African-American ethnicity, socioeconomic status, and breast cancer survival

A meta–analysis of 14 studies involving over 10,000 African-American and 40,000 white American patients with carcinoma of the breast

Abstract

BACKGROUND

African-American women are at increased risk for breast cancer mortality compared with white American women, and the extent to which socioeconomic factors account for this outcome disparity is unclear.

METHODS

A MEDLINE search was conducted to identify published studies that used a Cox proportional hazards regression model to evaluate the outcome of African-American women and white American women with breast carcinoma after adjusting for socioeconomic status. A meta–analysis was performed using specialized statistical software; the random-effects method of statistical evaluation was used because of the a priori impression that the studies reviewed would be at least moderately heterogeneous in study design and patient populations.

RESULTS

The initial literature search yielded 3962 studies. Fourteen studies met all criteria for inclusion in the meta–analysis, resulting in a sample size of 10,001 African-American patients and 42,473 white American patients with breast carcinoma. There was substantial variation in the method used for defining socioeconomic status. Summary statistics revealed a significant odds ratio of 1.22 (95% confidence interval, 1.13–1.30) for the adverse effect of African-American ethnicity on breast cancer mortality. Subset meta–analyses yielded similar results, supporting the robustness of this finding.

CONCLUSIONS

This meta–analysis revealed that African-American ethnicity is an independent predictor of a worse breast cancer outcome. The pooled analysis has added strength because of the aggregate sample size and indicates that the true biologic and/or therapeutic determinants of disparities in breast cancer outcome for different ethnic groups and for different socioeconomic strata are incompletely understood. Cancer 2002;94:2844–54. © 2002 American Cancer Society.

DOI 10.1002/cncr.10575

National population-based statistics demonstrate that breast cancer mortality is greater in African-American women compared with white American women, despite an inverse pattern reported for breast cancer incidence. 1 In addition, poverty rates and the prevalence of uninsured families are greater in the African-American population.2, 3 It is commonly assumed that these socioeconomic disadvantages account for the survival disparity, because decreased access to health care is likely to result in the delayed diagnosis and treatment of breast carcinoma. However, African-American patients with breast carcinoma are also at risk for developing high-grade, estrogen receptor negative disease,4–7 and they face an increased risk of being diagnosed with early-onset breast carcinoma.1 These features cannot be explained easily by socioeconomic factors, provoking questions regarding the possible existence of ethnicity-related variation in primary breast tumor biology.

Most series in the medical literature confirm the crude association between African-American ethnicity and an increased risk of breast cancer mortality, but the statistical significance of this association is diminished in multivariate analyses that adjust for socioeconomic status. However, individual outcome studies are likely to be underpowered in their ability to adequately control for the confounding effects of ethnicity and socioeconomic status on breast cancer survival rates. The purpose of this study, therefore, was to evaluate the independent predictive strength of self-reported African-American ethnicity on breast cancer survival after controlling for stage of disease at presentation and socioeconomic status based on the aggregate power of previously published studies that addressed this question and by using meta–analytical statistical methods.

MATERIALS AND METHODS

We conducted a MEDLINE literature search for all studies of breast cancer survival in African-American patients and white American patients published in the English language from 1980 to 2001. The results of the literature search, which was based on selected key words and phrases, are shown in Table 1. We identified 3962 studies, with substantial overlap between the different categories of the search. Multiple reviewers, all with prior experience in evaluating published literature on breast cancer, screened the literature search output. References from pertinent studies and review articles also were scanned to insure a comprehensive search.

Table 1. Results of Medline Literature Search

Key phrase	No. of studies retrieved
CA: carcinoma; SES: socioeconomic status.
Breast CA and African-American race	398
Breast CA and ethnicity	843
Breast CA and race	907
Breast CA and black women	432
Breast CA and SES	1382
Total	3962

A study was eligible for inclusion into the pooled analysis if it reported breast cancer survival rates stratified by African-American ethnicity versus white American ethnicity; if odds ratios for the relative survival rates were presented; and if the study used Cox proportional hazards regression analysis to adjust survival for stage of disease at diagnosis as well as some measure of socioeconomic status. Eligible reports were limited to those written in the English language and to studies conducted in the United States because of the obvious relevance of these features to our evaluation of ethnic variation in breast cancer outcome for two subsets of the American population. This search strategy yielded 14 studies 8–21 that were suitable for meta–analysis. One of these studies15 reported separate survival analyses for younger versus older patients with breast carcinoma; therefore, findings from that study are listed twice in the meta–analysis.

