, multicenter, doubleblind, placebo-controlled, phase III OCEANS study (NCT00434642) [30] have been reported previously [29]. In brief, OCEANS enrolled patients ≥18 years with histologically confirmed recurrent ovarian, primary peritoneal, or fallopian tube cancer who had experienced disease progression ≥6 months after completion of a first-line platinum-based regimen. Patients were required to have measurable disease at baseline according to RECIST. The study was approved by institutional review boards at all study sites, and all patients provided informed consent. Patients were randomly assigned to receive either bevacizumab or placebo in combination with GC, with stratification by platinum-free interval (6–12, N 12 months) and cytoreductive surgery for recurrent disease (yes, no). Gemcitabine (1000 mg/m2 on days 1 and 8) and
Scans sent prospectively to IRC Adjudication process Read 1 (radiologist)
Concordance? Read 2 (radiologist)
N Adjudication Read 3 (radiologist)
Y
Accept Read 1 or 2
N (Level 2) Agree with either Read 1 or 2?
Y (Level 1)
Adjudicator re-reads case Accept Read 3
Radiologist assessment of scans: • Best response • Date of progression
Additional clinical data
Medical oncologist (final review)
Final assessment by IRC: • Best response • Date of progression Fig. 1. IRC process. Imaging scans are provided prospectively to two IRC radiologists who perform primary reads to determine response and progression status (and assign a progression date, if necessary). If the two primary readers disagree on either response or progression status, then a third radiologist serves as adjudicator, reviewing the images. A medical oncologist then reviews the radiologist assessment, along with other clinical data, to document the final assessment by the IRC. Abbreviation: IRC, independent review committee.
C. Aghajanian et al. / Gynecologic Oncology 133 (2014) 105–110
carboplatin (area under the curve 4 mg/mL/min on day 1) were administered for 6 to 10 cycles of 21 days in duration. Bevacizumab (15 mg/kg) or placebo was administered on day 1 of each chemotherapy cycle. Following completion of chemotherapy, bevacizumab or placebo continued to be administered, once every 21 days, until PD or unacceptable toxicity. Patients were required to undergo computed tomography (CT) scans every 9 weeks, calculated from day 1 of cycle 1, regardless of treatment delays, treatment discontinuation, or any additional scans that were performed off-schedule, until disease progression assessed by investigators, according to modified RECIST version 1.0. The primary endpoint was PFS, defined as the time interval from randomization to disease progression or death due to any cause, whichever occurred first. Symptomatic clinical progression, but not cancer antigen 125 (CA-125) elevation alone, was defined as a progression event. Secondary endpoints included ORR, duration of objective response (DOR), OS, and safety. Of these secondary endpoints, the results for ORR and DOR based on IRC assessments are presented here. Objective response and its duration are defined in the Statistical analysis section. IRC assessment of response and progression The IRC process is summarized in Fig. 1. Radiologic images and relevant clinical data were sent prospectively to the IRC for all randomized patients, as they became available. Clinical data sent to the IRC were limited to baseline subject profile information identifying benign radiologic abnormalities that may mimic metastatic disease and information on prior radiation therapy history that could potentially affect the selection of target lesions; data on preexisting medical history, CA-125 levels, or other clinical data were not provided to independent radiologists for review. Images were reviewed by the IRC in a blinded fashion, in accordance with a prespecified charter, based on modified RECIST version 1.0 [14]. CT images were initially reviewed by two IRC radiologists. A third radiologist served as an adjudicator to resolve any differences in interpretation between the first two readers. Adjudication was required in cases when assessments by the two primary radiologists differed on either of two variables: best objective response or PD date. The adjudicator could either agree with one of the original radiologists or provide a third interpretation that would be considered the accepted image interpretation. Images, radiologist interpretations, and clinical data (i.e. from tumor response case report forms) were then reviewed by an oncologist who made a final determination of objective response and progression status, as the radiologist interpretation of imaging data is only one part of the independent review of overall response. Providing an overall
