Impact of race on care, readmissions, and survival for patients with glioblastoma: an analysis of the National Cancer Database

Abstract Background The objective of this study was to explore racial/ethnic factors that may be associated with survival in patients with glioblastoma by querying the National Cancer Database (NCDB). Methods The NCDB was queried for patients diagnosed with glioblastoma between 2004 and 2014. Patient demographic variables included age at diagnosis, sex, race, ethnicity, Charlson–Deyo score, insurance status, and rural/urban/metropolitan location of zip code. Treatment variables included surgical treatment, extent of resection, chemotherapy, radiation therapy, type of radiation, and treatment facility type. Outcomes included 30-day readmission, 30- and 90-day mortality, and overall survival. Multivariable Cox regression analyses were performed to evaluate variables associated with race and overall survival. Results A total of 103 652 glioblastoma patients were identified. There was a difference in the proportion of patients for whom surgery was performed, as well as the proportion receiving radiation, when stratified by race (P < .001). Black non-Hispanics had the highest rates of unplanned readmission (7.6%) within 30 days (odds ratio [OR]: 1.39 compared to White non-Hispanics, P < .001). Asian non-Hispanics had the lowest 30- (3.2%) and 90-day mortality (9.8%) when compared to other races (OR: 0.52 compared to White non-Hispanics, P = .031). Compared to White non-Hispanics, we found Black non-Hispanics (hazard ratio [HR]: 0.88, P < .001), Asian non-Hispanics (HR: 0.72, P < .001), and Hispanics (HR: 0.69, P < .001) had longer overall survival. Conclusions Differences in treatment and outcomes exist between races. Further studies are needed to elucidate the etiology of these race-related disparities and to improve outcomes for all patients.

Glioblastoma (GBM), a World Health Organization grade IV diffuse glioma of astrocytic lineage, 1 is the most commonly diagnosed primary malignant brain tumor with approximately 11 833 new diagnoses per year in the United States. 2 Most recent data show that treatment with a standard of care including resection, adjuvant temozolomide-based chemotherapy, and radiation yields 5-year and 10-year survival rates of 5.4% and 2.7%, respectively. 3 In recent years, efforts have been refocused on identifying the epidemiologic factors that contribute to the diagnosis, implemented treatment strategies, and survivorship in cancer patients, including in GBM. 2,[4][5][6][7][8][9][10][11] Past studies have shown associations between GBM risk or survival and insurance type [12][13][14] or socioeconomic status. 15,16 The datasets utilized in these studies all have limitations that analysis with the US nationwide collected dataset would otherwise yield. The National Cancer Database (NCDB) is one such central registry containing compiled data from over 1500 diverse treatment centers and represents more than 70% of newly diagnosed cancer cases nationwide. 17 It provides a more complete representation compared to both the population-based Surveillance, Epidemiology, and End Results (SEER) database, which represents only 28% of the US population, 18 and National Inpatient Sample, which similarly represents only 20% of hospital admissions in the United States. 19 Although prior work on GBM has shown differences in overall survival between races, no prior study has examined different aspects of care such as readmission rates. 20 The purpose of this study is to utilize this complete dataset to better assess the role of race/ethnicity in differences in care (eg, treatment type, readmissions) and clinical outcomes for GBM patients.

Methods
The NCDB was queried for adults at least 18 years of age diagnosed with primary GBM between 2004 and 2014. International Classification of Diseases for Oncology, Third Edition (ICD-O-3)-Oncology codes included morphologic codes 9440, 9441, and 9442 and topographical codes C71.0-C71. 9. Only patients with a histologically confirmed diagnosis of GBM were included. Baseline patient demographics, hospital characteristics, and treatment variables were compared and stratified by race/ethnicity. Patient demographic variables included age at diagnosis, sex, race (White, Black, Asian), ethnicity (Hispanic status [yes/no]), Charlson-Deyo score (identical to the Charlson Comorbidity), insurance status, and rural/urban/metropolitan location of zip code. Patients with reported race or ethnicities of "other" or "unknown" were excluded from this analysis. Hospital characteristics included treatment facility type: Community Cancer Program, Comprehensive Community Cancer Program, Academic/Research Program, and Integrated Network Cancer Program. Treatment and tumor characteristics included extent of surgical resection, unifocal or multifocal disease, type of radiation performed, and chemotherapy performed. Extent of surgical resection was categorized as no surgery, biopsy, subtotal resection, or gross total resection. Clinical outcomes examined included 30-day readmission, 30-and 90-day mortality, and overall survival. Exempt approval was obtained from the University Hospitals Institutional Review Board for this study.

