The impact of socioeconomic status and race on trial participation for older women with breast cancer

Abstract

BACKGROUND

Older women, and older minorities in particular, are under represented in breast cancer trials. Although socioeconomic status (SES) is associated with both race and age, to the authors' knowledge little is known regarding the impact of SES on trial enrollment among older women with breast cancer.

METHODS

The authors performed a case–control study comparing women who were participants in National Cancer Institute cooperative group breast cancer trials (cases) with a population-based sample of breast cancer patients (controls) obtained from the linked Surveillance, Epidemiology, and End Results (SEER)-Medicare data base. The sample was restricted to women age ≥ 65 years who were living in SEER areas. Proxies for SES included the proportion of the population below poverty level (by zip code) and unemployed (by county) as well those with Medicaid insurance coverage. A multivariable logistic regression model was used to test the association of SES with trial participation after accounting for other patient and county characteristics.

RESULTS

In bivariate analysis, trial participants were significantly less likely than community cancer patients to reside in high-poverty zip codes (20.9% vs. 24.9%, respectively; P < 0.001) or to have Medicaid insurance (2.0% vs. 10.0%; P < 0.0001). After adjusting for race, age, and county, trial participation remained inversely related to residing in areas with high poverty (odds ratio [OR] vs. residents of remaining counties, 0.78; 95% confidence interval [95% CI], 0.62–0.98), high unemployment rates (OR vs. residents of residents of counties in the lowest quartile, 0.50; 95% CI, 0.35–0.71), and having Medicaid insurance (OR vs. women without Medicaid, 0.22; 95% CI, 0.13–0.37); black race was not found to be related to trial participation (OR for black vs. white, 1.0; 95% CI, 0.67–1.47).

CONCLUSIONS

Low SES was associated inversely with trial enrollment for older women with breast cancer and appeared to account for the enrollment disparities between black patients and white patients. Future efforts to enhance enrollment of elderly women in cancer research should identify specific barriers related to SES that may be amenable to intervention. Cancer 2005. © 2004 American Cancer Society.

Recent work has demonstrated age-related and race-related disparities in breast cancer trial participants, with little improvement over the past decade. Despite the fact that older patients account for approximately two-thirds of cancer patients, they have been excluded from many clinical trials.1–5 Under representation of older individuals was documented in cancer trials conducted in 1992, when 39% of male participants and 26% of female participants were elderly; in 1993–1996, when 25% of participants were elderly; in 1997–2000, when 32% of participants were elderly; and, most recently, in 2000–2002, when 33% of participants were elderly.4, 6–8 Similarly, black women with breast cancer are less likely than white women to enroll in cancer trials, and the overall proportion of trial participants who were black actually declined between 1996 and 2002.6 Furthermore, age disparities were demonstrated within each racial group, indicating that older minority patients are particularly under represented in trials.6

These disparities have generated concern, because the prevalence of cancer-related prognostic factors (such as estrogen receptor status, histologic grade, and high-risk genetic mutations) and noncancer-related factors (such as comorbid illness and functional status) vary across age and racial groups.9–14 Although many trials will not accrue sample sizes large enough to conduct adequately powered subgroup analyses, including a broad spectrum of patients in trials can facilitate the identification of hypothesis-generating findings, such as race-related differences, in response to treatments that merit further investigation.15, 16

To our knowledge, few prior analyses of trial enrollment have accounted for patient socioeconomic status (SES). Because age and race are associated strongly with SES, it is unclear whether SES is an independent barrier to trial participation. Low SES may be a particularly important barrier to trial enrollment for older patients, because the elderly have a greater incidence of frailty, greater transportation and communication needs, and less disposable income than younger patients. SES has been identified as a strong predictor of breast cancer stage at presentation, breast cancer treatment, and breast cancer mortality, even after accounting for race and other patient factors.17–22 Whereas recent work has emphasized the importance of “disaggregating” race and SES, it is unclear whether SES and race are associated independently with breast cancer trial enrollment.17 It is important to understand the degree to which SES is related to racial disparities, because interventions that aim to increase racial diversity of trial participants could differ if SES is a major confounder in the relation between race and trial enrollment.

