Farjah F, Wood DE, Yanez ND, Vaughan TL, Symons RG, Krishnadasan B, Flum DR. Racial Disparities Among Patients With Lung Cancer Who Were Recommended Operative Therapy. Arch Surg. 2009;144(1):14-18. doi:10.1001/archsurg.2008.519
Health care system and provider biases and differences in patient characteristics are thought to be prevailing factors underlying racial disparities. The influence of these factors on the receipt of care would likely be mitigated among patients who are recommended optimal therapy. We hypothesized that there would be no significant evidence of racial disparities among patients with early-stage lung cancer who are recommended surgical therapy.
Design, Setting, and Patients
Retrospective cohort study of patients in the Surveillance, Epidemiology, and End Results–Medicare database who were diagnosed with stage I or II lung cancer between January 1, 1992, and December 31, 2002 (follow-up through December 31, 2005).
Main Outcome Measures
Receipt of lung resection and overall survival.
Among 17 739 patients who were recommended surgical therapy (mean [SD] age, 75  years; 89% white, 6% black), black patients less frequently underwent resection compared with white patients (69% vs 83%, respectively; P < .001). After adjustment, black race was associated with lower odds of receiving surgical therapy (odds ratio = 0.43; 99% confidence interval, 0.36-0.52). Unadjusted 5-year survival rates were lower for black patients compared with white patients (36% vs 42%, respectively; P < .001). After adjustment, there was no significant association between race and death (hazard ratio = 1.03; 99% confidence interval, 0.92-1.14) despite a 14% difference in receipt of optimal therapy.
Even among patients who were recommended surgical therapy, black patients underwent lung resection less often than white patients. Unexpectedly, racial differences in the receipt of optimal therapy did not appear to affect outcomes. These findings suggest that distrust, beliefs and perceptions about lung cancer and its treatment, and limited access to care (despite insurance) might have a more dominant role in perpetuating racial disparities than previously recognized.
Lung cancer is the second most common malignant neoplasm in the United States and is responsible for more deaths than any other cancer.1 Pulmonary resection provides the best chance of cure for patients with early-stage disease.2- 4 Black patients with early-stage lung cancer have lower 5-year survival rates than white patients, and this difference in outcome has been attributed to lower rates of resection among black patients.5
Barriers to the receipt of optimal cancer therapy are likely numerous and complex.6 Several potential factors underlying racial differences in the receipt of surgical therapy include differences in pulmonary function,7 access to care,8 refusal of surgery,9 beliefs about tumor spread on air exposure at the time of operation10 and the possibility of cure without surgery,11 distrust of the health care system and physicians,12 suboptimal patterns of patient and physician communication,13 and health care system and provider biases.14,15 Differences in patient characteristics (ie, lung function, performance status, and severity of comorbidity) and health care system and provider biases are often argued to be significant factors underlying racial disparities.
To better understand the relationship between hypothesized factors and racial differences in care and outcomes, we conducted a cohort study using the Surveillance, Epidemiology, and End Results (SEER)–Medicare database. Among patients who were recommended surgical therapy, the effects of health care system and provider biases on the receipt of optimal care might be mitigated, as might the potentially confounding effects of some measured and unmeasured patient characteristics. We hypothesized that resection and survival rates would not vary between white and black patients among a cohort of patients with early-stage lung cancer who were recommended surgical therapy.
A retrospective cohort study of patients diagnosed with lung cancer between January 1, 1992, and December 31, 2002, was conducted. The use of the SEER-Medicare database to investigate cancer care and the quality and validity of variables therein have been described extensively elsewhere.16- 22 The University of Washington Institutional Review Board approved this study and waived consent because the investigation involved existing, deidentified data.
Among 221 208 patients diagnosed with lung cancer between January 1, 1992, and December 31, 2002, sequential exclusions were made for patients with the following: advanced stage of disease or missing stage data (n = 177 910); diagnosis at autopsy or death (n = 134); ages younger than 66 years (n = 9284); diagnosis of a second malignant neoplasm up to 6 months after the lung cancer diagnosis (n = 1333); partial fee-for-service Medicare coverage and/or concurrent enrollment in a health maintenance organization (n = 10 389); or a histology code inconsistent with lung cancer (n = 32). An additional 4387 patients who were not recommended surgical therapy (19% of white patients and 26% of black patients) were excluded from the analysis, for a final cohort of 17 739 patients.
