Racial Differences in Surgical Evaluation, Treatment, and Outcome of Locoregional Esophageal Cancer: A Population-Based Analysis of Elderly
http://www.100md.com
▲還散笫雖悝◎
the Center for Clinical Decision Sciences, Department of Public Health, Erasmus MC, Rotterdam, the Netherlands
Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA
ABSTRACT
METHODS: We selected 2,946 white patients and 367 black patients who were older than 65 years and had clinically locoregional esophageal cancer in the Surveillance, Epidemiology, and End Results (SEER) registry (1991 to 1999). Treatment and outcome data were obtained from the linked SEER-Medicare databases. We used logistic regression analysis to estimate odds ratios (ORs) for being seen by a surgeon and for undergoing surgery. Cox proportional hazards analyses were performed to estimate hazard ratios (HRs) for survival adjusted for medical, nonmedical, and treatment characteristics.
RESULTS: The rate of surgery for black patients was half that of white patients (25% v 46%; OR, 0.38; P < .001), which was caused by both a lower rate of seeing a surgeon (70% v 78%; OR, 0.66; P < .001) and a lower rate of surgery once seen (35% v 59%; OR, 0.38; P < .001). These racial disparities were only partly explained by differences in patient and cancer characteristics, and not by nonmedical factors, such as socioeconomic status. The 2-year survival rate was lower for black patients (18% v 25%; HR, 1.18; P = .004), but this racial difference disappeared when corrected for treatment received (adjusted HR, 1.02; P = .80).
CONCLUSION: Underuse of potentially curative surgery is an important potential explanation for the poorer survival of black patients with locoregional esophageal cancer. Barriers to surgical evaluation and treatment need to be reduced, whether related to patient or healthcare system factors.
INTRODUCTION
Racial disparities in the use of cancer-directed treatments, including surgery, RT, and CT, have been documented for many cancers.6每10 Moreover, these disparities in treatment have been shown to account for differences in survival, even after adjustment or controlling for relevant medical and nonmedical factors.11每13 Racial disparities in patterns of care and outcome for esophageal cancer have received limited attention, however.14每16 Furthermore, factors mediating the observed racial disparities in use of cancer-directed treatment are poorly understood. Insight into these mediating processes may be obtained by studying receipt of treatment as a two-step process: referral to a specialist and treatment of referred patients.17 The magnitude of barriers at these two levels is important to guide any solutions that aim to reduce racial disparities in access to treatment.
We therefore undertook a population-based study of elderly patients diagnosed with locoregional esophageal cancer to determine whether there were racial disparities in the rates of surgical evaluation and of undergoing surgery once seen by a surgeon. We also studied whether such disparities in treatment could explain racial differences in survival, after statistical adjustment for medical and nonmedical characteristics.
METHODS
Because we used Medicare data, we included patients from the SEER registry who were 65 years or older. Patients were diagnosed with pathologically confirmed esophageal cancer between January 1, 1991, and December 31, 1999. We excluded patients for whom the date of death differed by more than 3 months between the SEER and Medicare databases, patients who were diagnosed from death certificate or autopsy, patients who were eligible for Medicare on the basis of end-stage renal failure or disability, patients for whom the month of diagnosis was not available, and patients with health maintenance organization enrollment within 8 months after diagnosis. We also excluded those with in situ disease, distant metastases, or unknown disease stage. Because the primary comparison of interest was between black and non-Hispanic white patients, we excluded patients of other races or ethnicities (n = 225). The study was approved by the Institutional Review Board of Dana-Farber/Partners Cancer Care.
Patient Characteristics
We classified patients by age at diagnosis, sex, year of diagnosis, census region (West, Midwest, Northeast, and South), and SES. SES was determined using the median income in the patient's census tract of residence using year 2000 data. If this information was missing, we used the zip code median income; if that was missing, we used the census tract per capita income; if that was missing, we used the zip code per capita income. If all of those were missing, the patient was classified in the lowest SES quintile.20 Because SES was not determined at the individual patient level, we refer to this variable as ecological SES. Cancer characteristics included histology (adenocarcinoma, squamous, mixed, or other carcinoma) and the stage of disease (categorized as local versus regional).
Stage of disease was based on clinical information in SEER, except for surgery patients, for whom staging information was updated with pathologic findings from surgery. This situation creates a potential misclassification bias because surgery patients who are initially staged as having clinically locoregional esophageal cancer may be reclassified as having distant disease based on pathologic findings. This phenomenon does not occur among patients treated with RT, CT, or supportive care only (ie, no cancer-directed treatments). To minimize the impact of this bias, we considered all such patients as having clinically regional disease (n = 283) because surgery typically is not indicated for clinically advanced esophageal cancer.10
Comorbidity was determined based on inpatient and outpatient Medicare claims from 13 months to 1 month before the date of diagnosis.21 Comorbidities were combined in the Charlson score22,23 and grouped as cardiovascular (previous myocardial infarction, heart failure, peripheral arterial disease, cerebrovascular disease), diabetes (with or without complications), pulmonary (chronic obstructive pulmonary disease), renal (mild to severe), or hepatic (mild to severe).24,25 The presence of comorbidity could not reliably be determined for patients younger than 66 years plus 1 month, because Medicare claims were only available from age 65 years and older.21 Missing values for the comorbidity index were statistically imputed based on the correlation with medical and nonmedical characteristics in logistic regression models.26
Treatment Characteristics
Patients were classified as ever having been treated at a teaching hospital if at least one bill for any health condition included a charge for medical education. We identified patients seen by a surgeon by searching for claims from a physician with any of the following Medicare provider specialty codes: 2 (general surgery), 28 (colorectal surgery), 33 (thoracic surgery), 49 (ambulatory surgical center), 91 (surgical oncology). In addition, all patients undergoing surgery were by definition seen by a surgeon. We studied the billing records of patients not undergoing surgery to determine whether they had ever seen a physician who had billed for an esophageal resection. In this way, both surgeons with and without specialty designation in Medicare would be captured.17
We considered surgery, RT, and CT administered up to 8 months after the registered date of diagnosis in the SEER database because this time frame is a reasonable interval during which patients would be expected to initiate cancer-directed treatment after a diagnosis of esophageal cancer. Surgery was identified from the Medicare and SEER databases. We used International Classification of Diseases (9th revision, clinical modification (ICD9-CM)27 codes, American Medical Association Current Procedural Terminology (CPT) codes, hospital Diagnosis-Related Groups (DRG) codes, and SEER identified cancer-directed surgery as a sensitive definition for any cancer-directed surgical procedures. We also considered a more stringent definition of surgery,28 which gave similar results.