SES is an important predictor of access to care, health status, and patient outcomes. To further our understanding of age-related and race-related disparities in breast cancer trial enrollment, it is necessary to further our understanding of SES. Women with lower SES may be less likely to enroll in trials for a number of reasons, including decreased access to care, a higher prevalence of comorbid illness, lower educational attainment, and decreased flexibility to commit the incremental time and expense required for trial participation.21, 23 A recent analysis of recruitment for an osteoporosis trial found that minority women were less likely to be eligible than white women in preliminary analysis; however, after accounting for SES, there was no difference in eligibility between black women and white women.23 Other studies of the trial recruitment process have demonstrated that patients who participate in clinical trials are more likely to be white, married, employed, and to have higher SES.24, 25

It is particularly important to explore barriers to trial enrollment for older minority patients with breast cancer, because recent work has demonstrated that age and race may be additive factors that result in even greater barriers than are faced by older nonminority individuals.6 However, to our knowledge little patient-level information is available regarding how SES and race affect trial enrollment for older patients with breast cancer. We therefore conducted a population-based study of older breast cancer patients to determine whether SES is associated with trial enrollment after adjusting for other patient factors. Because race and SES also may be associated with other geographic factors (such as population density, proximity to cancer centers and teaching hospitals, etc.), we explored how these other factors affected enrollment after accounting for race and SES.

MATERIALS AND METHODS

We performed a case–control study comparing older women who had breast cancer and had enrolled in clinical trials with older women who had breast cancer in a sample of the U.S. population. We used data from the National Cancer Institute (NCI) Clinical Trial Evaluation Program (CTEP), which is responsible for oversight and coordination of cooperative group trials, to identify women with breast cancer who were enrolled in cooperative group trials of therapeutic agents.26, 27 CTEP data are maintained with rigorous quality-control efforts, including data reviews, site visits, and possible sanctions for research centers that do not submit data. Required data elements include patient age, race/ethnicity, zip code, date of trial entry, and method of payment.27 Race/ethnicity is categorized as white, Hispanic, black, not of Hispanic origin, Native Hawaiian or other Pacific Islander, Asian, American Indian or Alaskan Native, or other. Payer is categorized as private insurance, Medicare, Medicaid, Medicare and private insurance, Medicare and Medicaid, military or veterans-sponsored, self-pay, uninsured, or other. Each trial participant is assigned to a county according to home zip code.

For our “nonparticipant” group, we obtained information about elderly cancer patients from the linked Surveillance, Epidemiology, and End Results (SEER)-Medicare data base. This population-based data base is the collaborative effort of the NCI, the SEER registries, and the Center for Medicare and Medicaid Services. The population-based SEER cancer registries account for approximately 14% of the U.S. population.28 Incident cancer patients reported to the SEER registries are cross-matched with a master file of Medicare enrollment; the resulting SEER-Medicare data base provides information on cancer patients who have fee-for-service Medicare coverage and reside in SEER areas. A previous study reported that approximately 94% of SEER cancer patients age ≥ 65 years were found in the Medicare enrollment file.29

Selection of Study Sample and Construction of Variables

Eligible trial participants included all women age ≥ 65 years with breast cancer who had enrolled in a Phase II or III cooperative group therapeutic trial during the years 1996 through 2001. We restricted the sample to patients who were living in SEER areas, because recent data suggest that elderly cancer patients who reside in SEER areas are more likely to be nonwhite, live in an urban area, and have a higher SES than elderly cancer patients in the remainder of the U.S.30 Hence, restricting our sample to patients residing in SEER areas ensured that participants and nonparticipants were drawn from the same population.