The majority of reports generated by the initial literature search were excluded from the meta–analysis because they were not survival studies. In addition, two studies 22, 23 appeared to report on patient populations that overlapped with other studies that already were included.14–16 In six studies,24–29 the data provided and/or the survival analysis methods used were not compatible with the statistical requirements of the meta–analysis.

For the 14 selected studies that met all inclusion criteria, a detailed review was conducted of their patient populations and the methods employed for stratifying stage and socioeconomic status. Furthermore, because there is known variation in ethnicity-related age distributions for breast cancer, we also reviewed the methods used by the included studies for stratifying patients by age. The crude measures of association between ethnicity and survival, as well as the stage-adjusted and socioeconomic status-adjusted odds ratios for survival, were recorded.

Because of our a priori suspicions regarding heterogeneity in the patient populations and in the methods used to assess socioeconomic status, it was determined that the random-effects method of DerSimonian and Laird 30 for calculating summary statistics would be more appropriate than the fixed-effects model. A summary odds ratio of mortality risk (adjusted for disease stage and socioeconomic status) for African-American patients compared with white American patients with breast carcinoma was calculated. The meta–analysis was performed using STATA™ software (Statsoft, Inc., Tulsa, OK). The Q statistic, a test of homogeneity between studies, also was calculated for the meta–analyses performed.

Subset meta–analyses were performed for studies that used more similar methods of assessing socioeconomic status and for studies that evaluated overall outcome versus disease specific outcome. Separate pooled summaries and Q statistics were generated for each analysis. These subset analyses were used to evaluate the sensitivity and robustness of the pooled findings.

RESULTS

Characteristics and results of the selected studies are demonstrated in Tables 2 and 3. In aggregate, 10,001 African-American patients and 42,473 white American patients were evaluated. Only one study 13 was case controlled; the remaining studies were retrospective analyses of various patient populations. The majority of studies used area-based measures (census tract or census block data) to define and assign socioeconomic status. There was substantial heterogeneity between studies in the methods employed to characterize socioeconomic status and age categories for the respective patient populations, as expected. Table 2 includes the findings of the studies that provided data regarding the distributions of socioeconomic strata among the African-American patients and the white American patients. Socioeconomic disadvantages were more prevalent among the African-American patients, resulting in individual studies that had relatively small sample sizes of affluent African-American patients with breast carcinoma. Two studies used sociodemographic surrogates to designate socioeconomic status; one study12 used marital status, and another study9 used a combination of marital status, occupation, education, and nutritional status. Three studies reported on ethnicity-related variation in breast cancer survival within equal-access health care systems, including military-based providers18 and managed care-based providers.17, 19

