Racial Differences in Breast Cancer Stage at Diagnosis in the Mammography Era

Abstract

Objectives. We assessed racial differences in breast cancer mortality by stage at diagnosis, since mammography became available.

Methods. We calculated adjusted odds of distant (versus local or regional) tumors for 143 249 White and 13 571 Black women aged 50 to 69 years, diagnosed with breast cancer between 1982 and 2007 and living in a Surveillance, Epidemiology, and End Results region. We compared linear trends in stage at diagnosis before and after 1998.

Results. Distant-stage cancer was diagnosed in 5.8% of White and 10.2% of Black participants. The Black–White disparity in distant tumors narrowed until 1998 (1998 adjusted difference = 0.65%), before increasing. Between 1982 and 1997, the proportion of distant tumors decreased for Blacks (adjusted odds ratio [AOR]/y = 0.973; 95% confidence interval [CI] = 0.960, 0.987) and Whites (AOR/y = 0.978; 95% CI = 0.973, 0.983), with no racial differences (P = .47). From 1998 to 2007, the odds of distant versus local or regional tumors increased for Blacks (AOR/y = 1.036; 95% CI = 1.013, 1.060) and Whites (AOR/y = 1.011; 95% CI = 1.002, 1.021); the rate of increase was greater for Blacks than Whites (P = .04).

Conclusions. In the mammography era, racial disparities remain in stage at diagnosis.

Despite a lower incidence of invasive breast cancer, Black women in the United States are more likely than are White women to die of the disease.^1,2 Since 1992, although breast cancer deaths have declined in both White and Black women, the overall disparity in mortality has increased.³ Stage at diagnosis is the strongest predictor of survival in breast cancer,^4–6 and Black women are more often diagnosed with advanced-stage disease than are White women.^7–10

Mammography is an important tool in the early detection of breast cancer.^11–13 First introduced in the United States in the early 1980s,¹⁴ mammography was initially most prevalent among White women. Racial disparities in mammography rates narrowed by the mid-1990s,^12,15 and Black women had rates equivalent to or greater than those of White women between 1996 and 2000.^10,15,16 From 2000 to 2005, mammography use declined nationally in women aged 50 to 64 years (78.6% to 71.8%), with a slightly larger decrement for White (−4.0%) than Black (−3.3%) women.¹³

The survival benefit of any screening program, including mammography, is related to its ability to detect tumors at earlier stages. Meta-analyses continue to find mortality benefit for mammography, although uncertainty remains regarding both the appropriate target population and the optimal screening interval.^11,17,18 Consistent with the expected effect of screening, an observational cohort analysis found that improvements in screening rates for both Black and White women during the 1990s contributed to diagnosis at an earlier stage in both groups.¹⁰

Nevertheless, despite generally equivalent rates of mammography for the past 15 years, the racial disparity in breast cancer mortality between Black and White women persists. Although previous meta-analyses suggested a mortality benefit for mammography, randomized controlled data regarding the efficacy of screening programs in minority populations are limited.^18,19 Because stage at diagnosis is an important predictor of survival in breast cancer, we assessed temporal changes in the distribution of stage at diagnosis between 1982 and 2007, in both Black and White women, adjusting for covariates known to affect stage at diagnosis.

METHODS

We obtained data on incident breast cancer cases from the Surveillance, Epidemiology, and End Results 9 (SEER-9) 2007 database.²⁰ Data from 1973 on were available in 9 registries: San Francisco–Oakland, California; Connecticut; metropolitan Detroit, Michigan; Hawaii; Iowa; New Mexico; Seattle (Puget Sound), Washington; Utah; and metropolitan Atlanta, Georgia. SEER registrars collected information on age at diagnosis, race (obtained from the medical registry), marital status, and stage at diagnosis. We obtained information on education and poverty in the county of residence from US Census 2000 data.²¹ County-level information on insurance came from the Small Area Health Insurance Estimates 2005 public-use database,²² and county-level estimates of body mass index (defined as weight in kilograms divided by the square of height in meters) and screening mammography from the Behavioral Risk Factor Surveillance System (BRFSS) 2006 database.²³