Nonparticipants, who we refer to as our population-based “community patients,” were defined as women age ≥ 65 years who were diagnosed with invasive breast carcinoma in 1996 (the most recent year in which full SEER-Medicare data were available). We also excluded patients who had no documented race (n = 203 patients) or who were Native American (n = 7 patients) due to the small number of individuals.

We used three proxies for SES. First, each patient's zip code was linked to 2000 Census data to derive the proportion below poverty level within the zip code area.31 We also used Medicaid insurance as a proxy, because this has been cited as a specific marker for low SES.32 For patients in the SEER-Medicare data base, Medicaid coverage was defined as having > 10 months of Medicaid coverage in the year prior to cancer diagnosis. For a sensitivity analysis, we repeated the analysis using a threshold of > 1 month of Medicaid coverage during the year prior to cancer diagnosis as the definition of Medicaid coverage. This did not change the results; therefore, we have reported only the results of the primary analysis. Finally, county unemployment rate was used as an SES proxy.

We obtained additional sociodemographic and geographic data from several sources. Because prior work has suggested that managed care can affect trial enrollment, we obtained county managed care penetration and index of competition (IOC) estimates for 2000 from the Interstudy County Surveyor Data Base.33–35 The IOC is defined as 1 minus the sum of the squares of each managed care organization's market share.36, 37 Values range from 0 to 1, with more competitive markets having values closer to 1. We also calculated the shortest linear distance between each patient's home and the nearest institution that enrolled patients in cooperative group breast cancer trials during the study period (“trial recruitment center”) using ZipFinder Deluxe (version 3.0).38 Additional county (presence of teaching hospital, population density) and state (proportion of population uninsured) characteristics were obtained from the Area Resource File and Interstudy County Surveyor Data Base.33, 39

Statistical Analysis

Pearson chi-square tests and t tests of independence were performed to assess the correlations between trial participation and various patient characteristics. Continuous variables were converted to categorical variables based on the distribution of the data in the full study sample. We then constructed a multiple logistic regression model with trial participation status as the dependent variable. Predictor variables included age (65–69 years, 70–74 years, 75–79 years, 80–84 years, ≥ 85 years), race (white, black, Hispanic, Asian/Pacific Islander), distance between home and nearest trial recruitment center (< 3 miles, ≥ 3 miles), percentage of the population below poverty level (< 0.13%, ≥ 0.13%), population density (< 1079, ≥ 1079), percentage uninsured population (≤ 15%, > 15%), managed care IOC (≤ 0.75, 0.76–0.82, ≥ 0.83), managed care penetration (< 0.07, 0.071–0.18, 0.19–0.39, ≥ 0.40), unemployment rate (≤ 2.90%, 2.91–3.40%, 3.40–5.59%, ≥ 5.60%), and presence of a teaching hospital in the patient's county of residence (no, yes). We used a reverse, stepwise selection process to construct the model, retaining variables with P < 0.05.

Possible interactions between race and the SES proxy covariates were explored (race*SES by zip and race*Medicaid). We also performed an analysis of the relation between race and trial participation after stratifying by SES; we restricted this analysis to black patients and white patients due to the small number of patients in the other racial groups. Variance inflation factors were used to assess for colinearity, with a value > 5 set as the threshold for further evaluation and possible exclusion of candidate variables from the model. Finally, we assessed the independent contributions of race and SES to the full model using likelihood ratio tests. The SAS (SAS Institute, Cary, NC) and Stata (Stata Corp, College Station, TX) software packages were used for our analyses.40, 41

RESULTS

Out of 37,191 patients who were enrolled in breast cancer clinical trials during the study period, 5025 patients resided in SEER areas. Among these patients, 875 women were age ≥ 65 years (17.4%), and 737 women were not missing zip code or race data. A total of 9604 patients from the SEER-Medicare data base had a first diagnosis of invasive breast carcinoma in 1996, and 7384 of those women were age ≥ 65 years and were not missing zip code or race data.