Information on RT was based on SEER records and Medicare data.29 In SEER, this information is coded as cancer-directed radiation and corresponds to radiation given within the first 4 months after diagnosis. In the Medicare files, we used an algorithm that combined an in-hospital radiation indicator, ICD9-CM diagnosis codes and procedure codes, with CPT codes, DRG codes, and revenue center codes.
Information on CT was based on Medicare data only.30 We applied an algorithm that combined ICD9-CM codes, with Health Care Common Procedure Coding System codes, CPT codes, DRG codes, and revenue center codes.
Treatment was classified as supportive care only (ie, no cancer-directed treatment within 8 months after diagnosis), RT, CT, or CRT (the sequential or concurrent receipt of CT and RT). Patients who underwent surgery were further subclassified as surgery (including those with surgery alone, surgery and any adjuvant treatment, or surgery and neoadjuvant RT), or surgery and neoadjuvant CT or CRT (including those with preoperative CT, or sequential or concurrent CRT).3每5,31
Outcome
We analyzed survival as the time from the month of diagnosis until death from any cause. A censoring date was determined for patients last known to be alive based on the last available Medicare coverage month.
Statistical Analysis
We assessed associations between race or ethnicity and patient characteristics with {chi}2 tests for categoric variables and the nonparametric Mann-Whitney test for ordered or continuous variables (ecological SES, age). We used logistic regression to calculate an odds ratio (OR) for the effect of race on surgical evaluation or surgery. Statistical interaction terms were used to study potentially differential effects of race by patient characteristics. For example, to assess whether race had the same association with surgery for males and females, we included the interaction term race x sex in a model that also included race and sex as main effects.
Survival curves were constructed with the Kaplan-Meier method and compared using the log-rank test. We compared survival between black and white patients and within subgroups according to whether the patients underwent surgery. We used Cox proportional hazards regression to study racial differences in survival, expressed as hazard ratios (HRs). We used statistical interaction terms to study whether the association between race and survival differed by patient characteristics.
We hypothesized that any differences in rates of surgery and survival might be explained by medical and nonmedical factors, which were associated with race. These factors were sequentially considered in multivariable regression models. First, we adjusted for age, sex, and the following medical characteristics: Charlson comorbidity index, cancer histology, and stage of disease. Second, we adjusted for age, sex, and medical and nonmedical characteristics, including ecological SES and geographical region. We finally created a Cox regression model, which adjusted for age, sex, medical and nonmedical characteristics, and treatment received.
Analyses were performed with SAS (SAS Institute, Cary, NC), SPSS (SPSS Inc, Chicago, IL), and S-plus (version 6; Insightful Inc, Seattle, WA). We present 95% CIs for unadjusted and adjusted ORs and HRs of race (black v white), which do not include the value 1 if statistically significant.
RESULTS
Surgical Evaluation and Treatment
Black patients were more likely to be treated in a teaching hospital (69% v 57%), but were assessed by a surgeon less often (70% v 78%). Surgery was performed in 25% of black patients, compared with 46% of white patients (Table 1). Black patients were more likely to have RT as their only treatment (20% v 13%) or to have no cancer-directed treatment at all (no RT, CT, or surgery; 26% v 15%; Table 1).
The 21 percentage point absolute difference in the rate of surgery by race corresponds to an unadjusted OR of 0.38 (P < .001; Table 2). This disparity could only partly be explained by medical characteristics (adjusted OR, 0.52). Nonmedical characteristics (region, ecological SES) did not further explain the disparity (adjusted OR, 0.55). Details of the latter regression model are presented in the Appendix. We found no statistically significant differential effects by race in the rate of surgery for the medical or nonmedical characteristics (interaction test, P > .05), except for geographic region (interaction test, P = .01, 3 df). This regional difference was attributable to an especially low rate of surgery among black patients in the Midwest region (17% v 48% for white patients).
Appendix. Detailed Description of the Multivariable Models Used to Adjust the Relationship of Race With Surgical Evaluation, Surgery, and Survival Among 3,313 Patients With Locoregional Esophageal Cancer From the SEER Registry 1991每1999
To better understand the processes leading to lower rates of surgery among black patients, we examined the proportion of black and white patients who were ever assessed by a surgeon. The difference of 70% for black patients versus 78% for white patients (Table 1) represents an unadjusted OR of 0.66 (Table 2). Medical and nonmedical characteristics could not explain this difference (adjusted OR, 0.66 and 0.62, respectively; Table 2 and Appendix). There was more racial disparity in being seen by a surgeon among those without comorbidity (interaction P = .01) and among men (interaction test, P = .02). The racial disparity was similar across regions (interaction test, P = .70), with 79% of black patients seen by a surgeon in the Midwest versus 87% of white patients.
We then analyzed the effect of race in the receipt of surgery among those surgically evaluated. Of 258 black patients seen by a surgeon, 90 (35%) underwent surgery, compared with 1,351 (59%) of 2,307 white patients (unadjusted OR, 0.38; P < .001; Table 2). This difference was partially explained by medical and nonmedical characteristics (adjusted OR, 0.57 and 0.62, respectively; Table 2 and Appendix). The effect of race varied by region (interaction test, P = .006; 3 df), with an especially low rate of surgery among blacks seen by a surgeon in the Midwest (21% v 55% for white patients). In sum, these analyses indicate that racial disparities in the rate of surgery had two causes: a lower rate of surgical evaluation and a lower rate of surgery among those seen by a surgeon.
Survival
During follow-up, 2,704 of the 2,946 white patients and 351 of the 367 black patients died. The median follow-up of 258 surviving patients was 5.1 year. The 6-month, and 1-, 2-, and 5-year survival rates were 58%, 39%, 18%, and 8% for black patients, compared with 64%, 43%, 25%, and 11% for white patients (log-rank P = .004; Fig 1). These racial differences represent an unadjusted HR of 1.18 (P = .004). The HR increased to 1.22 when we adjusted the racial difference in survival for age, sex, and medical characteristics (Table 2). This increase reflects that the risk profile of black patients was prognostically slightly more favorable than that of white patients (Table 1).
When we adjusted for ecological SES, the HR decreased from 1.22 to 1.18. When we also adjusted for region, the HR decreased further to 1.16. Adjusting for receipt of surgery decreased the HR to 1.11. In this regression model, undergoing surgery was associated with a strong beneficial effect (adjusted HR for the effect of surgery 0.64; 95% CI, 0.59 to 0.69). Adjusting for combinations of surgery, RT, and CT decreased the HR for racial difference to 1.02 (P = .80). In sum, some of the 22% higher mortality among black patients could be explained by nonmedical factors (HR decreased from 1.22 to 1.16), but a much greater proportion could be explained by treatment received (HR decreased from 1.16 to 1.02).