Table 2. Characteristics of 14 Selected Studies

Study	Study type	Patient Dx yrs	Study location	No. of African-American patients	No. of white American patients	Staging system	Age stratification (yrs)	1° SES measure	SES data source	SES stratification	Reported SES variation by ethnicity
Dx: diagnosis; SES: socioeconomic status; R: retrospective; AA: African American; WA: White American; sociodem: sociodemographic; NR: not reported; LRD local, regional, distant disease: CC. case–control; NO: New Orleans; SF: San Francisco. a Patients age ≥ 50 years. b Patients age < 50 years. c SES distribution reported for the combined data set of patients at all ages.
Bassett and Krieger, 1986 8	R	1973–1983	Washington State	251	1255	TNM	< 50 and > 49	Income	Census block	Below poverty line (%)	23% AA patients below poverty line, 11% WA patients below poverty line
Coates et al., 1990 9	R	1975–1979	Atlanta	469	1491	TNM	< 50, 50–64, and > 64	Sociodem	Self report	NR	NR
Gordon et al., 1992 10	R	1974–1985	Cleveland	253	1132	TNM	< 50 and > 49	Income	Census tract	Below poverty line (%)	90% AA patients below poverty line, 41% WA patients below poverty line
Ansell et al., 1993 11	R	1973–1985	Chicago	887	265	LRD	< 50 and > 49	Income	Census tract	Annual income < $20,000 vs. ≥ $20,000	AA patients mean income, $16,000; WA patients mean income, $24,000
Neale, 1994 12	R	1973–1978	Detroit	1558	9230	LRD	< 40, 40–54, 55–69 and > 69	Sociodem.	Self report	NR	NR
Eley et al., 1994 13	CC	1985–1986	Atlanta, NO, SF	612	518	TNM	20–49, 50–64, and 65–79	Income, sociodem.	Self Report	≤ 125% Below poverty line	30% AA patients below 125% poverty line, 7% WA patients below 125% poverty line
Perkins et al., 1996 14	R	1958–1987	Houston	801	2581	LRD	< 45 and > 44	Insurance	Hospital Designated	No pay/state aid	52% AA patients receiving no pay/state aid, 25% WA patients receiving No pay/state aid
Simon and Severson, 1996 15 a	R	1988–1992	Detroit	1275	6705	LRD	Age as a continuous variable	Income	Census Tract	Low-medium-high	73% AA patients with low SES, 15% WA patients with low SESc
Simon and Severson, 1996 15 b	R	1988–1992	Detroit	605	1917	LRD	Age as a continuous variable	Income	Census Tract	Low-medium-high	—
Franzini et al., 1997 16	R	1987–1991	Houston	163	964	LRD	< 30, 30–39, 40–49, 50–59, 60–69, and > 69	Insurance	Hospital Designated	SES Grades 1 (highest) to 6 (lowest)	60% AA patients with SES 5/6, (19% WA patients with SES 5/6
Howard et al., 1998 17	R	1986–1990	Three states	89	157	TNM	< 40, 40–49, 50–64, and > 64	Insurance, sociodem.	Equal access system	NR	NR
Wojcik et al., 1998 18	R	1975–1994	Department of Defense	698	5879	TNM	< 40, 40–49, 50–65, and > 65	Insurance	Equal access system	NR	NR
Yood et al., 1999 19	R	1986–1996	Detroit	273	613	TNM	< 55 and > 54	Insurance, income	Equal access system	Annual income < $35,000 vs. ≥ $35,000	AA patients median income, $26,000; WA patients median income, $44,000
El Tamer et al., 1999 20	R	1982–1995	Brooklyn	1297	448	TNM	Age as a continuous variable	Income	Census tract	Continuous variable	NR
Roetzheim et al., 2000 21	R	1994–1997	Florida	770	9318	LRD	Age as a continuous variable	Income, insurance	Census tract, Insurance	Annual income < $15,000, $15–24,000, $25–34,000, $35–49,000, and ≥ $50,000	17% AA patients with Medicaid or no insurance, 5% WA patients with Medicaid or no insurance
Total	—	—	—	10,001	42,473	—	—	—	—	—	—

Table 3. Survival Analysis Results of Studies Included in the Meta–Analysis; Mortality Odds Ratios for African-American patients Compared with White American Patients with Breast Cancera