The final study sample consisted for 737 trial participants and 7384 population-based patients with breast cancer (Table 1). There was a strong relation between age and trial participation. Although women age > 80 years represented 25% of community cancer patients, they accounted for only 6.4% of trial participants. There was also a significant relation between race and trial enrollment (P = 0.012 for comparison across all groups); black patients represented 4.9% of trial participants and 7.0% of community cancer patients.

Table 1. Characteristics of Study Sample and Bivariate Analysis

Characteristic	No. of patients	No of trial participants (%)	No. of community patients (%)	P value
NA: not available; HMO: health maintenance organization.
Total	8121	737 (100.0)	7384 (100.0)	NA
Patient characteristics
Age
65–69 yrs	2193	320 (43.4)	1873 (25.4)	< 0.0001
70–74 yrs	2172	215 (29.2)	1957 (26.5)
75–79 yrs	1855	155 (21.0)	1700 (23.0)
≥ 80 yrs	1901	47 (6.4)	1854 (25.1)
Race
Caucasian	7159	639 (86.7)	6520 (88.3)	0.012
African-American	552	36 (4.9)	516 (7.0)
Asian	260	40 (5.4)	220 (3.0)
Hispanic	150	22 (3.0)	128 (1.7)
Socioeconomic status
Medicaid
No	7366	722 (98.0)	6644 (90.0)	< 0.0001
Yes	755	15 (2.0)	740 (10.0)
Census percentage population below poverty level
< 0.13%	6131	583 (79.1)	5548 (75.1)	0.017
≥ 0.13%	1990	154 (20.9)	1836 (24.9)
County unemployment rate
≤ 2.90	1782	184 (26.2)	1598 (22.1)	< 0.001
2.91–3.40	2217	181 (25.8)	2036 (28.1)
3.40–5.59	1993	205 (29.2)	1788 (24.7)
≥ 5.60	1949	131 (18.7)	1818 (25.1)
Geographic/market characteristics
Distance to research center
< 3 miles	1985	200 (27.1)	1785 (24.2)	0.074
≥ 3 miles	6136	537 (72.9)	5599 (75.8)
County population density
< 1079	3324	370 (52.8)	2984 (40.8)	< 0.0001
≥ 1079	4617	331 (47.2)	4286 (59.2)
County uninsured
< 15%	3766	401 (57.2)	3365 (46.5)	< 0.001
≥ 15%	4175	300 (42.8)	3875 (53.5)
County HMO competitiveness index
≤ 0.76	4757	472 (67.3)	4285 (59.2)	< 0.001
0.76–0.82	1582	103 (13.7)	1479 (20.4)
> 0.83	1600	126 (18.0)	1474 (20.4)
County HMO Medicare penetration
< 8%	2009	297 (42.4)	1712 (23.6)	< 0.0001
8–19%	2217	165 (23.5)	2052 (28.3)
19–39%	1433	66 (9.4)	1367 (18.9)
> 39%	2280	173 (24.7)	2107 (29.1)
Teaching hospital in county
No	3331	325 (46.4)	3006 (41.5)	0.013
Yes	4610	376 (53.6)	4234 (58.5)

Trial participants were significantly less likely to have Medicaid insurance (2.0%) than community patients (10.0%; P < 0.001) (Table 1). Similarly, trial participants were significantly less likely than patients who lived in the community to reside in high-poverty zip codes (20.9% vs. 24.9%, respectively; P = 0.017). Trial participants were significantly less likely to reside in counties with high population density or high county managed care penetration or IOC (P < 0.001 for all comparisons).

In the multivariable model, several markers of SES were associated strongly with trial enrollment (Table 2). The odds ratio (OR) for Medicaid patient enrollment was 0.22 (95% confidence interval [95% CI], 0.13–0.37) compared with non-Medicaid patient enrollment. Patients who lived in high-poverty zip codes had an OR of 0.78 (95% CI, 0.62–0.98). Finally, patients who resided in counties with the highest unemployed proportion were the least likely to enroll in trials (OR vs. residents of counties in the lowest quartile, 0.50; 95% CI, 0.35–0.71).