107
assessment of response necessitated integration of clinical data with the response data based on RECIST. The following clinical data from the tumor response CRFs sent to the oncologist have included: documented clinical progression with supporting comments of signs and symptoms for the determination of clinical progression (provided by the investigator); ovarian cancer procedures performed on the patients; and lesions found on physical examination performed by the investigators. Statistical analysis The PFS data for patients who had not experienced PD and who had not died at the clinical cutoff date were censored at the date of the last tumor assessment (on or before the clinical cutoff date) or at the initiation of non-protocol-specified anticancer therapy. The PFS data for patients with no postbaseline tumor assessments and no death captured in the clinical database were censored at the date of randomization plus 1 day. The PFS analysis was performed by using a two-sided stratified log-rank test based on the intent-to-treat (ITT) population. Median PFS was estimated using Kaplan–Meier methodology. HRs were estimated using a stratified Cox regression model to assess the magnitude of the treatment effect. Stratification factors used at randomization were included in the model. Analysis of ORR was performed on the ITT population. Objective response was defined as the occurrence of complete response (CR) or partial response (PR) at any time point onstudy, confirmed by repeat assessment ≥ 4 weeks later. Patients not meeting these criteria, including those without postbaseline tumor assessments, were considered nonresponders. The Cochran–Mantel– Haenszel test was used to compare ORRs between treatment arms with the same stratification factors as in the PFS analysis. DOR was defined as the time from the first occurrence of a confirmed PR/CR until PD or death, whichever occurred first. DOR data were censored in the same way as for PFS. Descriptive statistics were provided for DOR within the nonrandomized subset of patients who experienced an objective response; formal hypothesis testing was not performed for this endpoint. Rates of concordance between the investigator-determined and IRCdetermined status for PD (yes, no), date of PD, and objective response (yes, no) were summarized for each treatment arm and for all patients. Discrepancy rates for each arm and their differential discordance rates were calculated as described by Amit et al. [25]. The early discrepancy rate (EDR) quantifies the frequency with which the investigators declared PD earlier than IRC among all patients with investigator-
Table 1 PFS, overall ORR, and DOR determined by investigators and the independent review committeea. Investigator assessment GC + placebo (n = 242)
GC + bevacizumab (n = 242)
IRC assessment GC + placebo (n = 242)
GC + bevacizumab (n = 242)
b
PFS Patients with events, n (%) Median, months (95% CI) Stratified HR (95% CI) Log-rank p value Objective response ORR, % (95% CI) CR rate, % PR rate, % Difference in ORRs, % CMH p value, stratified DOR Patients with events, n (%) Median, months (95% CI) Stratified HR (95% CI) Log-rank p value
187 (77.3) 8.4 (8.3–9.7)
151 (62.4) 12.4 (11.4–12.7) 0.484 (0.388–0.605) b0.0001
57.4 (51.2–63.7) 9.1 48.3
78.5 (73.3–83.7) 17.4 61.2
148 (61.2) 8.6 (8.3–10.2)
53.7 (47.4–60.0) 1.2 52.5
21.1 (13.0–29.2) b0.0001 n = 139 105 (75.5) 7.4 (6.3–8.3)
n = 190 119 (62.6) 10.4 (9.4–11.8) 0.534 (0.408–0.698) b0.0001
119 (49.2) 12.3 (10.7–14.6) 0.451 (0.351–0.580) b0.0001 74.8 (69.3–80.3) 0.8 74.0 21.1 (12.7–29.4) b0.0001
n = 130 80 (61.5) 6.0 (4.3–6.2)
n = 181 96 (53.0) 8.3 (6.5–9.9) 0.535 (0.393–0.727) b0.0001
Abbreviations: CI, confidence interval; CMH, Cochran–Mantel–Haenszel; CR, complete response; DOR, duration of objective response; GC, gemcitabine plus carboplatin; HR, hazard ratio; IRC, independent review committee; ORR, objective response rate; PFS, progression-free survival; PR, partial response. a Data on investigator-determined PFS, objective response, and DOR, and IRC-determined PFS were reported previously [29]. b Involved censoring for non-protocol-specified cancer therapy.