Study	Crude mortality odds ratio	Crude odds ratio 95%CI	Adjusted odds ratio, all-cause mortality	Adjusted odds ratio, all-cause mortality 95% CI	Adjusted odds ratio, disease specific	Adjusted odds ratio, disease specific Mortality 95% CI	Factors other than SES and stage entered into survival analysis
95% CI: 95% confidence interval; SES: socioeconomic status; NR: not reported. a Data are shown as crude, unadjusted mortality odds ratios with 95% CI and as mortality odds ratios adjusted for SES with 95% CI. b Patients age ≥ 50 years. c Patients age < 50 years.
Bassett and Krieger 8	1.3	1.03–1.74	1.10	0.83–1.46	NR	—	Age, histology, household size
Coates et al. 9	1.86	1.36–2.54	1.14	0.6–2.16	NR	—	Age, ulceration, occupation, time to treatment, body mass index/nutrition
Gordon et al. 10	1.3	1.04–1.61	1.10	0.83–1.47	1.03	0.81–1.30	Age, estrogen receptor, tumor size
Ansell et al. 11	1.2	1.00–1.41	NR	—	1.17	0.95–1.30	Age
Neale 12	NR	—	NR	—	1.14	1.04–1.24	Age, treatment, histology
Eley et al. 13	2.2	1.7–2.9	1.30	1.0–1.8	1.3	0.9–1.8	Age, treatment, comorbidity, tumor features
Perkins et al. 14	1.63	1.47–1.82	1.12	1.0–1.25	NR	—	Age, treatment
Simon and Sevenson 15 b	1.66	1.46–1.88	1.33	1.13–1.56	NR	—	Age, treatment, marital status, grade, hospital features
Simon and Severson 15 c	2.35	1.88–2.93	1.68	1.27–2.23	NR	—	Age, treatment, marital status, grade, hospital features
Franzini et al. 16	1.98	1.40–2.81	1.34	0.91–1.96	NR	—	Age, treatment, histology, protocol
Howard et al. 17	2.44	1.41–4.21	1.82	0.96–3.46	NR	—	Age, missed appointments, comorbidity, treatment delay, other
Wojcik et al. 18	1.41	1.17–1.70	1.41	1.22–1.80	NR	—	Age, treatment, marital status, family status, family history, lifestyle, grade, treatment delay
Yood et al. 19	1.6	1.1–2.2	1.00	0.7–1.5	NR	—	Age, marital status
El Tamer et al. 20	1.27	1.03–1.47	NR	—	0.97	0.78–1.21	Age, surgery. hospital
Roetzheim et al. 21	1.67	1.39–2.00	1.35	1.12–1.61	NR	—	Age, treatment, comorbidity

The odds ratios for breast cancer mortality in African-American patients compared with white American patients for the individual studies are shown in Table 3. Each study found a statistically significant survival disadvantage for the African-American patients based on the crude, unadjusted survival analyses. Table 3 also lists the adjusted mortality odds ratios for these studies. Most studies reported on all-cause mortality, and five studies reported on disease specific mortality. Two studies 10, 13 reported both all-cause and disease specific odds ratios; in each of those two studies, the all-cause mortality odds ratio was similar to the disease specific mortality odds ratio. In all cases, the association between ethnicity and outcome was reduced substantially when the various measures of socioeconomic status were taken into account.

Figures 1–5 demonstrate the meta–analysis results. In the aggregate, as well as in the subset analyses, the pooled estimates for outcome (expressed as the pooled mortality odds ratio) revealed that African-American ethnicity was a statistically significant, independent predictor of breast cancer outcome even after adjusting for socioeconomic status. In Figure 1, all studies were entered into the analysis, and the all-cause mortality odds ratio from each study was used, excluding the three studies 11, 12, 20 that reported only disease specific mortality. For this aggregate analysis, the mortality odds ratio for African-American patients compared with white American patients with breast carcinoma was 1.22 (95% confidence interval, 1.13–1.30). Overall findings were unchanged in the subset analyses, which were limited to all-cause mortality odds ratios (Fig. 2) and disease specific odds ratios (Fig. 3).

Figure 1. Meta–analysis of all studies. The mortality odds ratios (adjusted for socioeconomic status) are shown for African-American patients compared with white American patients with breast carcinoma (with all-cause mortality odds ratios entered into the analysis unless only disease specific mortality odds ratios were available): meta–analysis odds ratio, random effects model = 1.215; 95% confidence interval, 1.13–1.30; Q statistic, 21.58 with 14 degrees of freedom. The size of each box correlates with the relative sample size of each study population; horizontal lines represent 95% confidence intervals. Names at left are the lead authors for the referenced studies. Single asterisk, patients age < 50 years; double asterisks, patients age ≥ 50 years.