We performed a number of subgroup analyses. Survival analyses in the subgroup with squamous cell histology (n = 1,669; Table 3) showed similar results (Table 2). For example, the adjusted OR for receipt of surgery was 0.61 compared with 0.55 in the complete cohort, and the adjusted HR for survival was 1.00 compared with 1.02 in the complete cohort. There were not enough black patients with adenocarcinoma to carry out meaningful analyses in this subgroup. Furthermore, among those in the complete cohort who underwent surgery (n = 1,441), the adjusted HR was 1.05 (95% CI, 0.82 to 1.35; P = .68) and no racial difference was found among those not undergoing surgery (n = 1,872; adjusted HR, 1.00; 95% CI, 0.86 to 1.16; P = .99).
DISCUSSION
Black patients differed from white patients in many respects at presentation. Black patients were younger and had earlier stage disease, but they also had more comorbidity, a factor known to be associated with poorer survival.22 Black patients were also much more likely to have squamous cell histology.16 We therefore specifically studied the subgroup of patients with this histology and found similar racial disparities as those in the complete cohort (Table 3). Overall, black patients had a slightly more favorable prognostic profile with respect to medical factors. This is in contrast to other studies,32 and may be explained by our selection of patients with clinically locoregional disease rather than all patients with esophageal cancer. With respect to nonmedical factors, we found that a higher proportion of black patients were from the lowest ecological SES strata, which explained some of the racial disparity in survival.33,34 However, the effect of ecological SES was likely blunted in our population because all patients were covered by Medicare. In those younger than age 65, the effects of SES are likely larger, contributing more to racial disparities in treatment and outcome.
After controlling for medical and nonmedical factors, we still found a difference in survival by race (HR, 1.16). Among those undergoing surgery, the racial disparity was small, however (HR, 1.05), after adjustment for patient characteristics and other treatments (neoadjuvant chemotherapy or chemoradiotherapy). Among those not undergoing surgery, we found substantial differences in the use of other cancer-directed treatments, with a less frequent use of CT or CRT among black patients. When we adjusted for these characteristics, race had no residual effect (HR, 1.00). Therefore, our data suggest that treatment choice is the main factor accounting for racial disparities in outcomes for patients with locoregional esophageal cancer, rather than differential survival among similarly treated patients. This finding is in agreement with previous analyses for lung cancer,11 but in contrast with analyses for prostate cancer,10 for which racial differences in outcome remained after adjusting for treatment received.
We found racial disparities for surgical evaluation and for treatment once seen by a surgeon. This suggests that there are two opportunities to increase rates of surgery, either by increasing referral or by decreasing barriers to treatment once seen by a specialist.17 We observed the largest racial disparity in rate of surgery in the Midwest, which is also the SEER region with the largest proportion of black patients in our study. The relatively low rate of surgery was especially attributed to a lower rate among those seen by a surgeon, which indicates that barriers to treatment may especially exist at this step in the treatment selection process.
It is difficult to comment on the appropriate rate of surgery and other treatment in this cohort. Our analyses identified several factors in addition to race associated with a decreased use of surgery (eg, older age, squamous cell histology, and comorbidity; Appendix). However, it is interesting that treatment received fully accounted for racial differences in survival. Surgery revealed pathologically distant disease at a similar frequency among black and white patients (22%, 20 of 90; 19%, 263 of 1,351, respectively), and was similarly effective in improving survival (adjusted HR, 0.71; 95% CI, 0.56 to 0.91 and adjusted HR, 0.63; 95% CI, 0.58 to 0.68, respectively; interaction test, P = .33). Thus, it may be appropriate to assume a single effect for surgery in esophageal cancer patients (adjusted HR, 0.64). This is in contrast to a recent analysis of women with endometrial cancer, where the association between surgery and survival was substantially stronger among white women (HR, 0.26; 95% CI, 0.23 to 0.29) than among black women (HR, 0.44; 95% CI, 0.32 to 0.59).35
Our study has some limitations. First, the observational and administrative nature of the data precludes firm conclusions about the influence of race or ethnicity on treatment and outcome. The data do not capture all medical factors relevant to deciding whether a patient is a surgical candidate, such as performance status, and exact cancer location in the esophagus. For example, upper esophageal tumors are more likely to be of squamous cell type and treated primarily with radiotherapy. Still, location in the esophagus would not be expected to be different by race, and the findings were similar in the squamous subgroup. In addition, staging information is limited in SEER. Furthermore, SES is measured at the census tract level instead of at the individual level. Hence, the observed racial differences may still partly reflect socioeconomic differences, even after adjustment for ecological SES.36 Similarly, we were only able to capture comorbidities serious enough to result in the use of medical services before diagnosis, which likely underestimates the true presence of coexistent illnesses.21 Finally, we relied on statistical modeling to adjust for various confounding factors, but these models can only approximate underlying patterns.
Because of the nature of our data source, we cannot completely explain why racial differences in treatment occur; in particular, the role of patients' preferences versus specialists' biases is unclear. Others have found that black patients have less trust in their physicians,37 which reduces their willingness to undergo risky procedures,38 and may be more likely to believe that cure is possible without surgery.39 If racial differences in treatment reflect informed choices rather than barriers to care, one could argue that they are not cause for alarm.40 However, there is a growing consensus that when these choices adversely affect survival, we must identify and implement strategies to improve communication and gain the trust of these patients.41 Our finding that survival differed little by race among those who underwent surgery supports the value of interventions to increase the use of surgery among black patients with locoregional esophageal cancer. If other treatments, such as neoadjuvant CRT, prove to be of clinical value in this disease, it will be important to ensure that black patients do not selectively fail to realize the benefits of these newer treatments as well. The time has come for us to move beyond just trying to remove barriers to access to care and actually work to engage vulnerable patients and find ways to facilitate their participation in care.
Authors' Disclosures of Potential Conflicts of Interest
Acknowledgment
We thank Johan P. Mackenbach, Department of Public Health, Erasmus MC, Rotterdam, the Netherlands, and John D. Urschel, Department of Surgery, Beth Israel Deaconess Medical Center, Boston, MA, for comments on a previous version of this manuscript.