We also used BRFSS data to calculate mammography rates for Black and White women from 1990 to 2006 in all states with SEER-9 regions. In biennial population-based surveys, women were asked about mammography use in the past 2 years. We calculated rates of mammography for women aged 51 to 70 years because recommendations and insurance coverage for mammography at least every 2 years were generally consistent throughout the study period. Although the analytic data set age range was 50 to 69 years, the mammography data set comprised women aged 51 to 70 years, to include women aged 50 and 69 years who received mammograms within the past 2 years. Several investigators have found significant agreement (82%–94%) between self-report of mammography in the past 2 years and medical record data.^24–26

The SEER-9 2007 breast cancer database included 543 689 individuals with a first incident breast cancer diagnosed during 1973 to 2007. We excluded 3722 males and 83 687 individuals with carcinoma in situ or unstaged tumors, leaving 456 280 women with local, regional, or distant breast cancers. Staging criteria were uniform in all SEER sites. Of these women, 428 943 were non-Hispanic White or Black (derived from medical record abstraction). Because mammography became widely available in the early 1980s, we excluded women diagnosed between 1973 and 1981, yielding a sample of 353 554. Because mammography is considered most beneficial for women aged 50 to 69 years, and they have historically been the target of mammography efforts, we focused on the 157 820 women in this age group.

Variables

We assessed the proportion of breast cancers diagnosed at an advanced stage,^10,27–29 according to the SEER historic stage variable.³⁰ We focused on the proportion of distant tumors, as opposed to absolute rates of distant disease, to better assess changes in the distribution of tumors, which more accurately reflected the effect of screening programs on a population. In sensitivity analyses, we also assessed the number of distant tumors per population at risk for each race, year, and registry, with SEER estimates providing the number of persons at risk.

The main independent variable of interest was Black versus White race. Control variables were age at diagnosis (50–54, 55–59, 60–64, or 65–69 years), marital status (married vs unmarried vs unknown), SEER area, and year of diagnosis (all patient-level characteristics). We obtained estimates for the county of each patient’s residence for education (proportion within a county reporting no high school diploma, in tertiles), poverty (proportion living below the federal poverty level in 2000,²¹ in quartiles), uninsurance (proportion reporting no health insurance, in tertiles), and body weight (proportion with a body mass index > 30 kg/m², in quartiles). We also created 2006 estimates for each county of the proportion of women aged 51 to 70 years reporting having received a mammogram in the past 2 years (in quartiles).

Because the BRFSS, Small Area Health Insurance Estimates, and US Census 2000 data did not contain estimates for all counties represented in the SEER database, we included a dummy variable, missing, in the logistic model for each database. Of the 156 820 women in the analytic data set, 55 950 had missing county-level data for the BRFSS variables, 17 618 for the Small Area Health Insurance Estimates variable, and 1387 for the US Census variables. All of the individuals missing covariate data from the Small Area Health Insurance Estimates or US Census database were a subset of individuals missing covariate data from the BRFSS database.

Statistical Analyses

In unadjusted analyses, we described the proportion of White and Black women diagnosed with distant tumors by year of diagnosis. We used χ² tests to assess the association of race and other independent variables on diagnosis with advanced-stage disease.

We next compared the proportion of White and Black women diagnosed with distant tumors by year. In both unadjusted and adjusted analyses, the trend in the proportion of women with advanced-stage cancers appeared to change around 1998. To corroborate our visual inspection, we employed joinpoint regression of the analytic data set (with open-access software from SEER; http://surveillance.cancer.gov/joinpoint), which identifies statistically significant change in potential linear trends.³¹ Consistent with previous reported joinpoint analyses of invasive breast cancer incidence, we identified 1998 to 2000 as an era for significant change in overall rates of invasive breast cancer.^32,33 In light of the overlap of identified joinpoints and the nadir of Black–White disparity in proportion of advanced-stage disease in 1998, we used logistic regression to fit and compare linear trends before and after 1998 (1982–1997 vs 1998–2007) for Black and White women. We performed sensitivity analyses with 1997 and 1999 as time points for change in trend.