Figure 2. Meta–analysis of studies reporting all-cause mortality showing mortality odds ratios (adjusted for socioeconomic status) for African-American patients compared with white American patients with breast carcinoma: meta–analysis odds ratio, random effects model = 1.27; 95% confidence interval, 1.17–1.38; Q statistic, 15.16 with 11 degrees of freedom. The size of each box correlates with the relative sample size of each study population; horizontal lines represent 95% confidence intervals. The broken vertical line and diamond represent the mortality odds ratio and confidence interval for the pooled analysis, respectively. Names at left are the lead authors for the referenced studies. Single asterisk, patients age < 50 years; double asterisks, patients age ≥ 50 years.

Figure 3. Meta–analysis of studies reporting disease specific mortality. Mortality odds ratios (adjusted for socioeconomic status) are shown for African-American patients compared with white American patients with breast carcinoma: meta–analysis odds ratio, random effects model = 1.12; 95% confidence interval, 1.05–1.20; Q statistic, 3.20 with 4 degrees of freedom. The size of each box correlates with the relative sample size of each study population; horizontal lines represent 95% confidence intervals. The broken vertical line and diamond represent the mortality odds ratio and confidence interval for the pooled analysis, respectively. Names at left are the lead authors for the referenced studies.

Figure 4. Meta–analysis of studies using income/insurance measures of socioeconomic status. Mortality odds ratios (adjusted for socioeconomic status) are shown for African-American patients compared with white American patients with breast carcinoma: meta–analysis odds ratio, random effects model = 1.23; 95% confidence interval, 1.14–1.34; Q statistic, 20.0 with 12 degrees of freedom. The size of each box correlates with the relative sample size of each study population; horizontal lines represent 95% confidence intervals. The broken vertical line and diamond represent the mortality odds ratio and confidence interval for the pooled analysis, respectively. Names at left are the lead authors for the referenced studies. Single asterisk, patients age < 50 years; double asterisks, patients age ≥ 50 years.

Figure 5. Meta–analysis of studies reporting on breast cancer survival in equal-access health care systems. Mortality odds ratios (adjusted for socioeconomic status) are shown for African-American patients compared with white American patients with breast carcinoma: meta–analysis odds ratio, random effects model = 1.35; 95% confidence interval, 1.00–1.82; Q statistic, 3.95 with 2 degrees of freedom. The size of each box correlates with the relative sample size of each study population; horizontal lines represent 95% confidence intervals. The broken vertical line and diamond represent the mortality odds ratio and confidence interval for the pooled analysis, respectively. Names at left are the lead authors for the referenced studies.

A subset meta–analysis of the studies that stratified socioeconomic status based on income or insurance data only (and excluding the studies that used a sociodemographic surrogate) is shown in Figure 4; again, African-American ethnicity remained an independent predictor of worse outcome, with a pooled mortality odds ratio of 1.23 (95% confidence interval, 1.14–1.34).

Finally, the three studies that evaluated patients who were seen only through equal-access systems were evaluated in a separate meta–analysis. This subset analysis demonstrated the strongest association between African-American ethnicity and worse outcome, with a mortality odds ratio of 1.35, although the 95% confidence interval of 1.00–1.82 indicated a less stable association.

DISCUSSION

Population-based data in the United States have demonstrated a persistent association between African-American ethnicity and increased breast cancer mortality risk over the past several decades. Five-year breast carcinoma survival rates are 86% for white American patients compared with only 71% for African-American patients. Despite medical advances with mammographic screening and early detection, breast carcinoma mortality rates continue to rise in the African-American community, although they have started declining for white Americans. 1

It is well known that poverty rates are higher in the African-American community and that this subset of the population is represented disproportionately among the underinsured and noninsured; 23.6% of African Americans live below the U.S. poverty level, and 22% lack health insurance compared with 7.7% and 12% of white Americans, respectively. 2, 3 These socioeconomic disadvantages create obvious barriers to health care access, resulting in more advanced stages of disease at the time of initial diagnosis and a corresponding increased mortality rate.

