Previous studies have shown that the overall incidence of primary malignant brain tumor is greatest in Caucasians, although survival is better in African Americans. The objective of this study was to examine racial differences in survival after diagnosis with primary malignant brain tumor in a population-based sample of patients while adjusting for prognostic variables that differ by race.
The authors analyzed 21,493 patients (20,493 Caucasians and 1000 African Americans) who were diagnosed with primary malignant brain tumors from 1973 to 1997 (with follow-up through 1999) from the population-based Surveillance, Epidemiology, and End Results (SEER) Program. Chi-square tests were used to determine statistical significance of prognostic variables and race (using two-sided P values). Kaplan–Meier and Cox proportional hazards models were used to assess survival differences by race.
The univariable model for race showed no survival difference by race. The multivariable model demonstrated that African American patients were at a 13% increased risk of death from any cause compared with Caucasian patients. The racial difference was explained further by an interaction between race and surgery type in which there was an increased risk of death for African American patients who underwent subtotal resections or surgery not otherwise specified compared with Caucasian patients who underwent the same procedures.
Primary malignant brain tumor is among the most lethal malignancies, with a median survival of less than 2 years.1, 2 An estimated 18,300 patients will be newly diagnosed with primary malignant brain tumors and other nervous system tumors in 2003, and 13,100 deaths will occur.3 The incidence rate for malignant brain tumors and other central nervous system tumors is 6.6 per 100,000 in general, 8.0 per 100,000 for males and 5.4 per 100,000 for females,4 with a mean age at diagnosis in the middle to late 50s.5 The incidence of primary malignant brain tumor in general is at least twice as high for Caucasians compared with African Americans.1, 4–6 No risk factor that accounts for large numbers of brain tumors has been identified, although there has been some evidence for inherited factors, with a family history of malignant disease in approximately 16% of patients studied with brain tumors.2, 7, 8
Although the incidence is significantly higher in Caucasians compared with African Americans, African Americans seem to show improved survival after diagnosis and, hence, lower death rates.1, 4, 9 For all individuals who are diagnosed with primary malignant brain tumors, regardless of race, survival declines rapidly after the first 2 years from diagnosis, with a 5-year relative survival rate of 20–30%.4, 5 Previous studies of survival after diagnosis with brain tumor have been able to identify important prognostic variables. Histologic tumor type, age at diagnosis, race, tumor location in the brain, and surgery type are all important predictors of survival, with large variations in survival according to histologic tumor type.8, 10, 11 Treatment is also an important predictor of survival, with the majority of patients undergoing surgery alone or undergoing surgery in combination with radiotherapy and/or chemotherapy,2, 7 although only surgery and radiotherapy have been shown to help improve survival for patients with various tumor types. Racial differences in survival have been studied previously in female breast, colon, prostate, lung, bladder and gynecologic malignancies,12–17 but few studies have examined fully the differences in survival of brain tumor by race.10, 11, 18, 19 The objective of the current study was to examine racial differences in survival after diagnosis with primary malignant brain tumor in a population-based sample of individuals while adjusting for prognostic variables that differ by race.
MATERIALS AND METHODS
There were 21,493 patients studied, including 20,493 Caucasians and 1000 African Americans, who were diagnosed with first primary malignant brain tumors between 1973 and 1997 (with follow-up through December 31, 1999) whose information was attained from the population-based SEER Program data base. Patients with tumors of borderline malignancy were excluded. SEER is a national cancer surveillance program that collects information from all incident cancer cases from 14 population-based cancer registries and 4 supplemental registries covering approximately 26% of the United States population and is representative of national demographic characteristics.20 For the purposes of the current analysis, only data from the nine standard SEER sites were used. The nine standard sites are Metropolitan Atlanta, Metropolitan Detroit, the San Francisco/Oakland Standard Metropolitan Statistical Area (SMSA), the Seattle/Puget Sound SMSA, and the states of Connecticut, New Mexico, Hawaii, Utah and Iowa. Data have been collected from these main SEER sites since 1973 (1974 for Seattle/Puget Sound and 1975 for Atlanta), whereas the other five population-based registries and the four supplemental sites did not begin collecting data until the 1980s and 1990s. In addition to collecting demographic, treatment, and clinical information at the time of diagnosis, patients were followed annually for survival status.
