Disparities in the Use of Screening Breast MRI in Community Practice by Race, Ethnicity and Socioeconomic Status

Abstract

Background

Uptake of breast magnetic resonance imaging (MRI) coupled with breast cancer risk assessment offers the opportunity to tailor the benefits and harms of screening strategies for women with differing cancer risks. Despite the potential benefits, there is also concern for worsening population-based health disparities.

Methods

Among 316,172 women aged 35-69 years from five Breast Cancer Surveillance Consortium registries (2007-2012), we examined 617,723 negative screening mammograms and 1,047 screening MRIs. We examined the relative risks (RRs) of MRI use by women with <20% lifetime breast cancer risk and RR in the absence of MRI use by women with ≥20% lifetime risk.

Results

Among women with <20% lifetime risk, non-Hispanic white women were 62% more likely than non-white women to receive a MRI (95% confidence interval 1.32-1.98). Of these women, those with some college or technical school were 43% more likely and those who had at least a college degree were 132% more likely to receive an MRI compared to those with a high school education or less. Among women with ≥20% lifetime risk, there was no statistically significant difference in use of screening MRI by race or ethnicity, but high-risk women with a high school education or less were less likely to receive screening MRI than women who had graduated from college (RR 0.40; 95% CI 0.25-0.63).

Conclusions

Uptake of screening breast MRI into clinical practice has the potential to worsen population-based health disparities. Policies, beyond health insurance coverage, should ensure that use of this screening modality reflects evidence-based guidelines.

Background

While mammography has been the cornerstone of breast cancer screening, advanced breast imaging technologies, such as magnetic resonance imaging (MRI) coupled with breast cancer risk assessment algorithms, provide greater opportunity to improve the balance of benefits and harms by tailoring screening strategies for subgroups of women with differing risks of developing breast cancer. Consistent with a risk-based screening approach, in 2007 the American Cancer Society (ACS) recommended that women at high risk of developing breast cancer (i.e., a lifetime risk ≥20%) should be screened annually with MRI, in addition to mammography, starting at age 30.¹ For high-risk women, breast MRI has higher sensitivity than mammography for identifying cancer, but has modest specificity that leads to higher false-positive rates and additional work-up, even among women with a higher risk of developing breast cancer.^2-4 MRI is also more expensive, can trigger claustrophobia and anxiety for some women, and requires the use of intravenous gadolinium, with associated risks. The use of screening MRI in community practice has been increasing over time, yet MRI may be used in excess of guidelines by those at average risk of developing breast cancer and “underused” by those at higher risk, in contrast to clinical guidelines.^5,6

While there are well-documented disparities in screening mammography use,^7-12 little is known about how racial, ethnic, and socioeconomic characteristics influence the use of advanced screening modalities, like MRI, that are becoming increasingly available in community practice. The uptake of screening MRI in US radiologic practice has the potential to increase population-based health disparities by race, ethnicity, and socioeconomic status (SES). In particular, lack of guideline-consistent use of MRI by high risk, disadvantaged women may be associated with delayed diagnosis while use of MRI in excess of guidelines by average risk, more advantaged women may divert health care resources from more appropriate usage and may expose women to unnecessary biopsies, anxiety from additional diagnostic work-up, and risks associated with gadolinium exposure.

The goal of this paper is to examine whether use of screening MRI varies by race, ethnicity, and SES, as assessed by education, after accounting for lifetime risk of developing breast cancer, in community practice using data from the Breast Cancer Surveillance Consortium (BCSC). Because use of MRI screening should be risk-based, differences in use by race, ethnicity, and SES after accounting for risk provide strong evidence for disparities in care.¹³ Use of MRI by women with <20% lifetime risk of breast cancer in excess of guidelines could suggest “overuse,” whereas among women with ≥20% lifetime risk, lack of guideline-consistent use of MRI screening could suggest “underuse.” As risk-based screening becomes more widely adopted in clinical practice, ensuring that each woman receives the most appropriate level of screening should promote both individual and population-level health.

