The first two authors contributed equally to this article.
African American and poor patients have a dramatically worse prognosis for head and neck cancer
An examination of 20,915 patients
Article first published online: 6 OCT 2008
Copyright © 2008 American Cancer Society
Volume 113, Issue 10, pages 2797–2806, 15 November 2008
How to Cite
Molina, M. A., Cheung, M. C., Perez, E. A., Byrne, M. M., Franceschi, D., Moffat, F. L., Livingstone, A. S., Goodwin, W. J., Gutierrez, J. C. and Koniaris, L. G. (2008), African American and poor patients have a dramatically worse prognosis for head and neck cancer. Cancer, 113: 2797–2806. doi: 10.1002/cncr.23889
- Issue published online: 3 NOV 2008
- Article first published online: 6 OCT 2008
- Manuscript Accepted: 24 JUN 2008
- Manuscript Revised: 20 JUN 2008
- Manuscript Received: 21 APR 2008
- James and Ester King Tobacco Research Grant from the State of Florida
- head and neck;
- socioeconomic status;
Differences in cancer survival based on race, ethnicity, and socioeconomic status (SES) are a major issue. To identify points of intervention and improve survival, the authors sought to determine the impact of race, ethnicity, and socioeconomic status for patients with cancers of the head and neck (HN).
HN cancer patients diagnosed between 1998 and 2002 were examined using a linked Florida Cancer Data System and Florida Agency for Health Care Administration data set.
A total of 20,915 patients with HN cancers were identified, predominantly in the oral cavity and larynx. Overall, 72% of patients were male, 89.7% were white, 8.4% were African American (AA), and 10.6% were Hispanic. The median survival time (MST) was 37 months. MST varied significantly by race (white, 40 months vs AA, 21 months; P < .001), sex (men, 36 months vs women, 41 months; P = .001), and area poverty level (lowest, 27 months vs highest, 34 months; P < .0001). Only 32% of AA patients underwent surgery in comparison with 45% of white patients (P < .001). On multivariate analysis, independent predictors of poorer outcomes were race, poverty, age, sex, tumor site, stage, grade, treatment modality, and a history of smoking and alcohol consumption.
Carcinomas of the HN have an overall high mortality with a disproportionate impact on AA patients and the poor. Dramatic disparities by race and SES are not explained completely by demographics, comorbid conditions, or undertreatment. Earlier diagnosis and greater access to surgery and adjuvant therapies in these patients would likely yield significant improvement in outcomes. Cancer 2008. © 2008 American Cancer Society.
Head and neck (HN) cancers—including cancers of the larynx, oral cavity, and pharynx—represent 3% of all new cancer cases diagnosed in the US. Approximately 51,700 new cases of HN cancers will be diagnosed in 2008, with 21,800 deaths directly related to these types of tumors.1 We sought to determine the relative effects of race, ethnicity, and socioeconomic status (SES) on survival for patients diagnosed with HN cancers. In analyzing possible factors associated with differences in outcome, we hoped to identify possible points of intervention that may lead to improved patient survival. We analyzed a unique linked dataset of all HN cancer patients diagnosed in the state of Florida from 1998 to 2002. This dataset was generated by linkage between the Florida Cancer Data System (FCDS) and the Florida Agency for Health Care Administration (AHCA). To our knowledge, the dataset is among the largest available of patients with HN cancers that include comorbidity information.
MATERIALS AND METHODS
The 2007 FCDS dataset was used to identify all incident cases of HN cancer diagnosed in the state of Florida. The FCDS dataset was enhanced with data linked from the AHCA dataset. AHCA maintains 2 databases (Hospital Patient Discharge Data and Ambulatory Outpatient Data) on all patient encounters within hospitals and freestanding ambulatory surgical and radiotherapy centers in Florida. All hospitals have been required to report all discharges and outpatient encounters to AHCA since 1987. The AHCA datasets used in this study contain diagnoses, comorbid conditions, and procedures performed during every hospitalization or outpatient encounter for the period 1998 to 2002. Cases in the FCDS and AHCA datasets were linked on the basis of unique identifiers and confirmed with date of birth and sex. Postal codes listed in the FCDS-AHCA database were then used to determine community poverty levels according to the 2007 US Census Bureau report.2 Non-Florida residents were not included in the analysis because follow-up for such patients, particularly survival information, may be inaccurate in up to 10% of such patients (unpublished data).