108
C. Aghajanian et al. / Gynecologic Oncology 133 (2014) 105–110
A
CR
PR
78.5
80
57.4 P
Contents lists available at ScienceDirect
Gynecologic Oncology journal homepage: www.elsevier.com/locate/ygyno
Independent radiologic review: Bevacizumab in combination with gemcitabine and carboplatin in recurrent ovarian cancer☆ Carol Aghajanian a,f,⁎, Barbara Goff b, Lawrence R. Nycum c, Yan Wang d, Amreen Husain d, Stephanie Blank e a
Memorial Sloan-Kettering Cancer Center, Gynecologic Medical Oncology Service, 300 East 66th Street, New York, NY 10065, USA University of Washington School of Medicine, Department of Obstetrics and Gynecology, Box 356460, Seattle, WA 98195, USA Novant Health Forsyth Medical Center, Division of Gynecologic Oncology, 3333 Silas Creek Pkwy, Winston-Salem, NC 27103, USA d Genentech, Inc., Product Development, 1 DNA Way, South San Francisco, CA 94080, USA e New York University School of Medicine, Department of Obstetrics and Gynecology, 160 East 34th Street, New York, NY 10016, USA f Weill Cornell Medical College, 445 East 69th Street, New York, NY 10021, USA b c
H I G H L I G H T S • Tumor assessment by an independent review committee confirmed improved PFS and ORR for GC + BEV vs. GC + placebo. • Radiologic imaging in conjunction with RECIST provides a reliable means of assessing objective response and PD status in recurrent OC.
a r t i c l e
i n f o
Article history: Received 19 November 2013 Accepted 2 February 2014 Available online 6 February 2014 Keywords: Recurrent ovarian cancer Bevacizumab Phase 3 IRC Independent review
a b s t r a c t Objective. OCEANS, a randomized, placebo-controlled, phase III trial, found that adding bevacizumab to gemcitabine–carboplatin (GC) significantly improved investigator-determined progression-free survival (PFS) and objective response rate (ORR) in platinum-sensitive, recurrent ovarian cancer. To evaluate the reliability of assessment of progression and objective response per RECIST, radiologic and clinical data were assessed by an independent review committee (IRC). Methods. Radiologic images and clinical data were provided prospectively to the IRC for all randomized patients (N = 484). Data were reviewed in a blinded fashion per RECIST (modified v1.0). PFS and ORR were analyzed based on the IRC assessment. Concordance between investigator- and IRC-assessed progression and objective response was assessed. Results. The IRC analysis demonstrated a statistically significant increase in PFS (hazard ratio [HR] = 0.451; 95% confidence interval [CI] = 0.351 to 0.580, p b 0.0001) consistent with the benefit reported by investigators (HR = 0.484; 95% CI = 0.388 to 0.605, p b 0.0001). The concordance rate, defined by agreement on progression status, was 74.2% overall, and comparable between treatment arms (bevacizumab, 75.2% vs. placebo, 73.1%). IRCassessed ORR was significantly improved with bevacizumab (bevacizumab, 74.8% vs. placebo, 53.7%; p b 0.0001), consistent with the investigator-assessed results. The concordance rate for objective response was 79.8% overall, and comparable between treatment arms (bevacizumab, 78.9% vs. placebo, 80.6%). Conclusions. IRC-determined results were highly consistent with those determined by investigators, demonstrating that bevacizumab plus GC provides a significant improvement in PFS and ORR. These results suggest that investigators can reliably assess disease progression and objective response in recurrent ovarian cancer using RECIST, without the necessity of a full IRC review. © 2014 Elsevier Inc. All rights reserved.
