Glioblastoma multiforme (GBM) is the most frequent primary brain tumor in adults. Temozolomide was rapidly incorporated into first-line treatment following the publication of the pivotal European Organization for Research and Treatment of Cancer–National Cancer Institute of Canada phase 3 trial in 2005. However, in the trial, enrollment was limited to younger patients with good performance status. Therefore, this study performed a population-based survival analysis of patients with newly diagnosed GBM covering the period before and after the introduction of temozolomide.
Survival statistics and clinical and demographic variables were extracted from the Survival, Epidemiology and End Results Database for patients diagnosed with GBM from 2001 to 2007. Mean regional income for each patient was also collected. Survival was analyzed using the Kaplan-Meier method and proportional hazard models.
A total of 13,003 adult patients diagnosed with a GBM were identified. Prognostic variables included age <70 years, use of radiation, gross total resection, and residence in a high-income district (P < .001). Between 2001 and 2007, the median survival time increased from 7 to 9 months for the entire population. The 1-year survival increased from 29% to 39%. Prognosis of patients aged 70 or more years did not improve over this time. Over the study period, the absolute disparity in 1-year survival between low- and high-income districts increased from 6.6% to 10.1%.
Glioblastoma multiforme (GBM) is the most common primary brain tumor in adults. Disease incidence rises with age, with a mean age of presentation of 53 years.1 GBM is resistant to many anticancer therapies and is often rapidly fatal. For several decades, the standard of care was maximal safe resection followed by radiation therapy. In 2004, the initial results of the European Organization for Research and Treatment of Cancer and National Cancer Institute of Canada (EORTC-NCIC) phase 3 trial were presented, which demonstrated an overall survival (OS) advantage when concurrent and adjuvant temozolomide was added to radiation therapy. In that trial, enrollment was limited to patients with a good performance status and age <70 years.2
Because the EORTC-NCIC study examined a highly selected cohort, it is unknown whether the introduction of temozolomide affected GBM prognosis within the broader population. Furthermore, the results of the EORTC-NCIC trial have been questioned due to contamination with lower grade tumors.3 We therefore performed a population-based survival analysis to test the hypothesis that OS of subjects with newly diagnosed GBM has improved over the past decade. We further hypothesized that improvements in prognosis may be modulated by socioeconomic conditions.
MATERIALS AND METHODS
This study used the Survival, Epidemiology and End Results (SEER) database, which collects data from 17 population-based registries that cover approximately 26% of the population of the United States. Each individual registry collects data relating to patient demographics, primary tumor site and morphology, and first course of treatment (received within 4 months of diagnosis). These registries were selected to be a part of SEER due to their completeness and adequate representation of minority populations.4
From the SEER database, we identified adult patients (above 20 years of age) diagnosed with a GBM. Because the SEER population base was broadened in 2001, by inclusion of additional cancer registries, only cases diagnosed on or after 2001 were included. Cases with multiple primaries, unknown radiation or surgical treatment, primary spinal chord neoplasm, or those only diagnosed at autopsy were excluded.
Data on patient demographics, tumor grade, primary site, year of diagnosis, surgical and radiation treatment received within 4 months of diagnosis, and survival until death or follow-up as of December 31, 2007, were extracted. In addition, mean household income of the geographical location where each case lived was collected. For the purposes of analysis, continuous variables (eg, age and median geographical income) were converted into categorical variables.
Statistical analyses were performed using the Stata statistical package, version IC 11.1 (Stata, College Station, Tex). We used multiple logistic regression to model factors associated with receiving radiation therapy and surgical resection. Chi-square tests were used to assess correlations between categorical variables. Overall survival was defined from the time of initial diagnosis to the date of death and was calculated using the Kaplan-Meier method. The effects of demographic, pathologic, and treatment variables were tested using the log-rank test. Multivariate analysis was performed with a Cox proportional hazard model. Assumptions for performing the Cox analysis were verified using the log-log and Schoenfeld residuals methods. Patients who had missing data, and patients with multiple primary neoplasms were excluded from the multivariate analysis. All P values were 2-sided, and a P < .05 was considered statistically significant.
At total of 15,888 patients with GBM were identified through the database. Subjects were excluded if they had multiple primaries, unknown radiation treatment or surgery, primary spinal cord neoplasm, or were patients diagnosed at autopsy, leaving 13,003 subjects for further analysis. In total, 1668, 1765, 1845, 1958, 1911, 1860, and 1996 subjects were diagnosed in 2001, 2002, 2003, 2004, 2005, 2006, and 2007, respectively. The demographic characteristics of the subjects are presented in Table 1.
Table 1. Demographic Characteristics of Glioblastoma Population
Entire Population (%)
n = 5278 (%)
n = 7725 (%)
Sex (% male)
Gross total resection
Extent of resection unknown
The database was updated through December 31, 2007. At that point, 81.9% of all subjects had died, and, among those who died, GBM was listed as the cause of death in 94% of cases. For the entire population, median life expectancy estimated from the Kaplan-Meier curve was 8 months. The 1- and 2-year estimated OS rates were 35% and 13%, respectively.