However, as demonstrated in this meta-analysis and in other studies, socioeconomic disparities do not explain fully and clearly the unique patterns of breast cancer observed in African-American women. For example, one recent analysis of SEER-based data found that African Americans experience higher breast cancer mortality even after controlling for stage of disease at diagnosis, and sophisticated statistical modeling suggests a lower cure rate for African-American patients with breast carcinoma. 31 In addition, African-American patients with breast carcinoma have a younger age distribution compared with white American patients; among women age < 40 years, breast carcinoma incidence is higher for African-American women.1 Another study32 found that African-American patients with breast carcinoma in the premenopausal age range had a more advanced stage distribution compared with young white American patients and compared with older patients of either ethnic background. Furthermore, Swanson and Lin33 and Newman et al.34 demonstrated that young African-American patients with breast carcinoma tend to have the most aggressive patterns of disease and face a substantially worse prognosis compared with patients of other ethnic backgrounds at all ages. African-American patients with breast carcinoma also are at increased risk of being diagnosed with estrogen receptor negative tumors, high-grade tumors, and tumors with medullary histology.4–7

These features have motivated speculation that African-American ethnicity may be an independent predictor of breast cancer mortality because of ethnicity-related variation in tumor biology. Resolving this question has substantial public health and medical research implications. In the current era of expanding options for breast carcinoma risk reduction and for the treatment of patients with breast carcinoma, it is essential that we improve our understanding of clinical features that can stratify patients accurately into different levels of mortality risk.

The Human Genome Project has demonstrated that over 99% of the genetic code is identical across the human race 35; the remaining fraction of genetic information carrying the diversity sequences may be responsible for at least some of the ethnicity-related heterogeneity that is observed in disease expression and outcome. In the setting of breast carcinoma, we know that certain germ line founder mutations within the BRCA breast carcinoma susceptibility genes follow an ethnicity-related pattern of inheritance.36 The BRCA mutations appear to account for only a small proportion of breast malignancies that occur in the United States, but it is certainly possible that as-yet-unidentified founder mutations may account for some of the epidemiologic breast carcinoma patterns seen among African-American women. The parallel finding of a relatively younger age distribution seen for both African-American patients and native African patients with breast carcinoma supports this concept.37 Indeed, the histopathologic features discussed above that are known to be more prevalent among African-American patients with breast carcinoma are similar to the features that tend to characterize BRCA mutation-associated tumors. Gao et al.38 successfully identified some BRCA mutations that appear unique to groups of high-risk African-American families. Ongoing multicenter trials of hereditary breast carcinoma are likely to uncover additional mutations.

Still, it is likely that genetic factors only partially explain the comprehensive picture of breast carcinoma in African-American women. Other avenues for additional research include investigations of body mass index/obesity, nutritional factors, lifestyle behaviors, environmental exposures, and ethnicity-related variation in hormone metabolism. Data supporting the potential importance of these factors include findings from the National Health and Nutrition Examination Surveys demonstrating increased body mass index and obesity rates among African Americans 39, 40 and associations between obesity, serum estrogen levels, and breast carcinoma risk.41, 42 Socioeconomic status would be expected to influence nutrition and overall physical fitness.

The results of this meta–analysis should not minimize the importance of attempting to remedy the socioeconomic disadvantages that are more prevalent among African Americans. Proactive legislation that expands employment opportunities, improves housing standards, and increases health care access for the poor are necessary and essential items that should be kept high on the public policy agenda. Socioeconomic status is likely to be correlated intimately with environmental exposures, dietary practices, and lifestyle behaviors that can influence breast cancer incidence and mortality.

However, in evaluating breast cancer mortality rates among African Americans, we must be cautious about inappropriately using imprecise and poorly understood measurements of socioeconomic status as the primary explanation for the observed outcome disparity. It is incumbent upon us to identify the components of socioeconomic status that are the true determinants of tumor biology and predictors of breast cancer outcome.