We assessed the association of race and year of diagnosis with distant tumors in a logistic regression model adjusted for patient age, marital status, region, and county-level information on education, poverty, obesity, and insurance status. We treated year as a continuous variable and included an indicator variable for 1982 to 1997 versus 1998 to 2007 to assess change. We assessed the association of year with distant tumors before and after 1998 (1982–1997 vs 1998–2007) for White and Black women, and we assessed the significance of the pre- versus post–1998 variable. Significance reflected the comparison of the fitted linear trend in each era (1982–1997 vs 1998–2007). We tested the interaction of year by race before and after 1998. We calculated adjusted proportions of advanced-stage disease for White and Black women by direct standardization under the regression model.³⁴ In a second model, we also adjusted for county-level estimates of mammography use to assess whether this explained any differences noted.

In sensitivity analyses, we used a binomial regression model to assess the effect of race and year on rate of distant tumors. For each registry, we calculated the number of distant tumors by the population at risk for each race and year. Independent variables were race, era (pre- vs post-1998), and the interaction. All statistical tests were 2-sided. We conducted all analyses with SAS 9.2 (SAS Institute Inc, Cary, North Carolina).

RESULTS

Characteristics of the 156 820 women diagnosed with breast cancer during 1982 to 2007 are shown in Table 1. Black women were more often diagnosed with advanced-stage disease than were White women (10.2% vs 5.8%, respectively; P < .001). Women who were unmarried and who lived in counties with higher rates of uninsurance, poverty, lack of education, and obesity were more likely than others to be diagnosed with distant- versus earlier-stage disease.

TABLE 1—

Characteristics of Study Participants and Association With Distant Tumors at Breast Cancer Diagnosis: United States; Surveillance, Epidemiology, and End Results 9 Database; 1982–2007

	Participants (n = 156 820), No. (%)	Diagnosed With Distant Tumors, %	Pa
Individual characteristics
Race			<.001
White	143 249 (91.4)	5.8
Black	13 571 (8.6)	10.2
Marital Status			<.001
Not married	52 824 (33.7)	8.1
Unknown	4759 (3.0)	7.3
Married	99 237 (63.3)	5.1
Age, y
50–54	37 188 (23.7)	5.7	<.001
55–59	38 630 (24.6)	6.3
60–64	40 087 (25.6)	6.4
65–69	40 915 (26.1)	6.4
Locationb			<.001
San Francisco–Oakland, CA	24 765 (15.8)	5.7
Connecticut	24 686 (15.74)	6.1
Detroit, MI	27 828 (17.75)	7.2
Hawaii	2237 (1.43)	5.4
Iowa	19 610 (12.5)	6.5
New Mexico	9211 (5.9)	6.4
Seattle, WA	25 294 (16.1)	5.5
Utah	8958 (5.7)	5.7
Atlanta, GA	14 231 (9.1)	6.3
County of residence characteristics
No high school diploma, %			<.001
> 16.00	53 961 (34.4)	7.1
13.14–15.99	50 186 (32.0)	6.0
< 13.14	51 286 (32.7)	5.5
Missing	1387 (0.9)	5.8
Poverty, %			<.001
> 11.20	38 775 (24.7)	7.2
8.73–11.20	39 243 (25.0)	6.3
6.70–8.72	41 061 (26.2)	5.7
< 6.70	36 354 (23.2)	5.5
Missing	1387 (0.9)	5.8
No current health insurance, %			<.001
> 16.40	34 386 (21.9)	6.6
13.30–16.40	33 163 (21.2)	5.8
11.30–13.20	42 173 (26.9)	6.2
< 11.30	29 480 (18.8)	6.3
Missing	17 618 (11.23)	5.6
Obesity (BMI > 30 kg/m2), %			.008
> 27.00	24 723 (15.8)	6.5
22.64–27.00	25 146 (16.0)	6.2
20.50–22.63	25 986 (16.6)	6.1
< 20.50	25 015 (16.0)	6.5
Missing	55 950 (35.7)	5.9

Note. BMI = body mass index.

^aBased on the χ² test; comparison group had local or regional tumors.

^bSurveillance, Epidemiology, and End Results registry region.