Statistical Analyses
Descriptive statistics are presented and include means and standard deviations for continuous variables and frequency and proportions for categorical variables by race/ethnicity. Chi-squared tests were applied to test differences between racial and ethnic groups. Multivariable Cox regression analysis was performed to evaluate potential variables associated with the 4 defined outcomes by race and Hispanic status (ie, race and ethnicity). Additional covariables included in the multivariable analyses were facility type, age at diagnosis, sex, primary payer (Medicare and Medicaid are government insurances in which taxpayers represent the payer), urban/rural/metro, Charlson-Deyo score, surgical resection, unifocal/multifocal, radiation, and chemotherapy. Kaplan-Meier survival curves stratified by race and ethnicity were generated. Statistical significance was set at P value less than .05.

Results
Sociodemographic and treatment characteristics for the patient sample stratified by race and ethnicity (Hispanic status) are given in Table 1. A total of 103 652 patients diagnosed with GBM were identified (White non-Hispanic

Importance of the Study
We conducted this study to explore racial/ ethnic factors that may be associated with survival in patients with glioblastoma by querying the National Cancer Database (NCDB). We found that racial differences do exist in 30-and 90-day readmissions and mortality rates, as well as overall medial survival. This study is a useful addition to the current literature in large part due to the comprehensive scope of NCDB, as it encompasses approximately 70% of all new cancer diagnoses in the United States. Our study fills a gap in current literature surrounding race/ethnicity in glioblastoma treatment and outcomes and calls for future work to be done to further understand the reasons behind disparities in glioblastoma patient outcomes. n = 90 709, Black non-Hispanic n = 5704, Asian n = 1843, and Hispanic n = 5396). White non-Hispanics were significantly older at the time of diagnosis when compared to other races/ethnicities. Significant differences were present in the rate of surgical resection (White non-Hispanics had the highest rate of gross total resection, 30.7%), radiation therapy (Asian non-Hispanics had the highest rate, 70.9%), and chemotherapy (White non-Hispanics had the highest rate, 65.8%). Multivariable logistic regression models for 30-day readmission by race/ethnicity are given in Table 2. Black non-Hispanics had the highest rates of unplanned readmission within 30 days, as a significant difference was seen comparing Black non-Hispanics to White non-Hispanics on multivariable modeling (odds ratio [OR] 1.39, 95% confidence interval [CI] 1.15-1.6, P < .001). Significant associations were also seen with facility type, Charlson-Deyo score, surgical resection, and focality.
Results of multivariable logistic regression models for 30-day and 90-day mortality by race/ethnicity are given in Tables 3 and 4, respectively. Asian non-Hispanics had the lowest odds of 30-day and 90-day mortality when compared to other races. White non-Hispanics had the highest odds of 30-day and 90-day mortality (although Black non-Hispanics had the highest unadjusted 30-day mortality rate, Table 1). A significant difference in 30-day mortality was seen on multivariable analysis when comparing Asian non-Hispanics and White non-Hispanics (OR 0.52, 95% CI 0.28-0.91, P = .031). Significant differences were seen in 90-day mortality between White non-Hispanics and Asian non-Hispanics (OR 0.64, 95% CI 0.46-0.89, P = .009) and Hispanics (OR 0.648, 95% CI 0.528-0.794, P < .001), though not Black non-Hispanics (OR 0.88, 95% CI 0.73-1.06, P = .2).
Results of multivariable Cox Proportional Hazards modeling of race/ethnicity for overall survival are given in Table  5 Figure 1. Race was significantly associated with adjusted overall survival (P < .001), with White non-Hispanic having the lowest median survival (9.03 months) and Asian non-Hispanic having the highest (13.27 months). Kaplan-Meier survival curves by race and median survival, excluding from analysis individuals who died within 90 days, are demonstrated in Supplementary Figure 1. Race was still significantly associated with adjusted overall survival (P < .001), with White non-Hispanic having the lowest median survival (14.2 months) and Asian non-Hispanic having the highest (18.2 months).  OR, odds ratio; CI, confidence interval.