NOTES
E.W.S. was supported by a fellowship from the Royal Netherlands Academy for Arts and Sciences. C.C.E. was supported in part by a Dunkin' Donuts Rising Star award.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Leonard GD, McCaffrey JA, Maher M: Optimal therapy for oesophageal cancer. Cancer Treat Rev 29:275每282, 2003
2. Wu PC, Posner MC: The role of surgery in the management of oesophageal cancer. Lancet Oncol 4:481每488, 2003
3. Geh JI: The use of chemoradiotherapy in oesophageal cancer. Eur J Cancer 38:300每313, 2002
4. Kaklamanos IG, Walker GR, Ferry K, et al: Neoadjuvant treatment for resectable cancer of the esophagus and the gastroesophageal junction: A meta-analysis of randomized clinical trials. Ann Surg Oncol 10:754每761, 2003
5. Urschel JD, Vasan H: A meta-analysis of randomized controlled trials that compared neoadjuvant chemoradiation and surgery to surgery alone for resectable esophageal cancer. Am J Surg 185:538每543, 2003
6. Shavers VL, Brown ML: Racial and ethnic disparities in the receipt of cancer treatment. J Natl Cancer Inst 94:334每357, 2002
7. Ayanian JZ, Zaslavsky AM, Fuchs CS, et al: Use of adjuvant chemotherapy and radiation therapy for colorectal cancer in a population-based cohort. J Clin Oncol 21:1293每1300, 2003
8. O'Malley CD, Le GM, Glaser SL, et al: Socioeconomic status and breast carcinoma survival in four racial/ethnic groups: A population-based study. Cancer 97:1303每1311, 2003
9. Konety BR, Joslyn SA: Factors influencing aggressive therapy for bladder cancer: An analysis of data from the SEER program. J Urol 170:1765每1771, 2003
10. Godley PA, Schenck AP, Amamoo MA, et al: Racial differences in mortality among Medicare recipients after treatment for localized prostate cancer. J Natl Cancer Inst 95:1702每1710, 2003
11. Bach PB, Cramer LD, Warren JL, et al: Racial differences in the treatment of early-stage lung cancer. N Engl J Med 341:1198每1205, 1999
12. Shavers VL, Harlan LC, Winn D, et al: Racial/ethnic patterns of care for cancers of the oral cavity, pharynx, larynx, sinuses, and salivary glands. Cancer Metastasis Rev 22:25每38, 2003
13. Li CI, Malone KE, Daling JR: Differences in breast cancer stage, treatment, and survival by race and ethnicity. Arch Intern Med 163:49每56, 2003
14. Dominitz JA, Maynard C, Billingsley KG, et al: Race, treatment, and survival of veterans with cancer of the distal esophagus and gastric cardia. Med Care 40:I14每I26, 2002 (1 suppl)
15. Polednak AP: Survival of U.S. black and white patients with squamous cell cancer of the esophagus. J Natl Med Assoc 96:87每92, 2004
16. Streeter OE Jr, Martz KL, Gaspar LE, et al: Does race influence survival for esophageal cancer patients treated on the radiation and chemotherapy arm of RTOG #85-01 Int J Radiat Oncol Biol Phys 44:1047每1052, 1999
17. Earle CC, Neumann PJ, Gelber RD, et al: Impact of referral patterns on the use of chemotherapy for lung cancer. J Clin Oncol 20:1786每1792, 2002
18. Potosky AL, Riley GF, Lubitz JD, et al: Potential for cancer related health services research using a linked Medicare-tumor registry database. Med Care 31:732每748, 1993
19. Warren JL, Klabunde CN, Schrag D, et al: Overview of the SEER-Medicare data: Content, research applications, and generalizability to the United States elderly population. Med Care 40:IV-3每18, 2002 (8 suppl)
20. Bach PB, Guadagnoli E, Schrag D, et al: Patient demographic and socioeconomic characteristics in the SEER-Medicare database applications and limitations. Med Care 40:IV-19每25, 2002 (8 suppl)
21. Klabunde CN, Warren JL, Legler JM: Assessing comorbidity using claims data: An overview. Med Care 40:IV-26每35, 2002 (8 suppl)
22. Charlson ME, Pompei P, Ales KL, et al: A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. J Chronic Dis 40:373每383, 1987
23. Deyo RA, Cherkin DC, Ciol MA: Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 45:613每619, 1992
24. Bartels H, Stein HJ, Siewert JR: Preoperative risk analysis and postoperative mortality of oesophagectomy for resectable oesophageal cancer. Br J Surg 85:840每844, 1998
25. Coebergh JW, Janssen-Heijnen ML, Post PN, et al: Serious co-morbidity among unselected cancer patients newly diagnosed in the southeastern part of the Netherlands in 1993-1996. J Clin Epidemiol 52:1131每1136, 1999
26. Harrell FE Jr, Lee KL, Mark DB: Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 15:361每387, 1996
27. Cooper GS, Virnig B, Klabunde CN, et al: Use of SEER-Medicare data for measuring cancer surgery. Med Care 40:IV-43每48, 2002 (8 suppl)
28. Finlayson EV, Birkmeyer JD: Operative mortality with elective surgery in older adults. Eff Clin Pract 4:172每177, 2001
29. Virnig BA, Warren JL, Cooper GS, et al: Studying radiation therapy using SEER-Medicare-linked data. Med Care 40:IV-49每54, 2002 (8 suppl)
30. Warren JL, Harlan LC, Fahey A, et al: Utility of the SEER-Medicare data to identify chemotherapy use. Med Care 40:IV-55每61, 2002 (8 suppl)
31. Urschel JD, Vasan H, Blewett CJ: A meta-analysis of randomized controlled trials that compared neoadjuvant chemotherapy and surgery to surgery alone for resectable esophageal cancer. Am J Surg 183:274每279, 2002
32. Bach PB, Schrag D, Brawley OW, et al: Survival of blacks and whites after a cancer diagnosis. JAMA 287:2106每2113, 2002
33. Schrijvers CT, Mackenbach JP: Cancer patient survival by socioeconomic status in seven countries: A review for six common cancer sites. J Epidemiol Community Health 48:441每446, 1994
34. Cross CK, Harris J, Recht A: Race, socioeconomic status, and breast carcinoma in the U.S: What have we learned from clinical studies. Cancer 95:1988每1999, 2002
35. Randall TC, Armstrong K: Differences in treatment and outcome between African-American and white women with endometrial cancer. J Clin Oncol 21:4200每4206, 2003
36. Dale W, Vijayakumar S, Lawlor EF, et al: Prostate cancer, race, and socioeconomic status: Inadequate adjustment for social factors in assessing racial differences. Prostate 29:271每281, 1996
37. Boulware LE, Cooper LA, Ratner LE, et al: Race and trust in the health care system. Public Health Rep 118:358每365, 2003
38. Collins TC, Clark JA, Petersen LA, et al: Racial differences in how patients perceive physician communication regarding cardiac testing. Med Care 40:I27每I34, 2002 (1 suppl)
39. Cykert S, Phifer N: Surgical decisions for early stage, non-small cell lung cancer: Which racially sensitive perceptions of cancer are likely to explain racial variation in surgery Med Decis Making 23:167每176, 2003
40. Schulman KA, Berlin JA, Harless W, et al: The effect of race and sex on physicians' recommendations for cardiac catheterization. N Engl J Med 340:618每626, 1999
41. Katz JN: Patient preferences and health disparities. JAMA 286:1506每1509, 2001(Ewout W. Steyerberg, Crai)
Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA
ABSTRACT
METHODS: We selected 2,946 white patients and 367 black patients who were older than 65 years and had clinically locoregional esophageal cancer in the Surveillance, Epidemiology, and End Results (SEER) registry (1991 to 1999). Treatment and outcome data were obtained from the linked SEER-Medicare databases. We used logistic regression analysis to estimate odds ratios (ORs) for being seen by a surgeon and for undergoing surgery. Cox proportional hazards analyses were performed to estimate hazard ratios (HRs) for survival adjusted for medical, nonmedical, and treatment characteristics.