The SEER data abstractors examined medical records to determine whether surgical therapy was recommended.23 In this study, patients were considered to have been recommended surgical therapy if there was a documented recommendation in the medical records and/or if the patient underwent surgical therapy.
Race was the main predictor variable, ascertained using the Medicare Enrollment Database and classified as white, black, or other based on a prior recommendation.17 Income and education level were based on quartiles of median income and proportion of people with only a high school education living in the patient's zip code at the time of diagnosis, respectively. Geographic location was determined by the location of the SEER registry. Residence refers to the size of the population where the patient lived (metropolitan, ≥250 000 persons; urban, 20 000-249 999 persons; and rural, <19 999 persons). The previously validated Klabunde-modified Charlson Comorbidity Index was derived from carrier and outpatient claims in the year prior to diagnosis.24 Stage was based on the sixth edition of the American Joint Committee on Cancer's Cancer Staging Manual.25
Outcomes included resection and overall survival. Resection, radiation therapy, and chemotherapy were ascertained up to 6 months after diagnosis using The Healthcare Common Procedure Coding System, International Classification of Diseases, Ninth Revision26 procedure codes, and Revenue Center Codes as appropriate in the carrier claim and outpatient files (codes available on request). Death information was available in the Medicare Enrollment Database with follow-up through December 31, 2005.
Differences in continuous and categorical variables between black and white patients were tested using t tests for independent samples and χ2 tests, respectively. Logistic regression was used to evaluate whether an association existed between race and the receipt of resection after adjustment for potentially confounding factors. Unadjusted differences in overall survival were estimated using the Kaplan-Meier method. Cox proportional hazards models were used to evaluate the unadjusted and adjusted relationships between race and survival. Schoenfeld residuals were used to test the proportional hazards assumption, and extended (stratified) Cox models were fitted if the proportional hazards assumption was violated. Survival time was defined by the interval between the date of diagnosis and the date of death or censoring. Robust variance estimators were used for all of the regression analyses. All of the regression analyses were case-complete analyses. Seven percent of patients had at least 1 missing covariate value. The proportion of patients with missing covariate data did not vary between black and white patients (7% for both; P = .82), and patients with missing covariate data did not have a lower 5-year survival rate compared with those with no missing data (42% for both; P = .46).
Several post hoc sensitivity analyses were conducted. The relationship between race and receipt of surgical therapy would ideally be characterized in terms of relative risks or risk differences, but generalized linear models estimating relative risks and risk differences could not be fit. Results from the logistic regression analysis were reexamined after converting odds ratios to relative risks. When event rates are high, as was the case in this investigation (receipt of surgical therapy), an odds ratio tends to overestimate the risk ratio.27 The relationship between race and receipt of surgical therapy was also reevaluated in a time-to-event analysis censoring for deaths occurring among untreated patients within 6 months of diagnosis. If rates of death between diagnosis and treatment varied by race, then the results of the primary analysis might have been biased. Finally, all of the regression analyses were repeated after adjustment for clustering within the SEER registry because practice patterns might have been highly correlated within a given SEER registry. Adjustment for clustering at the provider or institution level was not possible because not all of the patients underwent surgical therapy and some patients received no therapy at all.
Stata special edition 9.2 statistical software (Stata Corp, College Station, Texas) was used for all of the statistical analyses. A 2-sided P value less than .01 was considered statistically significant.
A total of 17 739 patients with stage I or II lung cancer (median [range] age, 74 [66-103] years; 89% white and 6% black) were recommended surgical therapy (Table 1). Compared with white patients, black patients more frequently were men, were in the lowest quartiles of income and education, were not married at the time of diagnosis, were residing in a metropolitan area, and had more comorbid conditions. Racial differences in the distributions of stage and histological findings were not clinically significant.
Black patients underwent lung resection less frequently than white patients (69% vs 83%, respectively; difference, 14%; 99% confidence interval [CI], 11%-18%) despite both groups having been recommended surgical therapy (Table 2). Among those who did not undergo resection, there was no significant difference in the proportion of black vs white patients who underwent medical management (36% vs 40%, respectively; P = .26). After accounting for the potential confounding effects of variables listed in Table 1, black patients had 55% lower odds of having received a lung resection compared with white patients (Table 3). This association persisted in all of the sensitivity analyses, although the magnitude of the association was attenuated in some cases: odds ratios converted to relative risk (adjusted relative risk = 0.83; 99% CI, 0.78-0.88), censoring for early death (adjusted hazard ratio = 0.67; 99% CI, 0.60-0.75), and adjustment for clustering within the SEER registries (adjusted odds ratio = 0.45; 99% CI, 0.34-0.60).