Introduction Progression-free survival (PFS) is commonly used as the primary endpoint in clinical trials evaluating interventional therapy for ovarian
☆ Presented at the 43rd Annual Meeting on Women's Cancer, Society of Gynecologic Oncology, Austin, Texas, March 24–27, 2012. ⁎ Corresponding author at: Memorial Sloan-Kettering Cancer Center, Gynecologic Medical Oncology Service, 300 East 66th St, New York, NY 10065, USA. Fax: +1 646 888 4267. E-mail address: [email protected] (C. Aghajanian). 0090-8258/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ygyno.2014.02.003
cancer [1–8]. Compared with overall survival (OS), PFS measures require less follow-up time and are not confounded by the effects of non-protocol therapies administered after progressive disease (PD). PFS has been endorsed by international gynecologic oncology groups as a valid and clinically relevant endpoint for assessing new treatments in ovarian cancer [9–11]. Objective response is frequently utilized as a secondary endpoint in ovarian cancer trials. Determination of a patient's objective response or progression status (for PFS) depends on the accurate evaluation of radiologic tumor assessments [12,13]. Definitive criteria, such as the Response Evaluation
106
C. Aghajanian et al. / Gynecologic Oncology 133 (2014) 105–110
Study Comparing Efficacy and Safety of Chemotherapy and AntiAngiogenic Therapy in Platinum-Sensitive Recurrent Disease (OCEANS) study. This randomized, phase III study compared bevacizumab with placebo in combination with gemcitabine and carboplatin (GC) for platinum-sensitive, recurrent ovarian cancer. In the previously published primary analysis, OCEANS demonstrated a significant improvement in the primary endpoint, investigator-determined PFS, with median values of 12.4 and 8.4 months in the GC plus bevacizumab and GC plus placebo arms, respectively (hazard ratio [HR] = 0.484; 95% confidence interval [CI] = 0.388 to 0.605, p b 0.0001) [29]. A blinded IRC assessment of radiologic images and clinical data from OCEANS was performed as a sensitivity analysis and to evaluate the reliability of RECIST criteria for determining PD and objective response. As previously reported, the IRC analysis confirmed the PFS results obtained by the investigators [29]. Here we report a more detailed evaluation of the concordance between investigator- and IRC-determined analyses of multiple categories, including PD, objective response, PFS, disease progression dates, and adjudication rates.
Criteria In Solid Tumors (RECIST), were developed to improve consistency in radiologic evaluations [14,15]. However, bias in estimating the treatment effects can be introduced by inter- or intra-observer variability in the choice of target lesions, the ability to identify new lesions, and the interpretation of nontarget or nonmeasurable lesions [16–20]. Variation between evaluators' assessments may be particularly germane in ovarian cancer, where diffuse peritoneal implantation and frequent nontarget lesions, including ascites, can complicate radiologic assessments. Regulatory authorities have recommended that phase III registration studies utilizing a PFS endpoint employ an independent review committee (IRC) to assess radiologic images and confirm PD status [12,21]. IRC assessment may reduce variability through uniform training of radiologists, use of test cases to standardize image interpretation, and use of common imaging analysis tools [22]. The value of IRC assessments has, however, been called into question [23]. Along with the substantial costs and logistical burdens that an IRC imposes, it may also introduce new biases into the estimation of treatment effects. Since radiologic assessment usually stops after PD, when a patient is judged by investigators to have progressed, PD may not be subsequently confirmed by the IRC. Censoring events not confirmed by the IRC results in informative censoring, which can lead to a biased estimation of treatment effect (underestimation or overestimation), especially if the informative censoring occurs more frequently in one arm [23,24]. In contrast to potential pitfalls, meta-analyses have found that IRC results tend to be consistent with those determined by investigators [24–26]. Given the consistency between investigator and IRC assessments, and the potential for IRCs to introduce bias, several authors have suggested that IRCs for large, blinded trials be replaced with an IRC audit examining only a random subset of patients [27]. The FDA's Oncologic Drugs Advisory Committee recently found that IRC audits represented a reasonable alternative to full IRC review [24,28]. The matter is undergoing regulatory consideration. Given these issues, additional research is needed to examine the usefulness of IRCs in determining PD and objective response. An opportunity to investigate such questions is provided by the Ovarian Cancer
Methods Patients and study design The methodology and design of the randomized, multicenter, doubleblind, placebo-controlled, phase III OCEANS study (NCT00434642) [30] have been reported previously [29]. In brief, OCEANS enrolled patients ≥18 years with histologically confirmed recurrent ovarian, primary peritoneal, or fallopian tube cancer who had experienced disease progression ≥6 months after completion of a first-line platinum-based regimen. Patients were required to have measurable disease at baseline according to RECIST. The study was approved by institutional review boards at all study sites, and all patients provided informed consent. Patients were randomly assigned to receive either bevacizumab or placebo in combination with GC, with stratification by platinum-free interval (6–12, N 12 months) and cytoreductive surgery for recurrent disease (yes, no). Gemcitabine (1000 mg/m2 on days 1 and 8) and
Scans sent prospectively to IRC Adjudication process Read 1 (radiologist)
Concordance? Read 2 (radiologist)
N Adjudication Read 3 (radiologist)
Y
Accept Read 1 or 2
N (Level 2) Agree with either Read 1 or 2?