Prognostic variables of interest included race, age at diagnosis, surgery type, gender, histologic tumor type, and primary site of tumor in the brain. Race was categorized as Caucasian or African American, as recorded in the medical chart of each patient. All other races were excluded from the current analysis because of small sample sizes. Patient age at diagnosis was categorized as 18–39 years, 40–49 years, 50–59 years, 60–69 years, and 70 years or older for univariable analyses but was used as a continuous variable in the final multivariable modeling of survival. Surgery type was defined as biopsy only (incisional, needle, or aspiration); subtotal resection (local tumor destruction, partial resection of tumor); total resection (total resection of tumor, partial resection of primary tumor site in the brain, lobectomy, radial resection); and surgery, not otherwise specified (NOS). Because these data were limited to the first course of treatment, analyses by treatment other than surgery type were not attempted. Primary site of tumor in the brain was classified as the following: cerebrum; frontal lobe; temporal lobe; parietal lobe; occipital lobe; ventricle, NOS; cerebellum, NOS; brain stem; overlapping lesion of brain; and brain, NOS.
In the SEER data base, histologic tumor type is coded using International Classification of Diseases for Oncology (ICD-O) codes. The 11 specific subtypes of primary malignant brain tumors that were evaluated with their corresponding ICD-O morphology codes and names are shown in Table 1.
Table 1. Histologic Tumor Types and Classifications
Name (ICD-O code)
ICD-O: International Classification of Diseases for Oncology; NOS: not otherwise specified.
Medulloblastoma, NOS (9470); desmoplastic nodular medulloblastoma (9471); medullomyoblastoma (9472)
Ependymoma Grade 2–3
Ependymoma, NOS (9391); anaplastic ependymoma (9392); papillary ependymoma (9393)
First, differences by race and the prognostic variables of interest were assessed for statistical significance using chi-square tests, which generated two-sided P values. Differences in proportion by race for each category of each prognostic variable also were calculated. Survival for all analyses was defined as the time from diagnosis until death from any cause. Kaplan–Meier methods were used to estimate crude survival by race. The log rank test was used to assess the statistical significance of the differences between survival curves.21 To assess the prognostic significance of variables of interest, univariable Cox proportional hazards models22 were used to estimate a hazard ratio (HR) and a 95% conference interval (95% CI). Multivariable Cox proportional hazards models were used to calculate adjusted HRs and their 95% CIs to assess the importance of race as an independent predictor of survival after adjusting for other factors and interactions between those factors and race. The significance of interactions between race and all prognostic variables was assessed using an improvement in the fit log-likelihood ratio test. All prognostic variables were tested for agreement with the proportional hazards assumption of the survival model.23 All data were analyzed using SAS Version 8.1 (SAS Institute, Cary, NC) and SPlus Version 6.0 Professional (Insightful Corp., Seattle, WA) statistical software packages.24, 25
The average age at diagnosis for Caucasian patients was 54.8 years (95% CI, 54.6–55.0 years), and the average age at diagnosis for African American patients was 50.6 years (95% CI, 49.5–51.6 years). A t test for a difference in average age at diagnosis by race showed a statistically significant difference according to race (P < 0.0001), with a greater proportion of African American patients diagnosed at a younger age (49 years or younger). African American patients and Caucasian patients with first primary brain tumors had the same crude median survival of 10 months (Fig. 1A).
Table 2 shows the distribution of potential prognostic variables by race, as well as differences in proportions by race for each category of each variable. Distributional differences by race were significant for surgery type, histologic tumor type, primary site of tumor in the brain, and SEER registry site. The distribution of individuals by gender did not differ significantly between Caucasians and African Americans, although there were more men than women diagnosed with primary malignant brain tumors in both racial groups. More African American patients underwent a biopsy only, and more Caucasian patients underwent surgery, NOS. There was no difference according to race among patients with missing surgery type information (14% missing). A greater proportion of African American patients had glioma, other; astrocytoma other; astrocytoma Grade 1; oligodendroglioma Grade 3;, medulloblastoma; and ependymoma Grade 2–3 compared with Caucasian patients. African Americans had fewer tumors of the frontal lobe; parietal lobe; occipital lobe; and brain, NOS. However, most of the differences according to histologic tumor type and primary tumor site in the brain were small. SEER registry site varied considerably by race. Metropolitan Detroit and Metropolitan Atlanta had the greatest proportions of African Americans compared with any of the other registries; the variable was used in the multivariable models for adjustment purposes only.