RESULTS: The rate of surgery for black patients was half that of white patients (25% v 46%; OR, 0.38; P < .001), which was caused by both a lower rate of seeing a surgeon (70% v 78%; OR, 0.66; P < .001) and a lower rate of surgery once seen (35% v 59%; OR, 0.38; P < .001). These racial disparities were only partly explained by differences in patient and cancer characteristics, and not by nonmedical factors, such as socioeconomic status. The 2-year survival rate was lower for black patients (18% v 25%; HR, 1.18; P = .004), but this racial difference disappeared when corrected for treatment received (adjusted HR, 1.02; P = .80).
CONCLUSION: Underuse of potentially curative surgery is an important potential explanation for the poorer survival of black patients with locoregional esophageal cancer. Barriers to surgical evaluation and treatment need to be reduced, whether related to patient or healthcare system factors.
INTRODUCTION
Racial disparities in the use of cancer-directed treatments, including surgery, RT, and CT, have been documented for many cancers.6每10 Moreover, these disparities in treatment have been shown to account for differences in survival, even after adjustment or controlling for relevant medical and nonmedical factors.11每13 Racial disparities in patterns of care and outcome for esophageal cancer have received limited attention, however.14每16 Furthermore, factors mediating the observed racial disparities in use of cancer-directed treatment are poorly understood. Insight into these mediating processes may be obtained by studying receipt of treatment as a two-step process: referral to a specialist and treatment of referred patients.17 The magnitude of barriers at these two levels is important to guide any solutions that aim to reduce racial disparities in access to treatment.
We therefore undertook a population-based study of elderly patients diagnosed with locoregional esophageal cancer to determine whether there were racial disparities in the rates of surgical evaluation and of undergoing surgery once seen by a surgeon. We also studied whether such disparities in treatment could explain racial differences in survival, after statistical adjustment for medical and nonmedical characteristics.
METHODS
Because we used Medicare data, we included patients from the SEER registry who were 65 years or older. Patients were diagnosed with pathologically confirmed esophageal cancer between January 1, 1991, and December 31, 1999. We excluded patients for whom the date of death differed by more than 3 months between the SEER and Medicare databases, patients who were diagnosed from death certificate or autopsy, patients who were eligible for Medicare on the basis of end-stage renal failure or disability, patients for whom the month of diagnosis was not available, and patients with health maintenance organization enrollment within 8 months after diagnosis. We also excluded those with in situ disease, distant metastases, or unknown disease stage. Because the primary comparison of interest was between black and non-Hispanic white patients, we excluded patients of other races or ethnicities (n = 225). The study was approved by the Institutional Review Board of Dana-Farber/Partners Cancer Care.
Patient Characteristics
We classified patients by age at diagnosis, sex, year of diagnosis, census region (West, Midwest, Northeast, and South), and SES. SES was determined using the median income in the patient's census tract of residence using year 2000 data. If this information was missing, we used the zip code median income; if that was missing, we used the census tract per capita income; if that was missing, we used the zip code per capita income. If all of those were missing, the patient was classified in the lowest SES quintile.20 Because SES was not determined at the individual patient level, we refer to this variable as ecological SES. Cancer characteristics included histology (adenocarcinoma, squamous, mixed, or other carcinoma) and the stage of disease (categorized as local versus regional).
Stage of disease was based on clinical information in SEER, except for surgery patients, for whom staging information was updated with pathologic findings from surgery. This situation creates a potential misclassification bias because surgery patients who are initially staged as having clinically locoregional esophageal cancer may be reclassified as having distant disease based on pathologic findings. This phenomenon does not occur among patients treated with RT, CT, or supportive care only (ie, no cancer-directed treatments). To minimize the impact of this bias, we considered all such patients as having clinically regional disease (n = 283) because surgery typically is not indicated for clinically advanced esophageal cancer.10
Comorbidity was determined based on inpatient and outpatient Medicare claims from 13 months to 1 month before the date of diagnosis.21 Comorbidities were combined in the Charlson score22,23 and grouped as cardiovascular (previous myocardial infarction, heart failure, peripheral arterial disease, cerebrovascular disease), diabetes (with or without complications), pulmonary (chronic obstructive pulmonary disease), renal (mild to severe), or hepatic (mild to severe).24,25 The presence of comorbidity could not reliably be determined for patients younger than 66 years plus 1 month, because Medicare claims were only available from age 65 years and older.21 Missing values for the comorbidity index were statistically imputed based on the correlation with medical and nonmedical characteristics in logistic regression models.26
Treatment Characteristics
Patients were classified as ever having been treated at a teaching hospital if at least one bill for any health condition included a charge for medical education. We identified patients seen by a surgeon by searching for claims from a physician with any of the following Medicare provider specialty codes: 2 (general surgery), 28 (colorectal surgery), 33 (thoracic surgery), 49 (ambulatory surgical center), 91 (surgical oncology). In addition, all patients undergoing surgery were by definition seen by a surgeon. We studied the billing records of patients not undergoing surgery to determine whether they had ever seen a physician who had billed for an esophageal resection. In this way, both surgeons with and without specialty designation in Medicare would be captured.17
We considered surgery, RT, and CT administered up to 8 months after the registered date of diagnosis in the SEER database because this time frame is a reasonable interval during which patients would be expected to initiate cancer-directed treatment after a diagnosis of esophageal cancer. Surgery was identified from the Medicare and SEER databases. We used International Classification of Diseases (9th revision, clinical modification (ICD9-CM)27 codes, American Medical Association Current Procedural Terminology (CPT) codes, hospital Diagnosis-Related Groups (DRG) codes, and SEER identified cancer-directed surgery as a sensitive definition for any cancer-directed surgical procedures. We also considered a more stringent definition of surgery,28 which gave similar results.
Information on RT was based on SEER records and Medicare data.29 In SEER, this information is coded as cancer-directed radiation and corresponds to radiation given within the first 4 months after diagnosis. In the Medicare files, we used an algorithm that combined an in-hospital radiation indicator, ICD9-CM diagnosis codes and procedure codes, with CPT codes, DRG codes, and revenue center codes.