Table 2. Multivariate Model of Factors Associated with Trial Participation

Variable	Adjusted odds ratio	95%CI	P value
95%CI: 95% confidence interval; HMO: health maintenance organization.
Medicaid (%)
No	1.0
Yes	0.22	0.13–0.37	< 0.001
Census % below poverty
< 0.13%	1.0
≥ 0.13%	0.78	0.62–0.98	0.034
County unemployment rate
≤ 2.90%	1.0
2.91–3.40%	1.01	0.79–1.29	0.95
3.41–5.59%	1.56	1.21–2.01	0.001
≥5.60%	0.50	0.35–0.71	< 0.001
Age
65–69 yrs	1.0
70–74 yrs	0.65	0.54–0.79	< 0.001
75–79 yrs	0.54	0.43–0.66	< 0.001
≥ 80 yrs	0.14	0.10–0.19	< 0.001
Race
Caucasian	1.0
African-American	0.99	0.67–1.47	0.98
Asian	0.95	0.45–2.0	0.95
Hispanic	2.95	1.78–4.91	< 0.001
Distance to research center
≥ 3 miles	1.0
< 3 miles	1.29	1.07–1.56	0.009
Teaching hospital in county
No	1.0
Yes	1.80	1.45–2.24	< 0.001
County population density
< 1079	1.0
≥ 1079	0.60	0.47–0.77	< 0.001
County HMO Medicare penetration
< 8%	1.00
8–19%	0.36	0.28–0.46	< 0.001
19–39%	0.19	0.14–0.27	< 0.001
> 39%	0.90	0.64–1.26	0.53

There was a stepwise decrease in the odds of trial participation with increasing age, as expected. Compared with patients in the youngest age group (65–69 years), the odds of enrollment decreased from 0.65 (95% CI,0.54–0.79) in women ages 70–74 years to 0.54 (95% CI, 0.43–0.66) in women ages 75–79 years, and to 0.14 (95% CI, 0.10–0.19) in women age ≥ 80 years.

There was no relation between black race and trial participation in the multivariable model (OR, 0.99; 95% CI, 0.67–1.47). Hispanic patients were significantly more likely to enroll in trials than white patients (OR, 2.95; 95% CI, 1.78–4.91; P < 0.001). Proximity to research and teaching institutions also was related to trial enrollment. Women who resided within 3 miles of a research center were significantly more likely to enroll than individuals who lived farther away (OR, 1.29; 95% CI, 1.07–1.56). Similarly, women who resided in counties with a teaching hospital were significantly more likely to enroll in trials than women who resided in counties without teaching hospitals (OR, 1.80; 95% CI, 1.45–2.24; P < 0.001).

We constructed a series of models to explore the relation further between race and trial enrollment. After adjusting for age alone, black women were significantly less likely to be trial participants than white women (OR, 0.66; 95% CI, 0.47–0.94). When geographic factors were added to the model, the OR increased to 0.75 and was no longer significant (95% CI, 0.52–1.08). Finally, after adding SES variables to the model, the OR was 0.99 (95% CI, 0.67–1.47). The addition of the SES variables contributed a greater degree to the full model's prediction of trial enrollment status (likelihood ratio test for difference between the full model and the model without SES variables: chi-square with 2 degrees of freedom (df) = 53.2; P < 0.0001) compared with the addition of race (chi-square with 3 df = 14.9; P = 0.002).

