Bevacizumab received US Food and Drug Administration approval for use in recurrent glioblastoma based on promising radiographic response data, but without clear evidence that it prolongs survival. A population-based analysis was conducted to determine whether bevacizumab approval was associated with improved glioblastoma survival in the United States.
Surveillance, Epidemiology, and End Results (SEER) Program data were used to compare survival of glioblastoma patients who died in 2006, 2008 (both prior to approval of bevacizumab), and 2010 (after approval of bevacizumab).
The SEER database contained 1715 patients with glioblastoma who died in 2006, 1924 who died in 2008, and 1968 who died in 2010 who met study inclusion criteria. Median survival was 8 months for those who died in 2006, 7 months in 2008, and 9 months in 2010. The difference in survival between 2008 (pre-bevacizumab) and 2010 (post-bevacizumab) was highly significant. This difference is unlikely to be due to improvements in supportive care in this short interval, because there was no significant difference (P = .4440) between patients who died in 2006 versus those who died in 2008. Between 2008 and 2010, a statistically significant improvement in survival was seen in all age groups except those patients aged 18 to 39 years.
Glioblastoma is the most common primary brain tumor in adults. Despite recent advances in treatment, tumor recurrence after initial therapy is inevitable and typically leads to death within a matter of months. Treatment options at the time of glioblastoma recurrence have traditionally been associated with only modest response rates and survival times. Bevacizumab, a humanized monoclonal antibody that inhibits vascular endothelial growth factor A, was approved by the US Food and Drug Administration (FDA) for use in recurrent glioblastoma in May 2009, and subsequently has been widely used for this indication in the United States.
The accelerated FDA approval of bevacizumab for recurrent glioblastoma was based on the results of 2 single-arm trials, AVF3708g and National Cancer Institute (NCI) 06-C-0064E. Both trials showed that bevacizumab was well tolerated and associated with previously unprecedented rates of radiographically defined tumor response, but by nature of their single-arm designs, these trials could not definitively demonstrate that bevacizumab use positively impacted patient survival. To date, no large placebo-controlled trials of bevacizumab for recurrent glioblastoma have been performed, leaving open the question of to what extent bevacizumab therapy prolongs survival. Instead, recent clinical trials have focused on the efficacy of bevacizumab as part of up-front therapy for newly diagnosed glioblastoma,[4, 5] in combination with other treatments in recurrent glioblastoma,[6-10] and in patient populations such as children, or adults with World Health Organization grade III glioma. However, the initial optimism regarding bevacizumab in treatment of glioblastoma has been tempered by concerns that radiographic response to treatment may in some cases simply be “pseudoresponse,” a decrease in contrast-enhancement without a decrease of tumor activity as the tumor assumes a more invasive phenotype.[13-15] These concerns were magnified after the FDA revoked its previous approval of bevacizumab for metastatic breast cancer. As is the case for glioblastoma, FDA approval of bevacizumab for metastatic breast cancer had been originally granted on an accelerated basis following promising early data but without clear demonstration that it improved patient survival.
This study was undertaken to evaluate whether glioblastoma survival in the United States, as recorded by the cancer registries comprising the Surveillance, Epidemiology, and End Results (SEER) Program, increased following the FDA approval of bevacizumab. Previous SEER studies have successfully demonstrated an improvement in population-level survival after the FDA approval of temozolomide for newly diagnosed glioblastoma.[17, 18] We hypothesized that patients who died in 2010 would have lived significantly longer with disease than patients who died in 2008, because patients who died in 2010 would have had access to bevacizumab therapy. Further, we hypothesized that survival times of patients who died in 2006 and 2008 would have been similar, indicating that incremental improvements in supportive care, surgery, and radiation in the absence of novel therapeutic options were insufficient to improve glioblastoma survival over a 2-year period.
MATERIALS AND METHODS
Patient Identification and Selection
The SEER Program currently collects cancer incidence and survival data on 18 geographic areas in the United States which together represent approximately 28% of the US population. We requested and received permission from the NCI to use the most recent SEER data set, released April 2013. The SEER data was then used to identify patients 18 years of age and older diagnosed with glioblastoma from 2000 to 2010. For the purposes of this study, glioblastoma was indicated by International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3) codes 9440/3 “Glioblastoma, NOS,” 9441/3 “Giant cell glioblastoma,” and 9442/3 “Gliosarcoma.” Additional inclusion criteria included microscopic confirmation of disease, actively followed patients, known age at diagnosis, and case inclusion in the SEER research database. Exclusion criteria included patients identified only at autopsy, patients alive without known survival time, and patients with other cancer diagnoses prior to glioblastoma. Once all SEER exclusion criteria had been applied, the analysis group was limited to patients who died in 2006, 2008, or 2010. Figure 1 is a flowchart of the cohort selection process.