Y (Level 1)
Adjudicator re-reads case Accept Read 3
Radiologist assessment of scans: • Best response • Date of progression
Additional clinical data
Medical oncologist (final review)
Final assessment by IRC: • Best response • Date of progression Fig. 1. IRC process. Imaging scans are provided prospectively to two IRC radiologists who perform primary reads to determine response and progression status (and assign a progression date, if necessary). If the two primary readers disagree on either response or progression status, then a third radiologist serves as adjudicator, reviewing the images. A medical oncologist then reviews the radiologist assessment, along with other clinical data, to document the final assessment by the IRC. Abbreviation: IRC, independent review committee.
C. Aghajanian et al. / Gynecologic Oncology 133 (2014) 105–110
carboplatin (area under the curve 4 mg/mL/min on day 1) were administered for 6 to 10 cycles of 21 days in duration. Bevacizumab (15 mg/kg) or placebo was administered on day 1 of each chemotherapy cycle. Following completion of chemotherapy, bevacizumab or placebo continued to be administered, once every 21 days, until PD or unacceptable toxicity. Patients were required to undergo computed tomography (CT) scans every 9 weeks, calculated from day 1 of cycle 1, regardless of treatment delays, treatment discontinuation, or any additional scans that were performed off-schedule, until disease progression assessed by investigators, according to modified RECIST version 1.0. The primary endpoint was PFS, defined as the time interval from randomization to disease progression or death due to any cause, whichever occurred first. Symptomatic clinical progression, but not cancer antigen 125 (CA-125) elevation alone, was defined as a progression event. Secondary endpoints included ORR, duration of objective response (DOR), OS, and safety. Of these secondary endpoints, the results for ORR and DOR based on IRC assessments are presented here. Objective response and its duration are defined in the Statistical analysis section. IRC assessment of response and progression The IRC process is summarized in Fig. 1. Radiologic images and relevant clinical data were sent prospectively to the IRC for all randomized patients, as they became available. Clinical data sent to the IRC were limited to baseline subject profile information identifying benign radiologic abnormalities that may mimic metastatic disease and information on prior radiation therapy history that could potentially affect the selection of target lesions; data on preexisting medical history, CA-125 levels, or other clinical data were not provided to independent radiologists for review. Images were reviewed by the IRC in a blinded fashion, in accordance with a prespecified charter, based on modified RECIST version 1.0 [14]. CT images were initially reviewed by two IRC radiologists. A third radiologist served as an adjudicator to resolve any differences in interpretation between the first two readers. Adjudication was required in cases when assessments by the two primary radiologists differed on either of two variables: best objective response or PD date. The adjudicator could either agree with one of the original radiologists or provide a third interpretation that would be considered the accepted image interpretation. Images, radiologist interpretations, and clinical data (i.e. from tumor response case report forms) were then reviewed by an oncologist who made a final determination of objective response and progression status, as the radiologist interpretation of imaging data is only one part of the independent review of overall response. Providing an overall
107