Although neither interaction term (race and SES) reached statistical significance, we conducted stratified analyses due to the small sample sizes in some cells and the unequal distribution of SES across racial groups. White patients were significantly less likely to reside in high-poverty areas (19%) or to have Medicaid coverage (7.1%) compared with black patients (80.2% vs. 24.5%, respectively; P < 0.001 for both pairwise comparisons). The relation between SES and trial enrollment was consistent in both white patients and black patients (Table 3). For instance, the crude odds of trial enrollment for white patients in high-poverty zip codes versus white patients in nonhigh-poverty zip codes was 0.75 (95% CI, 0.60–0.93); among black patients, the OR was 0.72 (95% CI, 0.33–1.58). The adjusted OR (high poverty vs. nonhigh poverty) for white patients and black patients was 0.73 (95% CI, 0.57–0.94) and 0.63 (95% CI, 0.24–1.63), respectively. Similarly, among white patients, the unadjusted odds of trial enrollment for Medicaid patients versus non-Medicaid patients was 0.23 (95% CI, 0.13–0.42); whereas, among black patients, the OR was 0.08 (Table 3) (95% CI, 0.01–0.60).

Table 3. Breast Cancer Trial Enrollment According to Socioeconomic Status in White Women and Black Women

SES group	Total no. of patients	No. in trials (%)	Lower SES vs. higher SES
			Unadjusted		Unadjusted
			OR (95%CI)	P value	OR (95%CI)a	P value
OR: odds ratio; 95%CI: 95% confidence interval; SES: socioeconomic status. a The full model included adjustment for age, distance from research center, county unemployment rate, presence of teaching hospital in county, managed care penetration, and county population density.
White patients
Zip code
High poverty	1359	7 (7.1)	0.75 (0.60–0.93)	0.010	0.73 (0.57–0.94)	0.013
Other SES	5800	542 (9.3)	1.0		1.0
Medicaid
Yes	507	12 (2.1)	0.23 (0.13–0.42)	< 0.001	0.29 (0.17–0.54)	< 0.001
No	6652	627 (9.4)	1.0		1.0
Black patients
Zip code
High-poverty	443	27 (6.1)	0.72 (0.33–1.58)	0.41	0.63 (0.24–1.63)	0.34
Other SES	109	9 (8.3)	1.0		1.0
Medicaid
Yes	135	1 (0.7)	0.08 (0.01–0.60)	0.014	0.08 (0.01–0.77)	0.029
No	417	35 (8.4)	1.0		1.0

Overall, among patients with low SES, there were differences between black and white enrollment that were too small to reach statistical significance; whereas, among individuals with higher SES, there were no such race-related disparities (Table 4). Among residents of high-poverty zip codes, 6.1% of blacks and 7.1% of whites were trial participants (P value for difference = 0.45). Among residents of the remaining (low-poverty) zip codes, the absolute difference was about the same (8.3% of black enrollment and 9.3% of white enrollment; P = 0.70). Blacks with Medicaid insurance tended to be less likely participate in research trials than whites (0.7% and 2.1%, respectively; crude OR, 0.31; P = 0.26), although the differences were not statistically significant. After adjusting for age, gender, geographic location, and other factors in the multivariate models, the OR for trial participation among black women versus white women tended to be close to 1.0 for women who had higher SES (nonpoverty zip codes or non-Medicaid) compared with women who had lower SES (Table 4).

Table 4. Breast Cancer Trial Enrollment in Black Women and White Women According to Socioeconomic Status

SES group	Total no of patients	No. in trials (%)	Black women vs. white women
			Unadjusted		Adjusted
			OR (95%CI)	P value	OR (95%CI)a	P value
OR: odds ratio; 95%CI: 95% confidence interval; SES: socioeconomic status. a Adjusted for age, distance from research center, county unemployment rate, presence of a teaching hospital in the county, managed care penetration, and county population density. The analyses that were stratified by residence in high-poverty Zip codes also were adjusted for Medicaid insurance, whereas the models stratified by Medicaid insurance also were adjusted for residence in high-poverty Zip codes.
High-poverty Zip code
Black	443	27 (6.1)	0.84 (0.54–1.31)	0.45	0.83 (0.48–1.44)	0.51
White	97	97 (7.1)	1.0		1.0
Other SES Zip code
Black	109	9 (8.3)	0.87 (0.44–1.74)	0.70	0.97 (0.48–1.98)	0.94
White	5800	542 (9.3)	1.0		1.0
Medicaid
Black	135	1 (0.7)	0.31 (0.04–2.39)	0.26	0.56 (0.05–6.11)	0.64
White	507	12 (2.1)	1.0		1.0
No Medicaid
Black	417	35 (8.4)	0.88 (0.62–1.25)	0.48	1.00 (0.67–1.50)	0.99
White	6652	627 (9.4)	1.0		1.0