In unadjusted analysis, Black women with breast cancer were more likely than White women with breast cancer to be diagnosed with distant tumors in all study years, a difference that narrowed from 1982 to 1998, when the Black–White difference in the proportion of distant tumors was 1.44% (Figure 1, panel a). The Black–White disparity in advanced stage at diagnosis increased after 1998. After adjustment for covariates (Figure 1, b), the difference in the proportion of Black versus White women diagnosed with distant tumors in 1998 was nearly null (Black–White difference = 0.65%) and increased similarly in subsequent years. For Black women, the percentage of tumors diagnosed at distant locations reached a nadir in 1998 (adjusted proportion = 6.1%), whereas the nadir for White women occurred in 2002 (adjusted proportion = 4.8%). Figure 1 also displays the proportion of White and Black women aged 51 to 70 years and living in SEER-9 states who reported having a mammogram within the past 2 years. Mammography rates for both races peaked in 2000. Rates were lower for Blacks than Whites in the early 1990s, but after 1996, rates for Blacks were higher than or very similar to rates for Whites.

FIGURE 1—

Proportion of advanced-stage disease at time of breast cancer diagnoses and self-reported rates of mammography in the past 2 years among White and Black women over time (a) unadjusted analysis and (b) adjusted analysis.

Sources. Surveillance, Epidemiology, and End Results 9 database, 1982–2007,²⁰ for main data; Behavioral Risk Factor Surveillance System, 1992–2006,²³ for mammography rates.

In models assessing year at diagnosis as a continuous variable before and after 1998, the proportion of tumors diagnosed as distant decreased significantly from 1982 to 1997 for both Black and White women (adjusted odds ratio [AOR] for distant tumors/y among Black women = 0.973; 95% confidence interval [CI] = 0.960, 0.987; P < .001, and among White women = 0.978; 95% CI = 0.973, 0.983; P < .001). Comparison of the rate of decline in the proportion of distant tumors among Black and White women in this period found no statistically significant difference (P for trend comparison = 0.47). From 1998 to 2007, the proportion of breast cancers diagnosed at distant stage increased for Black women (AOR/y = 1.036; 95% CI = 1.013, 1.060; P = .002) as well as for White women, albeit more slowly (AOR/y = 1.011; 95% CI = 1.002, 1.021; P = .03). The rate of increase in the proportion of distant tumors after 1998 was significantly different for Black versus White women (P for trend comparison = 0.04), suggesting that the proportion of distant tumors increased more rapidly among Black than White women.

We also compared the rate of change in the proportion of distant tumors between periods (pre- vs post-1998) for each race. For both White and Black women, the rate of change in the proportion of distant tumors at diagnosis was significantly different in the pre- versus post-1998 era (both, P < .001; Table 2). Adjusting for county-level rates of mammography in 2006 did not change our findings. Sensitivity analyses with 1997 and 1999 as alternate time points for the era definitions showed the same statistical associations.

TABLE 2—

Adjusted Odds and Comparisons of Trend in Distant Tumors at Breast Cancer Diagnosis Across Race and Time: Surveillance, Epidemiology, and End Results 9 Database, 1982–2007

	Distant Tumor Diagnosis/y, AOR (95% CI)a	Pb
1982–1997
White women	0.978 (0.973, 0.983)	<.001
Black women	0.973 (0.960, 0.987)	<.001
White vs Black interaction before 1998		.47c
1998–2007
White women	1.011 (1.002, 1.021)	.03
Black women	1.036 (1.013, 1.060)	.002
White vs Black interaction after 1997		.04c
Rate of change, 1982–1997 vs 1998–2007
White women		<.001c
Black women		.001c

Note. AOR = adjusted odds ratio; CI = confidence interval.

^aAOR of distant tumors in a given year relative to the preceding year for the era defined.

^bBased on t test; analyses adjusted for marital status, age, location (Surveillance, Epidemiology, and End Results registry region), and county of residence education, poverty, insurance status, and obesity characteristics.

^cComparison of the slope of the fitted linear trend for each era and subgroup. We used logistic regression to define the rate of change for each defined subset (e.g., White women, 1982–1997). Values are the comparison of rates of change for each pairing described.