assessment of response necessitated integration of clinical data with the response data based on RECIST. The following clinical data from the tumor response CRFs sent to the oncologist have included: documented clinical progression with supporting comments of signs and symptoms for the determination of clinical progression (provided by the investigator); ovarian cancer procedures performed on the patients; and lesions found on physical examination performed by the investigators. Statistical analysis The PFS data for patients who had not experienced PD and who had not died at the clinical cutoff date were censored at the date of the last tumor assessment (on or before the clinical cutoff date) or at the initiation of non-protocol-specified anticancer therapy. The PFS data for patients with no postbaseline tumor assessments and no death captured in the clinical database were censored at the date of randomization plus 1 day. The PFS analysis was performed by using a two-sided stratified log-rank test based on the intent-to-treat (ITT) population. Median PFS was estimated using Kaplan–Meier methodology. HRs were estimated using a stratified Cox regression model to assess the magnitude of the treatment effect. Stratification factors used at randomization were included in the model. Analysis of ORR was performed on the ITT population. Objective response was defined as the occurrence of complete response (CR) or partial response (PR) at any time point onstudy, confirmed by repeat assessment ≥ 4 weeks later. Patients not meeting these criteria, including those without postbaseline tumor assessments, were considered nonresponders. The Cochran–Mantel– Haenszel test was used to compare ORRs between treatment arms with the same stratification factors as in the PFS analysis. DOR was defined as the time from the first occurrence of a confirmed PR/CR until PD or death, whichever occurred first. DOR data were censored in the same way as for PFS. Descriptive statistics were provided for DOR within the nonrandomized subset of patients who experienced an objective response; formal hypothesis testing was not performed for this endpoint. Rates of concordance between the investigator-determined and IRCdetermined status for PD (yes, no), date of PD, and objective response (yes, no) were summarized for each treatment arm and for all patients. Discrepancy rates for each arm and their differential discordance rates were calculated as described by Amit et al. [25]. The early discrepancy rate (EDR) quantifies the frequency with which the investigators declared PD earlier than IRC among all patients with investigator-
Table 1 PFS, overall ORR, and DOR determined by investigators and the independent review committeea. Investigator assessment GC + placebo (n = 242)
GC + bevacizumab (n = 242)
IRC assessment GC + placebo (n = 242)
GC + bevacizumab (n = 242)
b
PFS Patients with events, n (%) Median, months (95% CI) Stratified HR (95% CI) Log-rank p value Objective response ORR, % (95% CI) CR rate, % PR rate, % Difference in ORRs, % CMH p value, stratified DOR Patients with events, n (%) Median, months (95% CI) Stratified HR (95% CI) Log-rank p value
187 (77.3) 8.4 (8.3–9.7)
151 (62.4) 12.4 (11.4–12.7) 0.484 (0.388–0.605) b0.0001
57.4 (51.2–63.7) 9.1 48.3
78.5 (73.3–83.7) 17.4 61.2
148 (61.2) 8.6 (8.3–10.2)
53.7 (47.4–60.0) 1.2 52.5
21.1 (13.0–29.2) b0.0001 n = 139 105 (75.5) 7.4 (6.3–8.3)
n = 190 119 (62.6) 10.4 (9.4–11.8) 0.534 (0.408–0.698) b0.0001
119 (49.2) 12.3 (10.7–14.6) 0.451 (0.351–0.580) b0.0001 74.8 (69.3–80.3) 0.8 74.0 21.1 (12.7–29.4) b0.0001
n = 130 80 (61.5) 6.0 (4.3–6.2)
n = 181 96 (53.0) 8.3 (6.5–9.9) 0.535 (0.393–0.727) b0.0001
Abbreviations: CI, confidence interval; CMH, Cochran–Mantel–Haenszel; CR, complete response; DOR, duration of objective response; GC, gemcitabine plus carboplatin; HR, hazard ratio; IRC, independent review committee; ORR, objective response rate; PFS, progression-free survival; PR, partial response. a Data on investigator-determined PFS, objective response, and DOR, and IRC-determined PFS were reported previously [29]. b Involved censoring for non-protocol-specified cancer therapy.
108
C. Aghajanian et al. / Gynecologic Oncology 133 (2014) 105–110
A
CR
PR
78.5
80
57.4 P