Black patients had a lower unadjusted 5-year overall survival rate compared with white patients (36% vs 42%, respectively; P < .001). After adjusting for patient and disease characteristics, only a small and nonsignificant association between black race and death was evident (Table 3) despite the fact that black patients underwent resection 14% less often than white patients. This lack of association between race and death persisted in a sensitivity analysis adjusting for clustering within the SEER registries (adjusted hazard ratio = 1.03; 99% CI, 0.93-1.13). In an exploratory analysis, treatment received (including no therapy) was added to the model to investigate whether race had an effect on death through a pathway unrelated to the receipt of therapy. Black patients had a 6% lower hazard of death compared with white patients, although this association was not significant (adjusted hazard ratio = 0.94; 99% CI, 0.82-1.09). In another exploratory analysis, lung cancer cause-specific survival rates were evaluated but there was no racial variation in outcomes in the unadjusted (62% for black patients vs 64% for white patients; P = .17) or adjusted (hazard ratio = 1.01; 99% CI, 0.88-1.17) analyses.
Multiple factors likely result in racial differences in care and outcomes among patients with early-stage lung cancer, but the effect of any one factor on overall differences is not well understood. Health care system and provider biases and differences in patient characteristics are often thought to be predominant factors underlying racial disparities. Accordingly, we hypothesized that the influence of these factors on the receipt of optimal therapy would be mitigated among patients who were recommended surgical therapy; therefore, we did not expect significant racial variation in lung resection or outcomes. To our knowledge, this investigation is the first to use a nationally representative database to describe care and outcomes among patients with early-stage lung cancer who were recommended surgical therapy. Contrary to expectation, we observed a large (14%) difference between black and white patients in the proportion of patients who actually underwent lung resection, even though all of the patients were recommended surgical therapy. Another surprising finding was that this large difference in receipt of optimal therapy did not appear to affect long-term survival after adjustment for patient and disease characteristics. These findings suggest that reasons other than health care system and provider biases and patient characteristics may play a larger role in creating racial disparities than previously recognized.
There are several possible explanations for why large racial differences in care were observed among patients with lung cancer who were recommended surgical therapy. Black patients have been shown to refuse surgical therapy more frequently than white patients for reasons relating to beliefs about accelerated tumor spread at the time of operation,10 the possibility of cure without resection,11 and distrust of the health care system and providers.12 Although refusal rates have been reported to be as high as 18% among black patients,9 the generalizability of that finding is uncertain because the proportion of black patients in that study was higher than that reported in nationally representative cohorts (35% vs 10%, respectively).28 Our study could not determine why patients did not receive surgical therapy, but the findings are consistent with the notion that refusal rates among patients with lung cancer may be high among all patients and higher among black patients. Another explanation for why black patients underwent lung resection less frequently than white patients is limited access to care. All of the patients in this study had uniform health insurance and demonstrated the ability to use health services—as evidenced by a documented diagnosis and treatment recommendation—but not all of the patients may have had access to care facilities and/or providers able to provide optimal cancer therapy. Supporting that notion, a study of Medicare beneficiaries found that primary care physicians treating black patients had less access to high-quality subspecialists compared with physicians treating white patients.29 A third explanation for racial differences in care is that health care system and provider biases may exist even in a setting where optimal therapy is recommended. Studies evaluating patients with a diagnosis of lung cancer or a pulmonary nodule found that patients in racially discordant physician-patient pairs received less information and were less often prompted to actively participate in conversation compared with patients in racially concordant pairs.13 Additionally, black patients trusted their physician less often than white patients after an initial visit with a surgeon or oncologist, and this distrust appeared to be linked to perceptions that physician communication was not supportive, partnering, or informative.12 Finally, hypothesized factors not measured in our study or previously investigated—such as racial variation in distance, travel, family support, or social support6—might also explain racial differences in the receipt of lung resection.