Statistical analysis was performed with SPSS Statistical Package (version 15.0;(SPSS Inc., Chicago, Ill) and Stata 8.0 Statistical Software (StataCorp, College Station, Tex). Correlations between categoric variables were made using the chi-square test. Median and 5-year survival was calculated by the Kaplan-Meier method. Because the FCDS collects only primary cause of death, we analyzed only overall survival. Survival was calculated from the time of the initial diagnosis to the date of last contact (or date of death). The univariate effects of demographic, clinical, and treatment variables on survival were tested by the log-rank test for categoric values. A multivariate, Cox proportional hazards model was created, taking into account the variables that were found to be significant on univariate analysis. This final model was corrected for the effects of clustering of patients at particular treatment centers. In addition to demographic and clinical variables, patient comorbidities were added to the final model. Patient comorbidities were obtained from the linkage to the AHCA patient charts. Comorbid conditions included in this analysis are as follows: congestive heart failure, cardiac arrhythmias, valvular disease, pulmonary circulation disorders, peripheral vascular disorders, hypertension, paralysis, other neurologic disorders, chronic pulmonary disease, uncomplicated diabetes, complicated diabetes, hypothyroidism, renal failure, liver disease, peptic ulcer disease (excluding bleeding), acquired immunodeficiency syndrome (AIDS)/human immunodeficiency virus (HIV), lymphoma, metastatic cancer, solid tumor without metastasis, rheumatoid arthritis/collagen vascular diseases, coagulopathy, obesity, weight loss, fluid and electrolyte disorders, blood loss anemia, deficiency anemia, alcohol abuse, drug abuse, psychoses, and depression.
The staging criteria used by the FCDS are consistent with the Surveillance, Epidemiology, and End Results (National Cancer Institute) summary staging and differ from the TNM staging guidelines. In this study, local staging represents disease that does not extend beyond the primary organ, whereas those having positive lymph nodes at the time of resection were classified as having regional disease. Documentation of distant metastases during the perioperative period led to classification of affected patients as having distant disease.
Patient Demographics and Patient Characteristics
Over the 5-year period studied, 20,915 patients with tumors in the HN were identified. Demographics, social, and tumor characteristics of the entire study population, including subset analysis, are summarized in Table 1. The majority of the patients were men (n = 15,651; 72.3%), white (n = 19,478; 91.5%), and non-Hispanic (n = 19,419; 90.4%). The majority of subjects were ages 60 to 80 years (n = 16,049; 74%). Greater than 33% of tumors were situated within the oral cavity (n = 7819; 36%). Localized and regional disease were noted in equal numbers (n = 7900 [43.8%] and n = 7621 [42.3%], respectively). Approximately 50% of the study population lived in a community where 10% of the area population was living below the poverty line.
|Median age at diagnosis, y||Entire Cohort 65||Race||Ethnicity||Community Poverty Level*|
|White 65||African American 60||P < .001||Hispanic 63||Non-Hispanic 64||P .024||a 66||b 65||c 64||d 63||P < .001|
|n||%||% of Total||% of Total||% of Total|
|Age groups, y|
|Community poverty level*|
|Parotid and other||1592||7.3||7.4||6.2||8.1||7.2||8.3||8.2||6.9||6|
|Pharynx and other||5314||24.5||24||30.4||22.3||24.7||23.7||24.1||24.7||25|
Median survival rates of the entire study population, including subset analysis, are summarized in Table 2. The median survival time (MST) for the entire cohort was 37 months. Significantly higher survival rates were observed in women (41 months vs 36 months for men; P = .01), whites (40 months vs 21 months for African Americans [AAs]; P < .001), and Hispanics (47 months vs 37 months for non-Hispanics; P = .02). As age increased in the study population, a decrease in MST was observed. Living in an area in which ≥15% of individuals in the community reside below the poverty line conferred a poorer outcome. Among tumor sites, lesions located in the pharynx had the worst prognosis (MST of 26 months), whereas laryngeal tumors had the best outcomes (MST of 47 months). Survival was significantly higher with local disease stage than with distant disease (MST of 60 months vs 12 months; P < .001). Patients with well-differentiated tumors fared better than those with poorly differentiated tumors (56 months vs 33 months; P < .001).