Table 2. Difference by Race in Each Prognostic Variable for Patients Diagnosed between 1973 and 1997 with Primary Malignant Brain Tumors
NOS: not otherwise specified; SEER: Surveillance, Epidemiology, and End Results Program.
P values were determined with a test for differences in proportions for the categorized prognostic variables: differences in proportion indicate the proportion of Caucasians minus the proportion of African Americans, with a negative difference indicating that African Americans had greater proportions compared with Caucasians.
Surgery type (14% of patients missing)
Histologic tumor type
Astrocytoma Grade 1
Astrocytoma Grade 2
Astrocytoma Grade 3
Glioblastoma multiforme (astrocytoma Grade 4)
Oligodendroglioma Grade 3
Ependymoma Grade 2–3
Primary site of tumor in the brain
Overlapping lesion of brain
SEER registry site
Univariable, unadjusted Cox proportional hazards models were used to assess the importance of each prognostic variable (Table 3). The unadjusted HRs for race and gender showed no difference between Caucasians and African Americans or between males and females, respectively. The risk of death increased with increasing age at diagnosis. Patients who underwent subtotal resection; total resection; or surgery, NOS were at a significantly decreased risk of death from any cause compared with patients who underwent biopsy only. Patients with tumors in any of the histologic tumor type categories were at a significantly decreased risk of death compared with patients who had glioblastoma multiforme (astrocytoma Grade 4). Patients who had tumors in the ventricle, NOS; cerebellum, NOS; and brain stem were at decreased risk of death compared with patients who had tumors in the frontal lobe. Patients with tumors in all other primary sites in the brain were at increased risks of death compared with patients who had frontal lobe tumors.
Table 3. Hazard Ratio Estimates, with 95% Confidence Intervals, from Univariable Cox Proportional Hazard Survival Models for Patients Diagnosed with Primary Malignant Brain Tumors between 1973 and 1997
Variable of interest
95% CI: 95% confidence interval; NOS: not otherwise specified.
Age at diagnosis (yrs)
Histologic tumor type
Glioblastoma multiforme (Astrocytoma Grade 4)
Astrocytoma Grade 1
Astrocytoma Grade 2
Astrocytoma Grade 3
Oligodendroglioma Grade 3
Ependymoma Grade 2–3
Primary site of tumor in brain
Overlapping brain lesion
Multivariable Cox proportional models were used to assess confounding and interaction effects between race and other prognostic factors. The main effects only multivariable Cox proportional hazards model showed a 13% (race: HR = 1.13; 95% CI: 1.05–1.21) increased risk of death for African American patients compared with Caucasian patients (adjusted for gender, age at diagnosis, surgery type, histologic tumor type, primary site of tumor in the brain, and SEER registry site). Figure 1B shows the survival curves by race for the fully adjusted Cox models. The only significant interaction with race was found between race and surgery type. With the addition of the interaction of race and surgery type, the model was improved significantly compared with the main effects model only (log-likelihood ratio = 12.7 with 3 degrees of freedom; P < 0.01).
To investigate further the significance of this interaction, multivariable Cox proportional hazards models stratified by surgery type were performed; the HR of race for each of the four models is shown in Table 4. All multivariable Cox proportional hazard models were adjusted for the following variables: gender, age at diagnosis, histologic tumor type, primary site of tumor in the brain, and SEER registry site. The risk of death for African Americans who underwent biopsy only was 14% lower compared with Caucasians who underwent biopsy only, although the difference was not statistically significant. The risk of death for African Americans who underwent subtotal resection was 38% higher compared with Caucasians who underwent subtotal resection. The risk of death for African Americans who underwent total resection was 4% higher compared with Caucasians who underwent total resection, but the difference was not significant. The risk of death for African Americans who underwent surgery, NOS was 22% higher compared with Caucasians who underwent surgery, NOS.
Table 4. Race Hazard Ratio Estimates, with 95% Confidence Intervals, from a Multivariable Cox Proportional Hazards Model, Stratified by Surgery Type, for Patients Diagnosed with Primary Malignant Brain Tumors from 1973 to 1997a
All models were adjusted for the following prognostic variables: gender, age at diagnosis, histologic tumor type, primary site of tumor in the brain, and Surveillance, Epidemiology, and End Results Survey registry site.