Because bevacizumab is much more frequently used at the time of progression than at initial diagnosis, year of death rather than year of diagnosis was selected as the proxy for potential bevacizumab exposure. Bevacizumab was approved in May 2009, and thus patients who died in 2008 and 2010 were considered to have been treated in the pre-bevacizumab and post-bevacizumab eras, respectively. Chemotherapy information is not included in SEER data, so bevacizumab use could not be determined on an individual patient basis. In order to determine whether a statistically significant change in population-level survival could occur in a 2-year period which did not include the approval of a novel glioblastoma therapy, survival of patients who died in 2006 was compared to patients who died in 2008. This study was determined to be exempt from review by the Mayo Clinic Institutional Review Board, because it used only publicly available de-identified patient data.
Outcome and Statistical Analysis
The primary outcome of interest for this study was survival time, measured as a whole number of completed months of life from initial diagnosis to death, for patients who died during each year of interest. In SEER data, an individual patient survival time of 6 months could indicate a survival time between 6 months exactly and 1 day prior to 7 months of survival. From a population perspective, a median survival time of 6 months should be considered as more closely approximating 6.5 months than 6 months exactly given this variability.
SEER*Stat software was used to generate case listing data containing patient-specific demographic information such as age at diagnosis and treatment information such as whether radiation therapy was administered as part of the initial treatment plan and extent of initial surgical intervention. Age at time of diagnosis was subdivided into the age groups 18 to 39 years, 40 to 54 years, 55 to 69 years, and 70 or more years of age. All other demographic variables were taken directly from the SEER data set. A composite race/ethnicity variable was created dichotomizing patients as “White Non-Hispanic” and “All Other.” Radiation therapy use was dichotomized into patients confirmed to have received radiation therapy regardless of modality and patients not confirmed to have received therapy (whether it was not recommended, recommended but refused, or radiation status was unknown). Initial surgical intervention type was recoded into gross total resection (GTR), all other surgery, and no surgery. Patients who did not receive surgery were diagnosed with and treated for presumed glioblastoma during life but histologic confirmation was not obtained until autopsy. In order to evaluate survival in a patient group similar to that included in the European Organization for Research and Treatment of Cancer (EORTC) and National Cancer Institute of Canada Clinical Trials Group (NCIC) trial which established temozolomide as part of the standard of care for newly diagnosed glioblastoma, a “Stupp Cohort” subset was identified including only patients aged 18 to 70 years who underwent tissue confirmation of glioblastoma prior to death and were treated with a radiation-containing initial treatment regimen.
Descriptive statistics of median survival and interquartile range were generated for each year of death. Differences in survival distributions between 2006 and 2008, 2006 and 2010, and 2008 and 2010 were evaluated with Mann-Whitney U tests, given non-Gaussian survival time distributions. As an exploratory measure, the log-rank test was used to evaluate survival comparisons that were not significant in Mann-Whitney U analysis. Baseline demographic and treatment characteristics of the death year groups were evaluated with the chi-squared test for categorical outcomes and the Mann-Whitney U test for continuous outcomes with non-Gaussian distributions. Subpopulations of interest were selected for survival evaluation based on previous data demonstrating clear demographic or treatment associations with survival, and thus tests for interaction were not performed prior to subgroup analysis. Two-sided statistical tests with significance thresholds of P ≤ .05 were used in all analyses. All statistical analysis of the SEER case listing data was performed in JMP version 9.0 software (SAS Institute). The survival figure was generated in R software.
In total, 5607 adult patients with glioblastoma identified in SEER met criteria for inclusion in survival analysis. By year, 1715 patients (30.6%) died in 2006, 1924 patients (34.3%) died in 2008, and 1968 patients (35.1%) died in 2010. Patient demographic characteristics for patients who died in each year of interest are displayed in Table 1. The 3 cohorts were demographically similar aside from years of diagnosis. Details of initial tumor treatment are shown in Table 2. The rate of GTR fell from 2006 to 2008 and from 2008 to 2010, with a corresponding increase in other surgery types.