|Entire Cohort||Race||Ethnicity||Community Poverty Level*|
|mo||P||White, mo||African American, mo||P||Hispanic, mo||Non-Hispanic, mo||P||a, mo||b, mo||c, mo||d, mo||P|
|Age groups, y|
|Community poverty level*|
|Parotid and other||39||38||58||0.087||59||37||.379||44||37||52||29||.081|
AAs Have a Poor Prognosis
Univariate subset analysis suggested that AA race conferred a significantly poorer prognosis for HN cancer. Whereas sex distribution among races did not differ statistically, the MSTs for AA men (19 months vs 39 months; P < .001) and women (26 months vs 43 months; P < .001) were significantly shorter than those of their white counterparts.
AA patients were also diagnosed at a younger age, and presented with more advanced disease compared with whites. Greater than 80% of AAs were diagnosed before the age of 70 years, whereas only 63% of whites are diagnosed before that age (P < .001). The MST of AAs was significantly shorter than for whites among all age strata. Regardless of tumor location, AA patients had a shorter MST than the white cohort, with the exception of parotid or other salivary gland tumors, in which there was no significant difference noted. Approximately 50% of whites presented with regional or distant disease compared with 65% among AA patients (P < .001). Furthermore, AA patients had a higher proportion of poorly differentiated (33.5% vs 28.9%; P < .001) or undifferentiated tumors (3.4% vs 2.6%; P < .001) compared with whites. For all tumor stages and grades, the MST for AA patients was significantly shorter than for whites. Greater than half of the AA patients diagnosed with HN cancer lived in a community in which at least 15% of the local population was below the poverty line, compared with only 23% of whites (P < .001). At all poverty level strata, AA patients had a shorter MST than white patients (Table 2).
Differences in treatment modality and outcomes between AA and white patients were also observed. Although more white patients underwent surgical resection (45% vs 32%; P < 0.001), AA patients received more chemotherapy (26 vs 19%; P < .001) and radiotherapy (66% vs 56%; P < .001). Kaplan-Meier survival analysis of white and AA patients is shown in Figure 1a.
Ethnic Differences in HN Cancers
Differences in outcomes by ethnic background were also observed on univariate subset analysis. Hispanic patients had a longer MST than non-Hispanics regardless of sex (men: 43 months vs 36 months [P = .001]; women: 55 months vs 40 months [P < .001]). In all tumor locations, non-Hispanic patients had significantly lower MST than Hispanic patients. Furthermore, across all tumor stages and grades, non-Hispanics fared significantly worse than Hispanics.
So too, white non-Hispanic patients had a worse prognosis than white Hispanic patients (39 months vs 45 months; P = .002). Although not statistically significant, a similar trend was observed when non-Hispanic black patients were compared with Hispanic black patients (MST of 21 months vs 26 months; P = .09). More Hispanic patients lived in communities with a>10% incidence of poverty (63% vs 43%; P < .001). Despite this, Hispanics had a significantly longer MST than non-Hispanics. Whereas a larger proportion of Hispanics received chemotherapy (23% vs 19%; P < .001), no statistical difference was observed in frequency of radiotherapy (P = .104) or surgical intervention (P = .091). However, the MST in Hispanic patients was significantly better than in non-Hispanics for those patients undergoing surgery (63 months vs 51 months; P < .001), chemotherapy (33 months vs 26 months; P < .001), and radiotherapy (51 months vs 37 months; P < .001). Kaplan-Meier analysis of Hispanic and non-Hispanic survival is shown in Figure 1b.