Age, degree of surgical resection, and the use of radiation therapy were associated with improved survival (Fig. 1). All these associations were significant at the level of P < .0001. Median OS of males was 2 months longer than females (9 vs 7 months, respectively), P = .0002. Race was not significantly associated with all-cause mortality.
Survival improved over time, with median survival improving from 7 months to 9 months between 2001 and 2007. Likewise, there have been incremental improvements in 1- and 2-year survival rates (Fig. 2). This improvement in survival was confined to those under the age of 70 (Table 2). As expected, older patients had a significantly worse survival compared to younger patients. Older subjects were less likely to undergo a gross total resection (39.0% for those under 50 years vs 24.3% for those over 70 years, P < .001) and less likely to receive radiation therapy (81.4% vs 52.8%, respectively; P < .001) (Table 3). We therefore examined whether the prognosis for subjects aged ≥70 years, who had received optimal treatment (radiation therapy and gross total resection), had changed. Within this subpopulation, median survival time decreased from 9 months to 8 months over the decade, but this was not significant (P = .2).
Table 2. Median Overall Survival as a Function of Year of Diagnosis and Age
Median Survival (mo)
Median Survival (mo)
Median Survival (mo)
Table 3. Type of Surgery and Use of Radiation as a Function of Age at Time of Diagnosisa
Age Group, y
Surgery Performed (%)
Use of RT (%)
Abbreviations: GTR, gross total resection; NOS, not otherwise stated; RT, radiation therapy.
There was a significant association between type of surgery and age, and also between the use of radiation and age (both P < .001).
20-49 (n = 2299)
50-69 (n = 6378)
≥70 (n = 4326)
All ages (n = 13,003)
We also investigated whether the median household income of where the subjects lived influenced outcome. To this end, we evenly distributed the median annual household incomes into quartiles ($16,270-$40,779, $40,780-$46,199, $46,200-$55,299, $55,300-$79,890). Subjects living in districts with a high median household income lived on average 2 months longer compared to those living in areas with a low median household income (Table 4); P < .0001. There was no association between income quartile and average age. Patients living in districts with a higher median household income were also more likely to undergo a gross total resection (32.7% vs 30.4%; P < .001) and receive radiation therapy (72.2% vs 69.2%; P < .001).
Table 4. Relationship Between Overall Survival and Median Household Income of the Geographical Location Where Each Case Lived
Quartile 1 refers to the lowest income group (median household income $40,770 or less). Quartile 4 refers to the highest income group (median household income $55,340 or more).
There was a trend during the study period toward more extensive resection, with rates of patients undergoing biopsy-only decreasing from 20.6% in 2001 to 13.7% in 2007 (P < .001). The use of radiation did not change over this period, remaining at 70%.
We examined the interaction between these variables (sex, age at diagnosis, extent of surgical resection, use of radiation therapy, income quartile, and year of diagnosis) in a Cox proportional hazard ratio model. With the exception of sex, all variables remained significantly associated with OS (Table 5). Likewise, OS improved over time even within the subpopulation of patients “optimally treated” (ie, those who had undergone gross total resection and received radiation therapy). The findings suggest that the improvement in survival was not simply a function of improved surgery and more frequent referral to radiation treatment.
Table 5. Multivariate Analysis of Variables Associated With Overall Survival in Glioblastoma, from 2001 to 2007, Cox Model
95% Confidence Interval
Gross total resection
Surgery, not otherwise stated
We also investigated how prognosis has improved over time in the different economic areas. One- and 2-year survival has improved across all groups over time (P < .0002 for each of the quartiles assessed individually; Fig. 3). However, the disparity between lower income and higher income areas appears to be increasing over time
The results of this large population-based study demonstrate that prognosis in GBM has improved incrementally from 2001 and 2007, but only among subjects <70 years of age. Furthermore, we demonstrated that although the prognosis among patients in all geographic regions improved, those with a high median household income benefited the most. The “gap” between more affluent and more deprived areas appears to have widened between 2001 and 2007, although this did not reach statistical significance.
The strengths of the study include the large size of the database with its broad coverage of all age ranges and performance status, unlike studies based on single-institution experience or clinical trial participants. Our clear demonstration of the prognostic significance of surgical resection, age, and the use of radiation therapy, in keeping with the results of previous studies and randomized trials, further validates the database.
To the best of our knowledge, this is the first population-based study to demonstrate an improvement in OS for patients with GBM. Based on the results of single-arm, multi-institutional phase 2 studies, there has been recent speculation that prognosis of GBM is improving beyond that provided by temozolomide.5 Nonetheless, such studies are difficult to interpret due to the large influence that baseline characteristics such as age and performance status have on survival outcomes. By examining a very large population, the effect of these factors is likely nullified.