Table 1. Patient Demographic Characteristics by Year of Death
Year of Death
(n = 1715)
(n = 1924)
(n = 1968)
Age, median (interquartile range)
Age categories, N (%)
70 y and older
Sex, N (%)
Race, N (%)
Asian or Pacific Islander
American Indian/Alaska Native
Hispanic origin, N (%)
Year of diagnosis, N (%)
2004 or earlier
Table 2. Initial Tumor Treatment by Year of Death
Year of Death
(n = 1715)
(n = 1924)
(n = 1968)
Surgery, N (%)
Gross total resection
Radiation therapy, N (%)
Stupp Cohort, N (%)
The median survival time at death was 8 months (interquartile range [IQR] = 13) for the group as a whole, 8 months (IQR = 12) for those that died in 2006, 7 months (IQR = 13) for those who died in 2008, and 9 months (IQR = 14) for those who died in 2010. The difference in survival between 2006 and 2008 was not statistically significant (P = .4440), whereas the differences in survival between patients who would not have had access to bevacizumab (2006 and 2008 patient groups) and those who would have bevacizumab access (2010 group) was highly significant (each P < .0001). Figure 2 displays the proportion of patients alive with disease at monthly intervals up to 24 months before death in 2006, 2008, or 2010.
Table 3 shows median survival in 2006, 2008, and 2010 subdivided by patient characteristics. No age subgroup experienced a significant improvement in survival between 2006 and 2008, whereas significant differences in survival between 2008 and 2010 were seen in every age subgroup except the subgroup of those aged 18 to 39 years. Further, in 2008, the longest-lived 25% of patients in the subgroup of those aged 18 to 39 were alive with disease 32 months or longer prior to death, whereas in 2010, the longest-lived 25% were alive with disease 41 months or longer prior to death. The difference in survival between 2008 and 2010 in this age subgroup was statistically significant (P = .0342) when analyzed with the log-rank test, which is more sensitive to differences in groups at later time points. Race/ethnicity was analyzed as White Non-Hispanic versus All Others, given the high proportion of patients falling into the White Non-Hispanic group. Both groups saw an increase in median survival from 7 months in 2008 to 9 months in 2010. This improvement was significant in the White Non-Hispanic group and trended toward significance in the All Other group.
Table 3. Survival With Disease in Months by Year of Death and Selected Demographic and Treatment Factors
2006 (n = 1715)
2008 (n = 1924)
2010 (n = 1968)
2006 vs 2008 P Value
2008 vs 2010 P Value
Abbreviations: GTR, gross total resection; IQR, interquartile range.
All patients, median (IQR)
Age category, median (IQR)
70 and older
Race/ethnicity, median (IQR)
Surgery, median (IQR)
Radiation therapy, median (IQR)
“Stupp Cohort,” median (IQR)
Among patient groups defined by initial surgical treatment, median survival was longest in the patient group that underwent GTR, shorter in patients who underwent other surgeries, and shortest in the group that was treated without tissue diagnosis during life. The GTR surgical subgroup was the only demographic or treatment-defined patient subgroup in which survival significantly increased between 2006 and 2008. Survival in the GTR subgroup did not increase significantly between 2008 and 2010 (P = .4561), although it did approach significance in log-rank testing (P = .0594). During this same time period, survival significantly improved in the “Other Surgery” group, and a strong trend toward improvement was observed in the group in which tissue diagnosis was obtained postmortem.
In the Stupp Cohort of patients aged 18 to 70 years who were treated with surgery and radiation, median survival was 13 months in 2006, 14 months in 2008, and 15 months in 2010, although only the change in the latter time interval was statistically significant. In patients who did not qualify for the Stupp Cohort, median survival was identical at 4 months in both 2008 and 2010, but a statistically significant improvement over this interval was driven by an increase in relatively long-term survivors. In 2008, the longest-lived 25% of patients not qualifying for the Stupp Cohort were alive for 8 months or longer, while in 2010, the longest-lived 25% were alive 11 months or longer.
In summary, our analysis of the SEER registry indicates that following FDA approval of bevacizumab for treatment of recurrent glioblastoma in May 2009, overall survival for patients with glioblastoma improved. In this analysis, patients with glioblastoma who died in 2010 (and who had access to bevacizumab) had lived with disease significantly longer than patients who died in 2008 (before FDA approval of bevacizumab). This improvement in survival was not limited to a narrowly defined patient group; all adult patient age groups appeared to benefit. The magnitude of the improvement in survival between 2008 and 2010, although modest, was comparable to the 2.5-month improvement in median survival that led to FDA approval of temozolomide for newly diagnosed glioblastoma. Because SEER does not contain information about chemotherapy usage, chemotherapy practice patterns in each year of interest must be inferred. Prior to FDA approval of bevacizumab for recurrent glioblastoma, its use was limited to clinical trials and off-label therapy. After FDA approval, bevacizumab was enthusiastically received into clinical practice. Although not all patients with recurrent glioblastoma receive bevacizumab therapy, at a minimum its use is almost universally considered in this clinical setting. Thus, this analysis does not provide a true comparison of survival in patients with and without bevacizumab. Instead, it compares population outcomes in a time when physicians had the option of using bevacizumab in clinically selected patients to an earlier time in which they did not have this option.