Effects of Area Poverty on Prognosis of HN Cancers
Community poverty level also appeared to affect MST in HN cancer. Although men proportionately outnumbered women in each area poverty category, their survival was consistently shorter (P < .001). Patients living in communities with community poverty levels exceeding 15% presented with HN cancer at a significantly earlier age, and MST was decreased across all age groups in these communities. Furthermore, patients from these communities more frequently presented with advanced-stage disease. Despite receiving chemotherapy and radiotherapy at a higher frequency than any other demographic group, MST was significantly inferior in those patients from the areas with the highest poverty rates irrespective of therapeutic modality, surgery included. Kaplan-Meier survival analysis by area poverty level is shown in Figure 1c.
A larger proportion of AAs were reported to consume alcohol in this study (29% vs 17%; P < .001). For both those who consumed alcohol and those who did not, survival was worse for AA than for whites (14 months vs 24 months [P = .001] and 24 months vs 45 months [P = .001], respectively). A larger proportion of non-Hispanics than Hispanics was reported to consume alcohol (18% vs 15%; P < .001); however, there were no differences in survival noted among those who drank (P = .357). When nondrinking cohorts were compared, Hispanics had better survival than non-Hispanics (51 months vs 43 months; P = .011). As area poverty level increased, the frequency of alcohol consumption increased. For both those who consumed and those who did not consume alcohol, survival decreased as area poverty level increased.
The frequency of smoking in the white and AA cohorts did not differ significantly. The survival of white smokers was significantly higher than that of their AA counterparts (39 months vs 20 months; P = .001). Survival advantage was also observed in whites who did not smoke when compared with AA nonsmokers (56 months vs 33 months; P = .001). A larger proportion of non-Hispanics over Hispanics were observed to smoke (83% vs 74%; P < .001). Survival in the Hispanic cohort who smoked was higher than survival among non-Hispanic smokers (44 months vs 36 months; P < .001). When the cohorts who did not smoke were compared, no difference in MST was observed. As with alcohol consumption, poverty level correlated directly with tobacco usage.
Effects of Comorbidities on Survival in HN Cancer
The frequency of comorbidities and their effect on median survival in the study population was examined. Greater than half of the entire study population (60.5%) had a diagnosis of hypertension along with HN cancer. Conversely AIDS/HIV was the most infrequent comorbidity observed (1.0%). On univariate analysis, the presence of drug abuse in the patient's medical history did not significantly affect survival, whereas the 27 other comorbid conditions were observed to significantly affect survival (data not shown). On multivariate analysis, several comorbidities were found to be independent predictors of poorer outcomes (Table 3).
|Frequency, %||Hazards Ratio||P|
|Congestive heart failure||19.5||1.102||<.001|
|Pulmonary circulation disorders||2.0||1.097||.160|
|Peripheral vascular disease||12.1||1.016||.634|
|Chronic pulmonary disease||43.0||1.151||<.001|
|Peptic ulcer disease||4.4||1.008||.863|
|Rheumatoid arthritis/collagen vascular disease||21.4||0.754||<.001|
|Fluid and electrolyte disorder||40.4||1.698||<.001|
|Blood loss anemia||4.2||1.097||.042|
Results of multivariate analysis using the Cox regression model are summarized in Table 4. Results from comorbidities are shown in Table 3. Demographic variables that are independent predictors of survival in HN cancer include age, race, poverty level, and alcohol and tobacco usage. Differences between Hispanics and non-Hispanics failed to reach statistical significance in the final stepwise, multivariate model. Among clinical variables, tumor location, tumor stage, tumor grade, surgical extirpation, and adjuvant treatment modalities such as chemotherapy and radiotherapy, also proved to be independent predictors of survival.
|Hazards Ratio||95% CI||P|
|Age groups, y|
|Community poverty level†|
|a (wealthiest)||Reference group|
|Oral cavity||Reference group|
|Parotid and other||1.107||0.951-1.289||.19|
|Well differentiated||Reference group|
To better define the relation between race, ethnicity, and SES with other independent variables, a stepwise, multivariate analysis using the Cox regression model was also performed. Results are summarized in Table 5. Correcting for comorbid conditions significantly affected hazard ratios (HRs) both for AA and poor patients. The HR for AA race after correction for comorbidities decreased substantially from 1.607 to 1.387. Likewise, the HR for those living in poverty decreased from 1.202 to 1.139. Corrections for clustering were also performed.