Compared with Caucasian patients, African American patients had a 13% increased risk of death due to any cause, after adjusting for gender, age at diagnosis, surgery type, histologic tumor type, primary site of tumor in the brain and SEER registry site. Racial differences were explained further by an interaction between race and surgery type in which there was an increased risk of death from any cause for African Americans who underwent subtotal resection or surgery, NOS compared with Caucasians who underwent the same procedures. These data contradicted previous reports that African Americans had better or equal survival compared with Caucasians.1, 4, 12–16 These data did validate findings of previous studies of prognostic factors in brain tumor survival by showing that age at diagnosis, histologic tumor type, surgery type, and primary site of tumor in the brain had significant effects on the race risk estimates.2, 7, 8, 10, 11
Previous studies that evaluated racial differences in survival after diagnosis with primary malignant brain tumor showed improved survival after diagnosis and lower death rates1, 4, 9 but did not adjust for many prognostic variables of interest that vary by race. Age at diagnosis, histologic tumor type (which is correlated with tumor grade), surgery type, and primary site of tumor in the brain have been shown to be predictive of a patient's course of treatment and prognosis after diagnosis. In the current study, as a consequence of this adjustment (with the addition of SEER registry site), it was shown that African American patients were at a 13% increased risk of death from any cause compared with Caucasian patients, not a decreased risk of death. With further adjustment for the interaction of race and surgery type, significant increased risk of death was still seen for African American patients who underwent subtotal resection or surgery, NOS compared with Caucasian patients in these categories. Thus, adjustment of survival risk by variables that differ significantly by race and their interactions with race should be performed to calculate the most clinically and statistically accurate estimates.
Although many variables were included in the models used in the current analysis, there remained many other variables that were possible confounders with race that either were not available in the SEER data for this analysis or were available but had a large proportion of patients classified as NOS or other. For example, it has been shown that many other variables vary by race, such as access to health care (insurance information), socioeconomic status (SES), chemotherapy, and radiation, all of which would have provided additional information for the current analysis. Although SEER does not collect health care information or SES, the zip code of residence was recorded for all patients. Zip code information was not used to impute SES, because such imputation likely would result in nondifferential misclassification and would more likely misclassify minority populations.26 SEER does record information on chemotherapy, radiotherapy, and cause of death; however, these variables were such that they precluded a meaningful interpretation. In particular, each of these variables was subject to misclassification bias. Cause of death information was gathered from death certificates, which are known to be incorrect in many cases.27 Models were also fit using brain tumor specific death instead of death from any cause, and very little difference was seen in the estimates obtained from these models compared with the models shown in Tables 3 and 4. Therefore, an absolute survival endpoint of death from any cause was used instead of brain tumor specific death. Radiotherapy and chemotherapy information in the SEER data base was gathered only from hospital records; because many patients did not receive chemotherapy and radiotherapy treatments in hospitals, but rather received them at off-site centers and clinics, the presence of these data was inconsistent. Approximately 50% of the patients in this study were classified as surgery, NOS, although there was no statistically significant differential use of this category by race. For patients who were classified with glioma, other; astrocytoma, other; or oligodendroglioma, other, there was differential use of these categories by race, with Caucasian patients classified more frequently with oligodendroglioma, other, and with African American patients classified more frequently with glioma, other, or astrocytoma, other. Because of this differential by race, all multivariable models were adjusted for histologic type of tumor.
The current study has many strengths. It is well known that the SEER data are of very high quality. The information collected on each individual patient was monitored meticulously for its accuracy. Because the current study was population-based, generalizability of the results to the greater United States population can be made. The large sample size (21,493 individuals) in this study provided enough power to detect even the smallest of differences in death from any cause associated with race for patients with primary malignant brain tumors. This large sample size also allowed us to examine very rare brain tumors, which are not studied easily in smaller, hospital-based studies. In addition, the current study used the most recent SEER public-use data (2002 release).
In summary, no survival difference by race was apparent until adjustments were performed for important prognostic variables that differ by race. After adjustment for gender, age at diagnosis, surgery type, histologic type of tumor, primary site of tumor in the brain, and SEER registry site, African American patients were at a 13% increased risk of death from any cause compared with Caucasian patients. Racial differences were explained further by an interaction between race and surgery type in which there was an increased risk of death from any cause for African Americans who underwent subtotal resection or surgery, NOS compared with Caucasians who underwent the same procedures.