In sensitivity analyses assessing the effect of race and era (pre- vs post-1998) on the rate of distant tumors across registries, we similarly observed that the unadjusted rate of distant tumors for Black women increased significantly more than for White women after 1998 (data not shown). Available data did not permit adjustment for covariates.

DISCUSSION

We assessed changes in the distribution of stage at diagnosis for White and Black women since mammography became widely available. We found a sustained narrowing of the racial disparity in the proportion of advanced-stage disease until 1998, a time that corresponds approximately to the peak of mammography use. After 1998, the proportion of distant tumors increased for both races, but this increase was more rapid for Black than for White women.

An implicit assumption of contemporary mammography guidelines is that the intended effect of mammography—shifting disease distribution at diagnosis toward earlier stages—is uniform for women of all races. Our finding of a disproportionate increase in the proportion of distant tumors among Black women after 1998 suggests that this effect is not static through time. As previous research found, rising rates of mammography for both Black and White women during the 1990s were associated with a shift to earlier stage at diagnosis for both races.¹⁰ Our finding of a synchronous decrease in the proportion of distant-stage tumors at diagnosis for both Black and White women from 1982 to 1997 is consistent. What then explains the disproportionate increase in the proportion of later-stage disease in Black women since 1998?

Although differential screening use between Black and White women explains some of the difference in stage at diagnosis in the early screening era, Black women have had generally equal or higher rates of mammography since the mid-1990s.^35,36 Our analysis of a subset of BRFSS data corresponding to SEER-9 regions (Figure 1) corroborates the observed national trend.¹⁵ Previous findings that mammography test characteristics (i.e., sensitivity and specificity) are similar for Black and White women suggest that differential sensitivity of cancer detection between races is not a likely explanation.³⁷ Another possible explanation is changes in the distribution of factors known to affect stage at diagnosis: marital status, education, poverty, insurance, and obesity.^38–40 However, our findings persisted even after adjustment for these covariates, although for many of these factors, we had information only at the county level and only for a single year. Acknowledging the limitations of this adjustment, we propose that the disproportionate increase in the percentage of distant tumors for Black women since 1998 may be explained by an interaction of heterogeneous tumor biology with changes in population rates of mammography, saturation of mammographic efficacy at its current screening interval, or both.

The disproportionate increase in the proportion of distant tumors may be related to changes in overall rates of mammography in the setting of heterogeneous tumor biology. Black women are more likely to have poorly differentiated tumors²⁸ and higher rates of triple-negative breast cancer,³ both of which are associated with poorer prognosis. Differences in intrinsic tumor biology would be expected to yield different distributions of stage at diagnosis for Black versus White women. By extension, the same relative decline in mammography in a biologically heterogeneous population of tumors would be expected to disproportionately increase the percentage of distant tumors for the population of women with more aggressive disease.

It is striking to note that the near elimination of the Black–White disparity in 1998 coincided approximately with the peak of mammography use in the United States, according to National Health Interview Survey data.¹² Subsequently, rates of mammography in the United States have declined similarly for both Black and White women.^35,36 Our derivation of mammography rates in states corresponding to SEER-9 regions was likely imprecise, but generally showed a trend consistent with national data (peak in late 1990s with subsequent modest decline). This synchronous decline in mammography use for both races may have had a disproportionate impact on the stage distribution for Black women, who have biologically more aggressive disease. Black women, as a cohort, may therefore benefit more than White women from higher absolute rates of mammography use.

Alternatively, the greater increase in the proportion of distant tumors for Black women since 1998 may reflect a saturation of mammographic efficacy at the currently recommended screening interval (mammography every 1–2 years). BRFSS data suggest that rates of mammography have stabilized for both Black and White women since the late 1990s, both nationally³⁵ and within the subset of states corresponding to the SEER-9 regions (Figure 1). Others have argued that when the penetrance of a screening test reaches a plateau, incidence rates should decline as a result of a reduced pool of undiagnosed prevalent cancer cases.^41,42 This, however, assumes that the screening interval is shorter than the sojourn time, which is defined as the time from biological disease onset to clinical detection (i.e., a palpable mass). If the sojourn time is shorter than the screening interval—as is the case for biologically more aggressive tumors—the distribution of stage at diagnosis may not decline or reach a steady state. Indeed, biologically more aggressive breast tumors present at more advanced stages despite screening.^27,43 In other words, after the prevalent phase of a screening program, a fixed screening interval may only amplify intrinsic differences in biology within a population.