Despite large racial differences in the receipt of optimal therapy, there were no observable differences in survival between black and white patients. This finding is seemingly implausible because it is contrary to conventional wisdom, practice guidelines, and empirical evidence. Among all of the patients in the SEER-Medicare database with stage I or II lung cancer, stage-based survival rates were significantly higher for patients who received operative therapy compared with those who did not (stage I: 50% vs 15%, respectively; P < .001; stage II: 28% vs 8%, respectively; P < .001). Inadequate risk adjustment provides one explanation for this unexpected finding. The relationship between race and survival may have been confounded by unmeasured patient selection factors. For instance, forced expiratory volume is strongly associated with both the receipt of surgical therapy and survival, and it may vary by race.7 No racial difference in survival might be observed if white patients undergoing operations were at higher risk (defined by poorer pulmonary function) for death compared with black patients undergoing operations, counterbalancing the effect of more lung resections among white patients compared with black patients. Because the SEER-Medicare database does not record detailed information about patient selection, this study may not have provided adequate adjustment for all potential confounders—including forced expiratory volume, performance status, and severity of comorbid conditions. To the extent that this explanation is true, the issue of overuse and underuse of therapy is important to consider.30 Surgical therapy may have been overused among white patients or underused among black patients, although this determination cannot be made because practice guidelines are not based on risk stratification.2
The major limitation of this study is concern over the accuracy and validity of the recommended surgery variable within the SEER database. Although SEER data are considered highly valid,22 to our knowledge there are no published third-party evaluations of the validity and accuracy of this variable. The face validity of this variable could be evaluated because the recommendation to undergo surgical therapy and the receipt of surgical therapy were ascertained by 2 independent sources, SEER and Medicare, respectively. Although a patient who is recommended surgical therapy may or may not actually undergo resection, it is not likely that patients who are not recommended surgical therapy would undergo resection. Among patients who were not recommended surgical therapy, only 1% underwent resection; this proportion did not vary by race. One shortcoming of this variable is an inability to identify the type of provider (surgeon, oncologist, pulmonologist, etc) who made the recommendation and/or the context (both clinical and cultural) in which the recommendation was made. Another limitation of this study is that our findings may be generalizable only to elderly Medicare beneficiaries. However, given that the median age at diagnosis of lung cancer is 70 years28 and that Medicare provides health care coverage for more than 97% of Americans older than 65 years, this study might be considered to be representative of most patients with lung cancer. Although evidence shows that other racial groups such as Hispanic patients have disparate care and outcomes,31 we were not able to evaluate other racial groups because prior studies indicate that SEER and/or Medicare reliably measure only black or white race.17
In conclusion, even among patients with early-stage lung cancer who were recommended surgical therapy, there was large racial variation in the receipt of optimal cancer treatment after adjustment and despite uniform health care coverage. Although these findings do not refute the likely roles of health care system and provider biases and patient characteristics as important causal factors underlying health disparities, the findings do suggest that other factors (ie, distrust, perceptions and beliefs about lung cancer and its treatment, and limited access to subspecialty care) may have a more dominant role in causing disparities than previously recognized. The implication of these findings is that interventions designed to narrow gaps in health care should target structural aspects of care, providers, and patients and communities at risk for lung cancer and suboptimal care.
Accepted for Publication: August 24, 2008.
Correspondence: David R. Flum, MD, MPH, Department of Surgery, University of Washington, Box 356410, 1959 NE Pacific St, Seattle, WA 98195-6410 (firstname.lastname@example.org).
Author Contributions:Study concept and design: Farjah and Flum. Acquisition of data: Farjah and Flum. Analysis and interpretation of data: Farjah, Wood, Yanez, Vaughan, Symons, Krishnadasan, and Flum. Drafting of the manuscript: Farjah. Critical revision of the manuscript for important intellectual content: Wood, Yanez, Vaughan, Symons, Krishnadasan, and Flum. Statistical analysis: Yanez and Flum. Administrative, technical, and material support: Farjah, Symons, and Krishnadasan. Study supervision: Wood, Vaughan, and Flum.
Financial Disclosure: None reported.
Funding/Support: Dr Farjah was supported by Cancer Epidemiology and Biostatistics Training Grant T32 CA09168-30 and Ruth L. Kirschstein National Research Service Award F32 CA130434-01 from the National Cancer Institute. Additional resources were made available through the Department of Surgery and the Surgical Outcomes Research Center, University of Washington and the generosity of the Schilling family.
Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute or the National Institutes of Health. This study used the linked SEER-Medicare database. The interpretation and reporting of these data are the sole responsibility of the authors.
Additional Contributions: We acknowledge the efforts of the Applied Research Program, National Cancer Institute; the Office of Research, Development and Information, Centers for Medicare and Medicaid Services; Information Management Services, Inc; and the SEER Program tumor registries in the creation of the SEER-Medicare database.