This meta–analysis demonstrated that, even with sophisticated statistical tools (such as proportional hazards survival analysis), it is quite challenging to dissociate socioeconomic disadvantages from African-American ethnicity. Most studies will lack the statistical power to address breast carcinoma outcome in a meaningful sample size of relatively affluent African-American patients with breast carcinoma, and the measured association actually may be more of a comparison of breast carcinoma outcome between predominantly poor African-American patients and relatively wealthier white American patients. This sampling effect should be minimized in a pooled analysis. By enlarging the study population through meta–analysis, we found a lesser effect of socioeconomic status, and ethnicity emerged as the more significant predictor of mortality risk, although the magnitude of this ethnicity-related disparity (mortality odds ratio, 1.22) was not large. This finding was seen consistently in the subset meta–analyses performed. Most notably, a pooled analysis of the studies evaluating ethnicity-related breast cancer outcome in equal-access systems demonstrated the largest measure of association between African-American ethnicity and risk of mortality. The equal-access systems meta–analysis theoretically would have had the strongest power to eliminate confounding by ethnicity and socioeconomic status, because the patients who were treated in those systems had the most uniform access to insurance-covered health care.

The imprecision and inadequacy of existing measures of socioeconomic status are seen clearly in this meta–analysis. Only one study 13 had individualized, self-reported measures of income, education, occupation, and insurance status. The other studies generally relied on indirect assessment of income and education levels by using either census tract or census block data. A reappraisal of these area-wide measures of socioeconomic status and their implications regarding African Americans, therefore, is warranted. Sixty-five thousand census tracts cover the United States population and can include from 1000 to 8000 people; census blocks can cover between 300 and 3000 people.43 The incomes and highest education levels for residents of these areas are averaged. Investigators, therefore, can assign socioeconomic status to individual study participants by correlating their addresses and zip codes with these census area measures. Application of these census-based data has been validated as a potentially powerful means of studying socioeconomic factors.44 However, the use of area-based socioeconomic data may be less reliable for evaluating African-American families, because census-based areas may be too large to detect socioeconomic heterogeneity within African-American neighborhoods.

Some African-American families may select residential neighborhoods based on predominant ethnic composition as opposed to selecting a neighborhood with an average income that matches their own. Furthermore, when census-based areas comprised of predominantly African-American families are studied, small neighborhoods of more affluent families may be subsumed within the predominantly lower socioeconomic profile of the larger community. The effect of census-based community size on ability to detect income heterogeneity for African Americans was demonstrated by Bassett and Krieger, 8 who characterized neighborhoods by fraction of working-class, employed individuals. They found that census tract data from Seattle categorized only 2% of African Americans as living in predominantly working-class communities, whereas the use of the smaller scale census block areas characterized 14% of African Americans as living in predominantly working-class communities. Nonetheless, although census block measures may represent an improvement over census tracts in sensitivity to detect income heterogeneity, they remain imperfect. Some census blocks encompass neighborhoods of several thousand people and, thus, may camouflage substantial socioeconomic variation.

It has been postulated that the degree of income and social inequality within a community can influence health care outcomes for the population at both ends of the socioeconomic spectrum. 45–47 This model may account for the association between socioeconomic status and outcome of African-American patients with breast carcinoma reported by many individual studies. It is possible that the independent predictive value of socioeconomic status is confounded by the presence of resource inequality within communities when area-wide measures of income and education are used to stratify patients.

It is common for meta–analyses to include scoring systems for the quality of individual component studies, with randomized controlled trials carrying the greatest weight. A breast carcinoma treatment trial randomizing women on the basis of ethnicity clearly would be unethical, but we did attempt to identify randomized controlled trials that studied survival and that included data on both ethnicity and socioeconomic status. Two clinical trials cooperative groups have reported specifically on ethnicity-related outcomes, 48–50 but neither had individual, patient-reported socioeconomic data available. Roach et al.48 found that, within the Cancer and Leukemia Group B trials, survival for lymph node positive African-American and white American patients with breast carcinoma was similar after adjusting for chemotherapy dose, patient age, and estrogen receptor status. The National Surgical Adjuvant Breast Project (NSABP) reported in 199349 that African-American ethnicity was an independent predictor of higher mortality in both lymph node negative and lymph node positive subsets of patients in their B-06 trial comparing mastectomy with breast-conservation therapy. However, an updated review of NSABP trials and ethnicity50 revealed that African-American patients and white American patients who were treated on breast carcinoma treatment protocols with standardized therapy had comparable outcomes.