Information on CT was based on Medicare data only.30 We applied an algorithm that combined ICD9-CM codes, with Health Care Common Procedure Coding System codes, CPT codes, DRG codes, and revenue center codes.
Treatment was classified as supportive care only (ie, no cancer-directed treatment within 8 months after diagnosis), RT, CT, or CRT (the sequential or concurrent receipt of CT and RT). Patients who underwent surgery were further subclassified as surgery (including those with surgery alone, surgery and any adjuvant treatment, or surgery and neoadjuvant RT), or surgery and neoadjuvant CT or CRT (including those with preoperative CT, or sequential or concurrent CRT).3每5,31
Outcome
We analyzed survival as the time from the month of diagnosis until death from any cause. A censoring date was determined for patients last known to be alive based on the last available Medicare coverage month.
Statistical Analysis
We assessed associations between race or ethnicity and patient characteristics with {chi}2 tests for categoric variables and the nonparametric Mann-Whitney test for ordered or continuous variables (ecological SES, age). We used logistic regression to calculate an odds ratio (OR) for the effect of race on surgical evaluation or surgery. Statistical interaction terms were used to study potentially differential effects of race by patient characteristics. For example, to assess whether race had the same association with surgery for males and females, we included the interaction term race x sex in a model that also included race and sex as main effects.
Survival curves were constructed with the Kaplan-Meier method and compared using the log-rank test. We compared survival between black and white patients and within subgroups according to whether the patients underwent surgery. We used Cox proportional hazards regression to study racial differences in survival, expressed as hazard ratios (HRs). We used statistical interaction terms to study whether the association between race and survival differed by patient characteristics.
We hypothesized that any differences in rates of surgery and survival might be explained by medical and nonmedical factors, which were associated with race. These factors were sequentially considered in multivariable regression models. First, we adjusted for age, sex, and the following medical characteristics: Charlson comorbidity index, cancer histology, and stage of disease. Second, we adjusted for age, sex, and medical and nonmedical characteristics, including ecological SES and geographical region. We finally created a Cox regression model, which adjusted for age, sex, medical and nonmedical characteristics, and treatment received.
Analyses were performed with SAS (SAS Institute, Cary, NC), SPSS (SPSS Inc, Chicago, IL), and S-plus (version 6; Insightful Inc, Seattle, WA). We present 95% CIs for unadjusted and adjusted ORs and HRs of race (black v white), which do not include the value 1 if statistically significant.
RESULTS
Surgical Evaluation and Treatment
Black patients were more likely to be treated in a teaching hospital (69% v 57%), but were assessed by a surgeon less often (70% v 78%). Surgery was performed in 25% of black patients, compared with 46% of white patients (Table 1). Black patients were more likely to have RT as their only treatment (20% v 13%) or to have no cancer-directed treatment at all (no RT, CT, or surgery; 26% v 15%; Table 1).
The 21 percentage point absolute difference in the rate of surgery by race corresponds to an unadjusted OR of 0.38 (P < .001; Table 2). This disparity could only partly be explained by medical characteristics (adjusted OR, 0.52). Nonmedical characteristics (region, ecological SES) did not further explain the disparity (adjusted OR, 0.55). Details of the latter regression model are presented in the Appendix. We found no statistically significant differential effects by race in the rate of surgery for the medical or nonmedical characteristics (interaction test, P > .05), except for geographic region (interaction test, P = .01, 3 df). This regional difference was attributable to an especially low rate of surgery among black patients in the Midwest region (17% v 48% for white patients).
Appendix. Detailed Description of the Multivariable Models Used to Adjust the Relationship of Race With Surgical Evaluation, Surgery, and Survival Among 3,313 Patients With Locoregional Esophageal Cancer From the SEER Registry 1991每1999
To better understand the processes leading to lower rates of surgery among black patients, we examined the proportion of black and white patients who were ever assessed by a surgeon. The difference of 70% for black patients versus 78% for white patients (Table 1) represents an unadjusted OR of 0.66 (Table 2). Medical and nonmedical characteristics could not explain this difference (adjusted OR, 0.66 and 0.62, respectively; Table 2 and Appendix). There was more racial disparity in being seen by a surgeon among those without comorbidity (interaction P = .01) and among men (interaction test, P = .02). The racial disparity was similar across regions (interaction test, P = .70), with 79% of black patients seen by a surgeon in the Midwest versus 87% of white patients.
We then analyzed the effect of race in the receipt of surgery among those surgically evaluated. Of 258 black patients seen by a surgeon, 90 (35%) underwent surgery, compared with 1,351 (59%) of 2,307 white patients (unadjusted OR, 0.38; P < .001; Table 2). This difference was partially explained by medical and nonmedical characteristics (adjusted OR, 0.57 and 0.62, respectively; Table 2 and Appendix). The effect of race varied by region (interaction test, P = .006; 3 df), with an especially low rate of surgery among blacks seen by a surgeon in the Midwest (21% v 55% for white patients). In sum, these analyses indicate that racial disparities in the rate of surgery had two causes: a lower rate of surgical evaluation and a lower rate of surgery among those seen by a surgeon.
Survival
During follow-up, 2,704 of the 2,946 white patients and 351 of the 367 black patients died. The median follow-up of 258 surviving patients was 5.1 year. The 6-month, and 1-, 2-, and 5-year survival rates were 58%, 39%, 18%, and 8% for black patients, compared with 64%, 43%, 25%, and 11% for white patients (log-rank P = .004; Fig 1). These racial differences represent an unadjusted HR of 1.18 (P = .004). The HR increased to 1.22 when we adjusted the racial difference in survival for age, sex, and medical characteristics (Table 2). This increase reflects that the risk profile of black patients was prognostically slightly more favorable than that of white patients (Table 1).
When we adjusted for ecological SES, the HR decreased from 1.22 to 1.18. When we also adjusted for region, the HR decreased further to 1.16. Adjusting for receipt of surgery decreased the HR to 1.11. In this regression model, undergoing surgery was associated with a strong beneficial effect (adjusted HR for the effect of surgery 0.64; 95% CI, 0.59 to 0.69). Adjusting for combinations of surgery, RT, and CT decreased the HR for racial difference to 1.02 (P = .80). In sum, some of the 22% higher mortality among black patients could be explained by nonmedical factors (HR decreased from 1.22 to 1.16), but a much greater proportion could be explained by treatment received (HR decreased from 1.16 to 1.02).