Discussion
The purpose of this study was to examine the NCDB for an association between race and overall survival, 30-day and 90-day mortality for patients with GBM. Though the NCDB has its own limitations with regard to case coverage and reporting, its scope is quite comprehensive, covering approximately 70% of all new cancer diagnoses in the United States according to a recent comparison of national databases. 17 Our study demonstrates that survival for GBM patients is in part associated with race. Such an association has also been found in various other cancers. Multiple NCDB analyses reporting on other cancer types note race as a significant factor in survival, notably in breast cancer, 21 T-cell lymphoma, 22 uterine cancer, 23 and endometrial cancer. 24 Moreover, previous analyses on GBM have similarly demonstrated that non-Hispanic Whites have the poorest survival, even account for differences in treatment. A SEER database GBM study found that Hispanics have the highest survival, followed by Blacks, and then Asian/Pacific Islanders, relative to Whites. 20 White patients were also shown to have the poorest survival in other GBM datasets, including the Central Brain Tumor Registry of the United States. 2 However, these studies do not address variation in short-term survival (eg, 30-day and 90-day mortality), nor do they address issues related to quality of care as reflected by 30-day readmissions.
We hypothesized if considerations having to do with access to care might vary by race. Our analysis demonstrates that non-Hispanic Whites are more likely than Hispanics and non-Hispanic Blacks to receive standard of care treatment.
Specifically, treatment use is significantly lower in 2-3 race/ ethnicity groups compared to White non-Hispanics, including surgical resection and each adjuvant therapy (radiation therapy or chemotherapy). There are several reports that have observed the survival benefits of undergoing treatment in high-volume academic centers compared to lower-volume community centers. 25,26 Similar findings have been reported for patients undergoing isolated radiation therapy. 27 There has been recently published data on facility volume and outcomes specifically in GBM patients, demonstrating a substantial decrease in prolonged length of stay, readmission rates, and mortality in patients who sought treatment at academic and/or high-volume centers. 28 In fact, multidisciplinary "Tumor Board" conferences have shown to lead to better outcomes, including a higher likelihood of patients receiving adjuvant treatment. 29 With regard to the selection of treatment facility, our analysis shows that a majority of GBM patients seek treatment at either a Comprehensive Community Cancer Program or Academic/Research Program irrespective of race, which represents high-volume facility types. This is in line with past analysis demonstrating GBM patients who travel farther to receive care at high-volume centers have superior postoperative outcomes compared to patients who receive care locally at low-volume centers. 30 This may reflect the more comprehensive diagnostic capabilities of such centers, as well as the presence of oncologic care pathways that provide continuous and multidisciplinary care from the time of diagnosis through resection and adjuvant therapy. This may also reflect limitations in the ability of community programs to access services such as advanced MRI imaging or radiation therapy. Such access has previously been demonstrated as a factor in survival in other cancer types. 31 Our analysis found that differences in 30-day readmissions were present; non-Hispanic Whites are less likely than Hispanics and non-Hispanic Blacks to be re-admitted, suggesting disparities in short-term outcomes. This may reflect the existence of higher rates of perioperative complications not captured by the NCDB. Racial disparities in short-term perioperative outcomes have been widely reported for other oncologic procedures, 32-36 particularly when comparing Black to Caucasian populations. Additionally, readmissions may be correlated to treatment differences by race, specifically, in regard to pain management. Both oncologic and non-oncologic evidence suggest that Blacks, when compared to other races, have both higher pain scores and higher rates of inadequately treated pain. 37,38 While the NCDB does not have sufficient granularity in their data to tease out these specific reasons for readmission, multiple hypotheses that have been proven in the literature include higher rates of perioperative complications, inadequately managed periprocedural pain control, racial bias in discharge planning, differences in communication of discharge expectations, and inadequate provision of socioeconomic support services. [39][40][41] Additionally, a growing body of evidence suggests that racism, ie, systemic bias, rather than race itself may be the driving factor behind observed healthcare disparities. 42,43 Cumulatively these reasons may be contributing factors to the higher rates of 30-day readmission and mortality rates for the Black non-Hispanic population.