The reason for the improved prognosis in GBM is not revealed by our study, but is most likely related to the introduction of temozolomide in the upfront setting. When the initial results of the EORTC-NCIC study were announced at the annual meeting of the American Society of Clinical Oncology in June 2004, oncologists in the United States were able to prescribe temozolomide off-label, because it was approved by the US Food and Drug Administration (FDA) for refractory anaplastic astrocytomas in 1999. Formal FDA approval for upfront treatment of GBM was granted in 2005. Nonetheless, questions remain, because the trial's results were announced only in 2004, and the improvement in survival started at least in 2002. Possibilities include use of the drug prior to 2004 in both the upfront and recurrent settings based on the promising results of early phase trials, and improvements in surgical and radiotherapy technique. Although surgical technique appeared to improve during the course of this study (with more patients undergoing partial or gross total resection), year of diagnosis remained an independent variable in the multivariate analysis even after accounting for extent of surgery (Table 5). Another possible explanation is that the overall approach to the treatment of GBM (especially recurrent disease) is changing, becoming more aggressive and less nihilistic.
Based on our SEER data, estimated overall 2-year OS among cases who received radiation therapy was 16.7%, more than the 10.4% reported for the standard arm but significantly less than 26.5% reported for the experimental arm of the trial described in Stupp et al.2 This difference is partially explainable in terms of population characteristics. If patients over the age of 70 years are excluded, 2-year OS increases to 20.7%. When analyzed based on year of diagnosis, 2-year OS increased substantially over time, reaching 25.9% in 2005. We lack information on performance status and temozolomide usage. It is possible that if only patients with good performance status who received temozolomide were included in the SEER analysis, 2-year OS would be even higher.
It is disappointing to note there was no improvement in median survival for patients aged 70 or more years between 2001 and 2007. Advanced age is a well-known and powerful prognostic factor in this disease,6 suggesting that the biology of these tumors is different. It has been demonstrated that radiotherapy benefits the elderly to a similar degree as it does younger patients.7 Patients above the age of 70 were not enrolled into the EORTC-NCIC trial. In a large meta-analysis that assessed the use of chemotherapy in brain tumors from the era before temozolomide, elderly patients with high-grade gliomas benefited less than younger patients.8 In a single-institution retrospective case series of patients with GBM aged greater than 64 years, adjuvant chemotherapy did appear to benefit older patients.9 It is impossible to conclude from our study whether the prognosis of elderly patients failed to improve due to nonuse of temozolomide, lack of response to the drug in this population, lower quality health care, different biology, or some other factor.
An unexpected finding from this study is that the median survival time for patients ≥70 years who underwent a gross total resection and radiation therapy decreased from 9 months to 8 months. Although this was a nonsignificant finding, it is nevertheless surprising, especially because anecdotal reports suggest that patients aged ≥70 years frequently receive temozolomide in the United States, although no formal statistics exist. A possible explanation is that older patients may have poorer performance status. Temozolomide appeared to worsen survival in the small subgroup of patients in the EORTC-NCIC trial with poor performance status (World Health Organization grade 2). Alternatively, there may be an unknown confounder that we are not able to account for, which is a frequent problem with large data sets. Our results emphasize the need for research that specifically focuses on GBM in the elderly population.
Our finding of outcome disparities between patients living in districts with a lower versus higher median household income is a novel finding in GBM, but is consistent with trends in other diseases. In our study, subjects living in districts with a high median household income were more likely to receive radiotherapy and undergo a substantial resection. However, these findings do not fully explain the discrepancies. Low median income is generally associated with lower education levels, less social support, and a lower quality of health care. Moreover, given the high cost of temozolomide, it is likely that some uninsured patients living in low-income districts had impaired access to the agent. This may become a more pressing issue over the next few years as additional biologic agents such as bevacizumab are added to first- and second-line treatment for GBM. Although investigation of the causes of this disparity are beyond the scope of this study, it indicates that therapeutic interventions alone may be insufficient to ensure universal advances in health outcomes in GBM.
There are several limitations to the present study. A specific major limitation of the SEER database is lack of information regarding the use of chemotherapy, which likely contributed significantly to the improving prognosis of this patient population. Our survival analysis only includes patients diagnosed during a relatively short time span (2001-2007), during which the SEER population base was stable. Although the EORTC-NCIC phase 3 trial of upfront temozolomide was reported only in 2004, promising results from earlier smaller trials may have encouraged use of temozolomide in first- or second-line setting prior to this date. Population-based studies that use tools such as the SEER-Medicare linked databases, which include some chemotherapy information, are necessary to further investigate the impact of temozolomide and other agents on survival of patients with GBM in the population as a whole. Other known limitations of the SEER database include lack of information regarding performance status, neurological status, or symptoms; disease relapses; and subsequent treatments the patient may have received.4
In summary, we have demonstrated a small but incremental improvement in overall survival among patients with GBM during the first decade of the 21st Century. The improvement was less prominent among those living in areas with a low median household income. There has been no survival improvement among those above the age of 70 years. Further studies will seek to understand what factors are responsible for these differences between younger and older patients.
The Kimmel Cancer Center is supported by National Cancer Institute grant 2 P30 CA056036-09.