DISCUSSION

In the current study, we found that low SES was associated significantly with trial enrollment, even after adjusting for age, race, distance from the nearest cancer research center, and other county characteristics. Moreover, we found that the racial disparity (black vs. white) in enrollment was decreased after adjusting for SES, age, and other geographic characteristics. In addition, our stratified analysis demonstrated that, although blacks were far more likely to have low SES, the relation between low SES and trial enrollment was consistent among both blacks and whites. This result suggests that, whereas black women with breast cancer may be less likely than white women with breast cancer to enroll in research studies, this disparity may be mediated largely by SES. These findings build on earlier work that explored the etiology of racial disparities in trial participation. Although it has been hypothesized that black patients may be more likely to refuse to enroll in research studies, several analyses of patient enrollment have suggested that this is not the case; among individuals who are trial-eligible and who are asked to enroll, blacks are no more likely than whites to refuse.42–45 Our results extend this knowledge, suggesting that SES is a more likely mediator in trial enrollment disparities than willingness to enroll.

There are several potential mechanisms for the relation between SES and trial enrollment. SES may be a marker for trial eligibility. Low SES likely is correlated with later stage at diagnosis for many cancer types and also with a greater prevalence or severity of comorbid conditions—factors that may preclude trial eligibility. Patients with low SES may have decreased access to care and may have fewer options with regard to seeking care at centers actively engaged in clinical research. Another potential mechanism is that patients with lower SES may face greater barriers with regard to both logistics (e.g., travel, childcare responsibilities, inflexible work-hours) and communication (e.g., language, literacy), rendering it more difficult for them to adhere to strictly specified regimens that are part of trial protocols. In addition, SES may not be considered purely a social factor: Prior work has suggested that social class is related to some important biologic factors that affect breast cancer prognosis, such as hormone receptor status.9, 20 Low SES, with its associated higher parity rates, may lead to increased exposure to unopposed estrogen and, subsequently, to a greater likelihood of developing hormone receptor-negative tumors.9 These biologic differences also may affect trial eligibility or response to therapy.

Results from the current analysis also suggested that SES was a stronger barrier to trial enrollment than race. In addition, our stratified analyses suggested that SES was related to trial enrollment among both black patients and white patients, although the number of black patients was too small to reach statistical significance. We also found that, after stratifying by SES, there appeared to be smaller racial disparities in the higher SES groups than in the lower SES groups, although the small sample sizes precluded statistical significance. It is important to disaggregate race and SES when considering breast cancer care, because prior studies have suggested that they are independent predictors of disease stage at presentation for breast cancer patients.17–19 Other studies have explored the impact of race on mortality among breast cancer patients and found that racial disparities were erased after accounting for stage, residence in high-poverty areas, and having Medicaid insurance.20, 21 Even among the select group of women who enroll in clinical trials, it was found that SES was predictive of disease-free survival (DFS), whereas race was unrelated to DFS.22 Our current results are consistent with this body of work and suggest that race may be a barrier to trial enrollment, particularly for women with low SES, although future studies with larger sample sizes should explore this phenomenon.