Patients with glioblastoma receive multidisciplinary care, including treatment by oncologists, neurologists, neurosurgeons, and radiation oncologists. Although each of these fields is constantly advancing, the approval of bevacizumab was the change most likely to have affected glioblastoma survival between 2008 and 2010. A previous SEER analysis of glioblastoma survival demonstrated a long period without any notable improvement, despite presumed incremental improvement in supportive care, surgical technique, and radiation therapy methods. More recently, SEER studies demonstrated that survival improved following FDA approval of temozolomide, which was last chemotherapy approved for use in glioblastoma prior to bevacizumab approval.[17, 18]
The longer survival observed for patients who had access to bevacizumab (2010 group) compared with those who did not have access (2008 group) was unlikely to be related to improvements in supportive care over this 2-year interval. In this regard, we observed no statistically significant improvement in survival between patients who died in 2006 and those who died in 2008, a time interval equal to that in which the significant change between 2008 and 2010 was documented. In fact, median survival was 1 month greater in 2006 than in 2008, a finding that likely reflects the granularity of the SEER data more than a clinically meaningful change, because the difference was not statistically significant and the survival curves for these years in Figure 2 are closely apposed.
This study provides the strongest evidence to date that bevacizumab therapy improves survival in patients with glioblastoma. A variety of factors have made this survival benefit difficult to demonstrate in clinical trials. First, clinical trials of bevacizumab in recurrent glioblastoma have often been confounded by crossover to bevacizumab at the time of tumor progression, or they have compared 2 bevacizumab-containing regimens. Most brain tumor clinicians believe, based on clinical experience, that bevacizumab therapy is beneficial to patients with progressive glioblastoma, and thus at this point a randomized trial of bevacizumab versus placebo without crossover would be difficult, if not impossible, to conduct in countries in which bevacizumab is readily available. In addition, it had been hypothesized that the addition of bevacizumab to radiation and temozolomide at the time of glioblastoma diagnosis would prolong survival relative to radiation and temozolomide alone, and thus demonstrate the survival benefit of bevacizumab without the need to conduct a randomized trial in the progressive glioblastoma setting. This hypothesis led to 2 large international randomized placebo-controlled trials, neither of which demonstrated an overall survival benefit of bevacizumab in this setting.[4, 5] On the basis of the negative newly-diagnosed glioblastoma trials and the positive results of this analysis, it is not possible to discern whether bevacizumab is equally effective in the newly diagnosed and progressive glioblastoma setting, or selectively effective in progressive glioblastoma.
A limitation of this analysis is that SEER does not include patient-level data such as performance status and MGMT methylation status, which are of great importance in patients with progressive brain tumors. There are no data to suggest that these characteristics changed on a population level over the period considered by this analysis, and thus they should not confound year-to-year comparisons. Furthermore, some SEER-reported data are not fully objective, and may confuse subgroup comparisons if standards change over time. For example, GTR in SEER is based on surgeon report. Between 2006 and 2010, the rate of reported GTR fell steadily, and the GTR patient subgroup was the only demographic or treatment subgroup in which survival improved between 2006 and 2008. It is unlikely that surgical progress truly resulted in a lower GTR rate. Instead, it is probable that surgeons became more circumspect in estimating extent of resection as postoperative magnetic resonance imaging became more routine, and earlier GTR groups contain higher proportions of patients who truly had residual disease after surgery than later GTR groups. These issues preclude the identification of clear patient subsets that do and do not benefit from bevacizumab treatment. Further studies, ideally prospective trials, will be needed to inform patient selection.
In conclusion, FDA approval of bevacizumab was associated with a statistically significant improvement in glioblastoma survival in the United States. Although the observed increase in survival was modest on a population basis, this analysis likely underestimates the change in median survival seen in the subpopulation of patients clinically selected for bevacizumab therapy. Although a cause-and-effect relationship between bevacizumab use and increased survival cannot be definitively proven in a retrospective population-based analysis, this study strongly suggests that bevacizumab is a useful treatment for recurrent glioblastoma, and that the initial radiographic response data that led to FDA approval of bevacizumab for recurrent glioblastoma successfully predicted a benefit in overall survival for this patient population.