Despite the higher rates of treatment and lower readmissions, White non-Hispanics exhibited the lowest overall   survival, demonstrated by our multivariable cox proportional hazards model. Although this may be partially attributable to older ages at diagnosis in White non-Hispanics, our multivariable model did adjust for age. We found that survival was highest in Asians, followed by Hispanics and Blacks relative to Caucasians. Importantly, this finding was durable even after excluding from analysis the patient who died within the first 90 days, that is, Whites continue doing worse even beyond the 90-day window, suggesting biology may underlie these differences. This study does have a number of limitations, as database studies are inherently limited by selection bias, missing data, and confounding factors. The overwhelming majority of GBM patients in the NCDB are White non-Hispanic. Other similar such studies utilizing the NCDB have samples that are up to 90.5% White patients. 12 However, this in part reflects the known higher incidence in White patients. The Central Brain Tumor Registry of the US's 2011-2015 primary brain tumor report shows that the incidence rate of GBM was higher in White patients (3.47) compared to Black patients (1.80) and Asian patients (1.57). 2 While the NCDB captures data for approximately 70% of all new cancer diagnoses, only hospitals with Commission on Cancer accreditation are included, which only applies to about 30% of 5000 hospitals in the United States. This may present a selection bias for the study population, as racial differences may be present in the use of high-volume Commission on Cancer accredited hospitals. 44 Furthermore, only patients with a histologically confirmed diagnosis were included in this dataset, which may limit inclusion of the elderly, those with a late diagnosis, and disadvantaged individuals without access to a histologically confirmed diagnosis. This could also be an indirect contributor to racial differences in survival. Exclusion of patients with undiagnosed cancers or death prior to diagnosis may also influence findings. Additionally, analyzing mortality as an outcome for a disease with as rapid a course as GBM presents the potential confounder of time of presentation influencing the duration of postoperative survival measured. With regard to missing data, most patient data did not have Karnofsky Performance Status, MGMT methylation status, extent of resection, extent of adjuvant therapy, despite all having an important relationship with survival data 45,46 ; therefore, these had to be excluded in the multivariable analyses. As with virtually all multi-institutional registries, miscoding or erroneous entry of data variables must be acknowledged. Additional important predictors such as molecular markers were not present in this dataset, which can have significant impacts on survival as well as may have racial associations which should be investigated in future studies. Finally, race is an elusive concept, sometimes imposed by external sources (the racial NCDB may not reflect all races) or those internally defined. Healthcare outcomes can differ across communities and encompass socioeconomic factors as well as cultural and access issues that are built into the racialism that creates some and fractures many other communities. 47

Conclusions
Ultimately, our findings reiterate the well-known disparities that exist in healthcare as a function of race and should alert care providers and researchers to the continued need for improvement in studying and addressing these issues. Given that GBM is the most common primary brain tumor, it is likely that such findings related to racial/ethnic differences may be true in other primary brain tumors, as our group has demonstrated similar results in gliosarcoma. 48 We call for further work to be done to understand the sources behind the observed disparities in GBM patient treatment, readmissions, and outcomes, so these disparities can ultimately be overcome. Future studies should be focused on addressing such disparities, and such national databases can continue to contribute to this work in the manner in which they collect patient data.