|Hazards Ratio||95% CI||P|
|Race + demographics†||1.607||1.473-1.753||<.001|
|Race + demographics + comorbidities‡||1.387||1.268-1.516||<.001|
|Race +demographics + comorbidities + clinical characteristics§||1.390||1.254-1.552||<.001|
|Race + demographics + comorbidities + clinical characteristics + treatment‖||1.365||1.229-1.516||<.001|
|Ethnicity + demographics||0.935||0.805-1.086||.377|
|Ethnicity + demographics + comorbidities||0.899||0.797-1.013||.080|
|Ethnicity + demographics + comorbidities + clinical characteristics||0.570||0.776-0.999||.048|
|Ethnicity + demographics + comorbidities + clinical characteristics + treatment||0.890||0.788-1.006||.062|
|Poverty + demographics||1.202||1.119-1.291||<.001|
|Poverty + demographics + comorbidities||1.139||1.062-1.222||<.001|
|Poverty + demographics + comorbidities + clinical characteristics||1.114||1.010-1.229||.030|
|Poverty + demographics + comorbidities + clinical characteristics + treatment||1.116||1.008-1.235||.034|
Differences in cancer survival based on race, ethnicity, and SES remain a major issue despite recognition of these inequalities for more than 30 years. In an effort to understand outcomes for HN cancer patients and potentially to improve survival, we examined a population-based registry to identify global prognostic factors important in the survival of patients diagnosed with HN cancer. In this analysis, emphasis was placed on the effects and interplay of race, ethnicity, and area poverty level on overall survival. Although our report is consistent with others that have also found racial and socioeconomic disparities in cancer, this study is unique in that patients from all age groups, as well as their comorbid conditions, were included in the analysis. To our knowledge, this study represents the largest, most comprehensive analysis of these variables on the outcomes of patients with HN cancer reported to date.
We observed that race significantly affects outcomes for patients with HN cancer independent of demographics, SES, comorbidities, clinical characteristics, and treatment modality. These results support the findings of racial disparities that have been previously demonstrated for several other malignancies.3–7 In the cohort reported herein, AA patients are diagnosed at a younger age and present with more advanced disease. Although they are more likely than whites to undergo chemotherapy and radiotherapy, they are less likely to be operated on for their tumors. It may be that larger numbers of AA patients are poor operative candidates because of comorbidities, or have a higher incidence of disease that is inoperable at diagnosis because of extent of disease burden, leading to more frequent use of palliative chemotherapy and radiotherapy. The survival of AA patients who do undergo surgery, however, is also significantly shorter than that of surgically treated whites (Table 2).
In addition to differences in survival by race, we also observed that living in an area of high poverty was an independent predictor of worse outcomes regardless of demographics, comorbidities, clinical characteristics, and treatment modality. This observation is consistent with studies of other malignancies that have identified SES as 1 of the main factors responsible for differences in outcome. We observed that those patients from poor areas are diagnosed at a younger median age, and regardless of age group, these patients have worse outcomes as evidenced by shorter MST. As area poverty level increased in whites, their survival decreased. No differences in survival were observed in AA across the different area poverty levels.
In the univariate model, the results suggested that being AA, as compared with white, is a predictor of poor outcome. The stepwise multivariate analysis showed that AAs indeed had a relatively poor prognosis even after correcting for all other variables. However, a substantial decrease in HR is observed after the addition of comorbid conditions into the stepwise multivariate regression model. Furthermore, the stepwise model indicates that regardless of how they are treated for HN cancer (surgery, chemotherapy, and/or radiation), being AA results in a poor overall outcome. Some factor or factors other than the tested variables, such as genetic or environmental differences, is thus likely responsible for the observed outcome differences. Increased use of tobacco and alcohol, however, does explain part of the reduced survival of AAs as compared with whites.