A major strength of prospective, randomized clinical trials is the standardization of therapy, which protects patients from potential clinician bias in allocation of treatment. Studies that demonstrate improved breast cancer outcome for African-American patients who are treated on standardized protocols motivate questions regarding the possibility that discriminatory practices in off-protocol treatments may contribute to ethnicity-related disparities in outcome. It certainly may be surmised that differences in background and life experiences between physicians and patients can result in miscommunication regarding treatment options, so that physician bias accounting for suboptimal treatment delivery is not necessarily related to outright racism. Physician bias in treatment recommendations for African-American patients has been documented in the cardiothoracic literature 51–53 and was reported recently in a subset analysis of the National Cancer Institute Black/White Cancer Survival Study.54 These biases (regardless of cause) must be eliminated.

Our meta–analysis did include one case–control study that was reported by Eley et al. 13 That project, which was sponsored by the National Cancer Institute, was a direct comparison of survival in age-matched and location-matched, African-American and white American patients with breast carcinoma. It also was the only study in the meta–analysis that had individualized, self-reported patient data regarding income, education, and employment history. Therefore, we felt that the study by Eley et al. probably was the highest quality study within the meta–analysis. The remaining studies all were retrospective reviews of African-American patients and white patients with breast carcinoma that attempted to control for socioeconomic status in a variety of ways.

Socioeconomic status as a single feature clearly makes a complex contribution to breast cancer incidence and outcome. On the one hand, breast cancer appears to be a disease of greater magnitude in industrialized Western nations, 55 and population-based breast cancer incidence rates as well as mortality rates increase in parallel with the degree of national socioeconomic wealth. Even within the United States, Wagener and Schatzkin56 have demonstrated that these population-based rates tend to be associated directly with socioeconomic status, and the more affluent communities tend to carry a greater breast cancer burden. However, those authors also found that the absolute value of this disparity has been declining, because mortality rates have declined for wealthier populations and have increased for poorer populations over the past 3 decades. Liu et al.57 demonstrated similar findings correlating increased breast cancer risk with higher socioeconomic status in a study of California communities. Conversely, as demonstrated by the studies that were included in the current meta–analysis, socioeconomic disadvantages tend to be associated with a worse outcome in patient populations that have an established breast carcinoma diagnosis.

In conclusion, individual studies have confirmed the well-known population-based finding that African-American patients with breast carcinoma have a worse outcome compared with white American patients with breast carcinoma. Proportional hazards survival analyses in these individual studies that attempt to adjust for some measure of socioeconomic status frequently (although inconsistently) demonstrate that ethnicity is eliminated as an independent predictor of outcome. Closer scrutiny of the socioeconomic stratification methods, however, reveals that these studies generally do not include significant proportions of more affluent African-American patients and, thus, probably were not powered statistically to address ethnicity and socioeconomic status. The pooled meta–analysis of these survival studies (which should minimize sample size effect) reveals that ethnicity is an independent predictor of mortality. This meta–analysis finding does not necessarily indicate that African Americans have biologically different types of breast tumors. However, it does indicate that biologic and genetic predictors of breast cancer outcome that are associated with different ethnic groups and with different socioeconomic backgrounds are measured poorly and are understood incompletely. Both ethnicity and socioeconomic-related variations in breast cancer outcome may be affected by residual confounding from other factors. The true determinants of outcome may be related to nutrition, genetics, environmental exposures, lifestyle, and/or variation in the delivery of treatment. Future research should focus on accurate identification and measurement of these factors.