Living near a facility that is conducting trials is an important predictor of trial enrollment: We found that women who lived within 3 miles of a research center were significantly more likely to be trial participants. This is consistent with a prior study of enrollment in clinical trials among patients with human immunodeficiency virus in which it was found that, after adjusting for other patient factors, patients who lived > 1 mile from the nearest research center were significantly less likely to enroll than patients who lived closer than 1 mile away.47 Similarly, a study of elderly patient enrollment for a heart failure trial found that the most commonly cited reason for patient refusal was living “too far away” from the study center.48

It is important to note that the relation between having Medicaid insurance and trial enrollment may be mediated through a number of factors in addition to having income that is low enough to meet Medicaid eligibility requirements. A recent study found that having Medicaid insurance was related significantly to DFS, whereas race was not.21 The authors of that study noted that nearly half of the Medicaid patients were residents of either long-term care facilities or nursing homes, which raises the issue of whether Medicaid also serves as a proxy for comorbid conditions or impaired functional status.21 Furthermore, Medicaid may be a proxy for disability among patients age ≥ 65 years, whereas many younger women with Medicaid coverage may have low SES but are less likely to be disabled. In addition, Medicaid policies in some states may not provide coverage explicitly for routine care costs for patients enrolled in clinical trials.49 Because payer reimbursement policies are a frequently cited barrier to recruiting patients into clinical studies, future work should explore the impact of Medicaid policies on trial enrollment.50

There are several important considerations that must be taken into account in interpreting the current findings. Medicaid may have low sensitivity for identifying patients with low SES, because many people with low SES may not apply for or receive Medicaid coverage.32 Hence, some patients who are classified as not having Medicaid in the SEER-Medicare data may have low SES and would be eligible for Medicaid. However, this misclassification would yield conservative estimates of the impact of Medicaid on trial enrollment by biasing toward the null. It is also important to note that the use of patient zip code to approximate SES provides an “ecologic” measure of patients' areas of residence, which can differ from those derived from census block data or from actual individual income figures.32 In addition, the SEER-Medicare data did not include information about trial participation status of the community cancer patients. Consequently, we estimate that approximately 1.7% of the SEER-Medicare patients may have been trial participants. However, this small level of misclassification would tend to bias results toward the null, leading to more conservative estimates of differences between participants and nonparticipants. Although race is coded reasonably accurately in SEER-Medicare data, Hispanic ethnicity may not be as reliable.32 The reliability of Hispanic classification in the CTEP data is unclear. Hence, it is possible that the significant relation we noted between Hispanic ethnicity and trial participation may be attributable in part to misclassification in the SEER-Medicare data base. Finally, although statistical tests for interactions and colinearity between race and each proxy measure for SES were not significant, the sample size was small enough that interactions may have been missed.

Although we found that older women with lower SES seem to face a higher hurdle before they can enroll in research studies, it is important to note that the rate of trial participation is inadequate for older patients of all races and socioeconomic groups. Recognizing the need to facilitate the enrollment of older individuals in research, several initiatives have been implemented, including a revision of the Medicare reimbursement policy in 2000 to authorize payment for trial-related clinical costs and the funding by the National Cancer Institute and the National Institute on Aging for joint ventures to initiate trials designed for older women with breast cancer.51–53 Trials that focus on the elderly are imperative, because recent work has suggested that ineligibility was a common reason for inadequate enrollment of older patients in breast cancer trials.54

The current results suggest that low SES is a barrier to research participation, and it is important to understand why this occurs. Further research will be needed to determine whether the relation between SES and trial enrollment is mediated through barriers to receiving or adhering to care, disability, education, physician and patient attitudes, or other factors. Our results suggest that overcoming these barriers may provide a greater benefit to minority patients, who are more likely to have low SES than white patients. That is, delineating the importance of SES does not obviate the finding that blacks bear the burden of the effect of low SES. Interventional studies of trial-enrollment strategies, such as the provision of transportation, childcare, and educational outreach, may help to close the SES gap in trial enrollment and to increase the ability of women at all income levels to participate in breast cancer trials.

Acknowledgements

In this study, the linked Surveillance, Epidemiology, and End Results (SEER)-Medicare data base was used. The interpretation and reporting of these data are the sole responsibility of the authors. The authors 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.