By using the same stepwise, multivariate model, area poverty level was an independent predictor of poor outcomes in HN cancer after controlling for a host of other factors. Similarly, the addition of comorbid conditions into the stepwise multivariate regression model decreased the HR in the poor patients. The American Cancer Society (ACS) suggested that low SES in AAs was responsible for many inequalities.8 Moreover, the ACS, in collaboration with the National Cancer Institute and Centers for Disease Control, issued a landmark report highlighting the key issues.9 The poor were found to have worse outcomes because of financial hardship and lack of insurance, and therefore poor access to healthcare. This report also suggested that the poor endured more pain and suffering when diagnosed with cancer, and that they did not benefit from cancer education and outreach programs. Finally, the report averred that poor patients regard cancer diagnoses with a sense of resignation and futility, and therefore are less likely to seek medical treatment.9 In our final Cox regression model, even when the poor were treated for HN cancer (surgery, chemotherapy, and/or radiation), poorer overall outcomes were also observed.
Using univariate analysis, Hispanics have a longer MST than non-Hispanics (47 months vs 37 months; P = .02) (Table 2). On multivariate analysis, however, differences between Hispanics and non-Hispanics failed to reach statistical significance (HR of 0.89; P = .062) (Table 4). Upon further analysis using a stepwise model, the addition of treatment variables into the model corrected for any differences observed between Hispanics and non-Hispanics. This suggests that non-Hispanics with HN cancer have a poorer prognosis when untreated. However, once non-Hispanics have access to healthcare and are evaluated and treated with surgery, chemotherapy, and/or radiotherapy, there is no significant difference in the prognosis between these 2 groups. This is in contrast to AAs. Indeed, even after correcting for treatment in these 2 groups, race and SES continue to be independent predictors of survival.
Tobacco use and alcohol consumption have long been linked with cancer, especially of the head and neck.10–12 We have observed that a larger proportion of AAs and non-Hispanics consume alcohol. This was associated with a shortened survival in the former group. Furthermore, as area poverty level increases, the proportion of alcohol consumption also increases (Table 1). However, even among those who do not drink, area poverty level was associated with increased mortality, which is consistent with poverty being an independent predictor of poor outcomes. No racial differences were observed in the frequency of tobacco usage. However, survival in the AA cohort was consistently lower regardless of whether or not they smoked. Surprisingly, although overall survival in Hispanics was greater than for non-Hispanics, there were no ethnic differences in survival noted for those who did not smoke. This suggests that smoking has a more deleterious effect on survival in non-Hispanics than in Hispanics.
This report presents a large, comprehensive analysis of HN cancer patients using a linked FCDS and AHCA database. This linkage dramatically improved the power of the study because it provided additional data such as comorbidities, enhanced follow-up, socioeconomic information, and improved treatment information. Our study was restricted to 1998 through 2002 because linkage of these datasets was not possible in datasets for prior years. The FCDS, which currently includes >2.7 million records, is a population-based registry of all cancer cases diagnosed or treated in the state of Florida, which represents approximately 6% of the total US population. The data collected from large cancer registries provides insight into tumor behavior and allow us to examine outcomes from current treatment strategies.13–18 Although this represents an excellent database for comparative outcomes analysis, it is not without limitations. Using area poverty as a proxy for SES may result in misclassification of some patients whose postal code does not accurately reflect the true income level of the individual. In addition, FCDS records only primary cause of death; as a result, we were unable to include disease-specific survival in our examination. The status of surgical resection margins is not included in the database and could not be included in the analysis. Furthermore, although data regarding radiotherapy and chemotherapy were examined, information on specific therapeutic regimens and dosages were also not available.
In conclusion, racial and SES disparities continue to exist in HN cancer survival. This inequality is not explained completely by demographics, comorbid conditions, or undertreatment because poor outcomes continued to be observed after correcting for these factors. Earlier diagnosis, particularly in those from low SES groups and among AA patients, is needed to improve outcomes. Furthermore, examination into potential differences in tumor biology among cohorts needs to be undertaken to identify better therapies for these patients.
- 2Current Population Reports. Washington, DC: US Census Bureau; 2007: P60–P226., .
- 9American Cancer Society. Cancer in the Poor: a Report to the Nation. Atlanta, GA: American Cancer Society; 1989.