Others showed that shortening the screening interval from biennial to annual reduced the incidence of late-stage breast cancer in women aged 40 to 49 years, who are predicted to have sojourn times shorter than 2 years.^44,45 Specific data regarding sojourn times for breast cancer in Black women are lacking, but sojourn times are likely shorter for Black than White women because of the racial differences in tumor biology. Black women may benefit from shorter screening intervals than are currently recommended for all women.

Limitations

Information on several covariates was extracted at the county level, which may not be maximally sensitive to heterogeneity within counties. We also extracted covariate data from the general population, which may not reflect the distribution of covariates in women aged 50 to 69 years. Several covariates were extracted from single years, which may not reflect distribution through time. We did not have patient-level, longitudinal mammography data, which may explain the lack of significant change in our findings after adjustment for mammography. Perhaps more relevant would be rates of repeat mammography across race; existing data on repeat mammography in Black and White women are limited and conflicting.²

We did not adjust for tumor biology for 3 reasons. Data on variables such as hormone receptor status were not available throughout the study period. Tumor characteristics were highly associated with each other, raising concern for autocorrelation and endogeneity in an analytic model with stage at diagnosis as the dependent variable.²⁹ Finally, we considered the more clinically relevant question to be the effect of screening programs on stage at diagnosis with an endogenous biological distribution of breast cancers. That is, what is the cumulative effect of screening in a biologically heterogeneous population of tumors? Indeed, the recently growing literature assessing temporal changes in breast cancer biology during the screening era^41,42 reflects an expanding recognition of the potential clinical relevance of the relationship between screening, tumor biology, and time.

We assessed the proportion of distant-stage tumors at diagnosis as our primary outcome measure, which we believe most accurately captures the distributional effects of mammography in a population. Nevertheless, as a relational outcome, the proportion of distant tumors remains sensitive to changes in overdetection^46,47 that may not reflect an actual increasing burden of distant tumors. Our exclusion of ductal carcinoma in situ tumors limited the potential confounding contribution of overdetection to our findings.

We doubt that recent declines in invasive breast cancer secondary to reductions in hormone replacement therapy⁴⁸ affected our findings. In the United States, rates of hormone replacement therapy were highest in White women; any cessation-related declines in early-stage tumors would be expected to increase the proportion of distant tumors more for White than for Black women.^49–52 We observed the opposite trend, suggesting that our findings may, in fact, underestimate true differences. Finally, sensitivity analyses conducted at the registry level identified a disproportionate increase in the rate of distant tumors for Black women compared with White women after 1998, further suggesting that the identified increase in the proportion of distant tumors for Black women was not related to an artifactual isolated reduction in the burden of local and regional cancers.

Conclusions

Along the diagnostic and therapeutic spectrum of breast cancer, several variables may explain the survival disparity between Black and White women. In addition to stage at diagnosis, racial differences in breast cancer survival are also affected by differences in timely evaluation of abnormal mammograms⁵³ and receipt of recommended therapy.² Although our findings have potential implications regarding differences in breast cancer survival, our goal was more proximal: to assess the relative effects of screening on stage at diagnosis distribution through time. The extent to which mammography use explains the racial disparity in stage at diagnosis is variably reported, with estimates ranging from 10% to nearly the entire stage disparity.^28,53–55 The heterogeneity of these data and the lack of robust randomized clinical data assessing the mortality benefit of mammography in Black women highlight an important information deficit in the application of mammography.⁵⁶

To the extent that stage at diagnosis remains an important predictor of survival in breast cancer, our finding of an increasing proportion of distant tumors in Black women since 1998 is particularly troublesome because of the marked difference in 5-year survival rates for local versus distant tumors (98.0% vs 23.1%).⁵⁷ In light of the implications for minority patients with breast cancer, continued exploration of racial disparities in stage at diagnosis is warranted.