We performed a number of subgroup analyses. Survival analyses in the subgroup with squamous cell histology (n = 1,669; Table 3) showed similar results (Table 2). For example, the adjusted OR for receipt of surgery was 0.61 compared with 0.55 in the complete cohort, and the adjusted HR for survival was 1.00 compared with 1.02 in the complete cohort. There were not enough black patients with adenocarcinoma to carry out meaningful analyses in this subgroup. Furthermore, among those in the complete cohort who underwent surgery (n = 1,441), the adjusted HR was 1.05 (95% CI, 0.82 to 1.35; P = .68) and no racial difference was found among those not undergoing surgery (n = 1,872; adjusted HR, 1.00; 95% CI, 0.86 to 1.16; P = .99).
DISCUSSION
Black patients differed from white patients in many respects at presentation. Black patients were younger and had earlier stage disease, but they also had more comorbidity, a factor known to be associated with poorer survival.22 Black patients were also much more likely to have squamous cell histology.16 We therefore specifically studied the subgroup of patients with this histology and found similar racial disparities as those in the complete cohort (Table 3). Overall, black patients had a slightly more favorable prognostic profile with respect to medical factors. This is in contrast to other studies,32 and may be explained by our selection of patients with clinically locoregional disease rather than all patients with esophageal cancer. With respect to nonmedical factors, we found that a higher proportion of black patients were from the lowest ecological SES strata, which explained some of the racial disparity in survival.33,34 However, the effect of ecological SES was likely blunted in our population because all patients were covered by Medicare. In those younger than age 65, the effects of SES are likely larger, contributing more to racial disparities in treatment and outcome.
After controlling for medical and nonmedical factors, we still found a difference in survival by race (HR, 1.16). Among those undergoing surgery, the racial disparity was small, however (HR, 1.05), after adjustment for patient characteristics and other treatments (neoadjuvant chemotherapy or chemoradiotherapy). Among those not undergoing surgery, we found substantial differences in the use of other cancer-directed treatments, with a less frequent use of CT or CRT among black patients. When we adjusted for these characteristics, race had no residual effect (HR, 1.00). Therefore, our data suggest that treatment choice is the main factor accounting for racial disparities in outcomes for patients with locoregional esophageal cancer, rather than differential survival among similarly treated patients. This finding is in agreement with previous analyses for lung cancer,11 but in contrast with analyses for prostate cancer,10 for which racial differences in outcome remained after adjusting for treatment received.
We found racial disparities for surgical evaluation and for treatment once seen by a surgeon. This suggests that there are two opportunities to increase rates of surgery, either by increasing referral or by decreasing barriers to treatment once seen by a specialist.17 We observed the largest racial disparity in rate of surgery in the Midwest, which is also the SEER region with the largest proportion of black patients in our study. The relatively low rate of surgery was especially attributed to a lower rate among those seen by a surgeon, which indicates that barriers to treatment may especially exist at this step in the treatment selection process.
It is difficult to comment on the appropriate rate of surgery and other treatment in this cohort. Our analyses identified several factors in addition to race associated with a decreased use of surgery (eg, older age, squamous cell histology, and comorbidity; Appendix). However, it is interesting that treatment received fully accounted for racial differences in survival. Surgery revealed pathologically distant disease at a similar frequency among black and white patients (22%, 20 of 90; 19%, 263 of 1,351, respectively), and was similarly effective in improving survival (adjusted HR, 0.71; 95% CI, 0.56 to 0.91 and adjusted HR, 0.63; 95% CI, 0.58 to 0.68, respectively; interaction test, P = .33). Thus, it may be appropriate to assume a single effect for surgery in esophageal cancer patients (adjusted HR, 0.64). This is in contrast to a recent analysis of women with endometrial cancer, where the association between surgery and survival was substantially stronger among white women (HR, 0.26; 95% CI, 0.23 to 0.29) than among black women (HR, 0.44; 95% CI, 0.32 to 0.59).35
Our study has some limitations. First, the observational and administrative nature of the data precludes firm conclusions about the influence of race or ethnicity on treatment and outcome. The data do not capture all medical factors relevant to deciding whether a patient is a surgical candidate, such as performance status, and exact cancer location in the esophagus. For example, upper esophageal tumors are more likely to be of squamous cell type and treated primarily with radiotherapy. Still, location in the esophagus would not be expected to be different by race, and the findings were similar in the squamous subgroup. In addition, staging information is limited in SEER. Furthermore, SES is measured at the census tract level instead of at the individual level. Hence, the observed racial differences may still partly reflect socioeconomic differences, even after adjustment for ecological SES.36 Similarly, we were only able to capture comorbidities serious enough to result in the use of medical services before diagnosis, which likely underestimates the true presence of coexistent illnesses.21 Finally, we relied on statistical modeling to adjust for various confounding factors, but these models can only approximate underlying patterns.
Because of the nature of our data source, we cannot completely explain why racial differences in treatment occur; in particular, the role of patients' preferences versus specialists' biases is unclear. Others have found that black patients have less trust in their physicians,37 which reduces their willingness to undergo risky procedures,38 and may be more likely to believe that cure is possible without surgery.39 If racial differences in treatment reflect informed choices rather than barriers to care, one could argue that they are not cause for alarm.40 However, there is a growing consensus that when these choices adversely affect survival, we must identify and implement strategies to improve communication and gain the trust of these patients.41 Our finding that survival differed little by race among those who underwent surgery supports the value of interventions to increase the use of surgery among black patients with locoregional esophageal cancer. If other treatments, such as neoadjuvant CRT, prove to be of clinical value in this disease, it will be important to ensure that black patients do not selectively fail to realize the benefits of these newer treatments as well. The time has come for us to move beyond just trying to remove barriers to access to care and actually work to engage vulnerable patients and find ways to facilitate their participation in care.
Authors' Disclosures of Potential Conflicts of Interest
Acknowledgment
We thank Johan P. Mackenbach, Department of Public Health, Erasmus MC, Rotterdam, the Netherlands, and John D. Urschel, Department of Surgery, Beth Israel Deaconess Medical Center, Boston, MA, for comments on a previous version of this manuscript.
NOTES
E.W.S. was supported by a fellowship from the Royal Netherlands Academy for Arts and Sciences. C.C.E. was supported in part by a Dunkin' Donuts Rising Star award.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Leonard GD, McCaffrey JA, Maher M: Optimal therapy for oesophageal cancer. Cancer Treat Rev 29:275每282, 2003
2. Wu PC, Posner MC: The role of surgery in the management of oesophageal cancer. Lancet Oncol 4:481每488, 2003
3. Geh JI: The use of chemoradiotherapy in oesophageal cancer. Eur J Cancer 38:300每313, 2002
4. Kaklamanos IG, Walker GR, Ferry K, et al: Neoadjuvant treatment for resectable cancer of the esophagus and the gastroesophageal junction: A meta-analysis of randomized clinical trials. Ann Surg Oncol 10:754每761, 2003
5. Urschel JD, Vasan H: A meta-analysis of randomized controlled trials that compared neoadjuvant chemoradiation and surgery to surgery alone for resectable esophageal cancer. Am J Surg 185:538每543, 2003
6. Shavers VL, Brown ML: Racial and ethnic disparities in the receipt of cancer treatment. J Natl Cancer Inst 94:334每357, 2002
7. Ayanian JZ, Zaslavsky AM, Fuchs CS, et al: Use of adjuvant chemotherapy and radiation therapy for colorectal cancer in a population-based cohort. J Clin Oncol 21:1293每1300, 2003
8. O'Malley CD, Le GM, Glaser SL, et al: Socioeconomic status and breast carcinoma survival in four racial/ethnic groups: A population-based study. Cancer 97:1303每1311, 2003
9. Konety BR, Joslyn SA: Factors influencing aggressive therapy for bladder cancer: An analysis of data from the SEER program. J Urol 170:1765每1771, 2003
10. Godley PA, Schenck AP, Amamoo MA, et al: Racial differences in mortality among Medicare recipients after treatment for localized prostate cancer. J Natl Cancer Inst 95:1702每1710, 2003
11. Bach PB, Cramer LD, Warren JL, et al: Racial differences in the treatment of early-stage lung cancer. N Engl J Med 341:1198每1205, 1999
12. Shavers VL, Harlan LC, Winn D, et al: Racial/ethnic patterns of care for cancers of the oral cavity, pharynx, larynx, sinuses, and salivary glands. Cancer Metastasis Rev 22:25每38, 2003
13. Li CI, Malone KE, Daling JR: Differences in breast cancer stage, treatment, and survival by race and ethnicity. Arch Intern Med 163:49每56, 2003
14. Dominitz JA, Maynard C, Billingsley KG, et al: Race, treatment, and survival of veterans with cancer of the distal esophagus and gastric cardia. Med Care 40:I14每I26, 2002 (1 suppl)
15. Polednak AP: Survival of U.S. black and white patients with squamous cell cancer of the esophagus. J Natl Med Assoc 96:87每92, 2004
16. Streeter OE Jr, Martz KL, Gaspar LE, et al: Does race influence survival for esophageal cancer patients treated on the radiation and chemotherapy arm of RTOG #85-01 Int J Radiat Oncol Biol Phys 44:1047每1052, 1999
17. Earle CC, Neumann PJ, Gelber RD, et al: Impact of referral patterns on the use of chemotherapy for lung cancer. J Clin Oncol 20:1786每1792, 2002
18. Potosky AL, Riley GF, Lubitz JD, et al: Potential for cancer related health services research using a linked Medicare-tumor registry database. Med Care 31:732每748, 1993
19. Warren JL, Klabunde CN, Schrag D, et al: Overview of the SEER-Medicare data: Content, research applications, and generalizability to the United States elderly population. Med Care 40:IV-3每18, 2002 (8 suppl)
20. Bach PB, Guadagnoli E, Schrag D, et al: Patient demographic and socioeconomic characteristics in the SEER-Medicare database applications and limitations. Med Care 40:IV-19每25, 2002 (8 suppl)
21. Klabunde CN, Warren JL, Legler JM: Assessing comorbidity using claims data: An overview. Med Care 40:IV-26每35, 2002 (8 suppl)
22. Charlson ME, Pompei P, Ales KL, et al: A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. J Chronic Dis 40:373每383, 1987
23. Deyo RA, Cherkin DC, Ciol MA: Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 45:613每619, 1992
24. Bartels H, Stein HJ, Siewert JR: Preoperative risk analysis and postoperative mortality of oesophagectomy for resectable oesophageal cancer. Br J Surg 85:840每844, 1998
25. Coebergh JW, Janssen-Heijnen ML, Post PN, et al: Serious co-morbidity among unselected cancer patients newly diagnosed in the southeastern part of the Netherlands in 1993-1996. J Clin Epidemiol 52:1131每1136, 1999
26. Harrell FE Jr, Lee KL, Mark DB: Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 15:361每387, 1996
27. Cooper GS, Virnig B, Klabunde CN, et al: Use of SEER-Medicare data for measuring cancer surgery. Med Care 40:IV-43每48, 2002 (8 suppl)
28. Finlayson EV, Birkmeyer JD: Operative mortality with elective surgery in older adults. Eff Clin Pract 4:172每177, 2001
29. Virnig BA, Warren JL, Cooper GS, et al: Studying radiation therapy using SEER-Medicare-linked data. Med Care 40:IV-49每54, 2002 (8 suppl)
30. Warren JL, Harlan LC, Fahey A, et al: Utility of the SEER-Medicare data to identify chemotherapy use. Med Care 40:IV-55每61, 2002 (8 suppl)
31. Urschel JD, Vasan H, Blewett CJ: A meta-analysis of randomized controlled trials that compared neoadjuvant chemotherapy and surgery to surgery alone for resectable esophageal cancer. Am J Surg 183:274每279, 2002
32. Bach PB, Schrag D, Brawley OW, et al: Survival of blacks and whites after a cancer diagnosis. JAMA 287:2106每2113, 2002
33. Schrijvers CT, Mackenbach JP: Cancer patient survival by socioeconomic status in seven countries: A review for six common cancer sites. J Epidemiol Community Health 48:441每446, 1994
34. Cross CK, Harris J, Recht A: Race, socioeconomic status, and breast carcinoma in the U.S: What have we learned from clinical studies. Cancer 95:1988每1999, 2002
35. Randall TC, Armstrong K: Differences in treatment and outcome between African-American and white women with endometrial cancer. J Clin Oncol 21:4200每4206, 2003
36. Dale W, Vijayakumar S, Lawlor EF, et al: Prostate cancer, race, and socioeconomic status: Inadequate adjustment for social factors in assessing racial differences. Prostate 29:271每281, 1996
37. Boulware LE, Cooper LA, Ratner LE, et al: Race and trust in the health care system. Public Health Rep 118:358每365, 2003
38. Collins TC, Clark JA, Petersen LA, et al: Racial differences in how patients perceive physician communication regarding cardiac testing. Med Care 40:I27每I34, 2002 (1 suppl)
39. Cykert S, Phifer N: Surgical decisions for early stage, non-small cell lung cancer: Which racially sensitive perceptions of cancer are likely to explain racial variation in surgery Med Decis Making 23:167每176, 2003
40. Schulman KA, Berlin JA, Harless W, et al: The effect of race and sex on physicians' recommendations for cardiac catheterization. N Engl J Med 340:618每626, 1999
41. Katz JN: Patient preferences and health disparities. JAMA 286:1506每1509, 2001(Ewout W. Steyerberg, Crai)