Methods

Study Registries

The BCSC is a consortium of breast imaging registries in community-based settings with linkages to tumor and/or pathology registries, and a centralized Statistical Coordinating Center (SCC). This study used data from 5 registries: Carolina Mammography Registry (CMR), Group Health (GH) Cooperative (Washington state), New Hampshire Mammography Network (NHMN), San Francisco Mammography Registry (SFMR), and Vermont Breast Cancer Surveillance System (VBSS).¹⁴ The data from these five registries reflect mammography and screening MRI practice as it is performed in the community.¹⁵ Each registry and the SCC received institutional review board approval for either active consent or passive permission or a waiver of consent to enroll participants, link study data, and perform analytic studies. All registries and the SCC have received a Federal Certificate of Confidentiality and other protection for the identities of women, physicians, and facilities that are subjects of this research.

Study Population

We included women aged 35-69 years who had a negative screening mammogram (i.e., to ensure that any subsequent MRI was not being done for diagnostic purposes). The BCSC defines a negative screening mammogram as a BIRADS final assessment of: 1, 2, or 3 (with no recommendation for biopsy, FNA, or surgical consult).

At the time of breast imaging, women completed a questionnaire to ascertain age, race, ethnicity, level of education, first-degree family history of breast cancer, history of breast procedures, and other breast cancer risk factors. We excluded women with a prior history of breast cancer or ductal carcinoma in situ, or prior breast augmentation (because the sensitivity of mammography may be lower for these women who may therefore opt for MRI for other reasons beyond risk¹⁶). We also excluded women with missing data to calculate risk of breast cancer, defined further below. Methods used to identify and assess screening mammograms, screening MRIs, patient characteristics, and outcomes have been described previously.^5,15,17

Women were included if they received a mammogram in a BCSC registry from 2007 through 15 months before the most recent date of complete MRI imaging data capture available for each registry (CMR: 12/31/2012, SFMR: 12/31/2012, VBSS: 12/31/2012, NHMR: 9/30/2010, GH: 5/31/2011). To avoid artificially underestimating the use of screening MRI by including facilities where MRI was not readily available, we only included women who received a mammogram at a facility participating in the BCSC that reported at least five MRI exams during the study period.

Definition of Risk

Lifetime risk of developing breast cancer was calculated using the Gail algorithm, which includes age, race, previous breast biopsy results, presence of atypia on any prior biopsy, age at menarche, age at first live birth, and history of breast cancer in first-degree relatives.^18,19 Lifetime risk was calculated on the basis of covariate values at the time of a woman’s first screening mammogram during the study period. We defined women as being at high-risk of developing breast cancer if their lifetime risk of breast cancer was ≥20%, as guidelines suggest that these women should be screened annually by MRI in addition to mammography starting at age 30.¹ We defined women as being at “average” risk if their lifetime risk of developing breast cancer was <20%.

Covariates

In addition to lifetime risk of breast cancer, our principal covariates were race and ethnicity (categorized as non-Hispanic white, non-Hispanic black, Hispanic, and other), and SES, assessed by level of education (categorized as high school graduate or GED or less, some college or technical school, college graduate or higher). We also examined age at the time of their first mammogram during the study period (categorized as 35-39, 40-49, 50-59, 60-69 years), and year of first mammogram during the study period.

Outcome Assessment

For each woman, we included all of her screening mammograms during the study period and then looked for use of a screening MRI within 15 months of each screening mammogram (either before or after), to provide some allowance for scheduling given the annual recommendation. We conducted the primary analysis at the “woman level” by considering women to have received a screening MRI if an exam was documented within 15 months of any eligible mammogram. Women with no screening MRI exams observed within 15 months of any screening mammogram were defined as non-users. For high-risk women, for whom guidelines suggest they receive an MRI, lack of an MRI could be considered as potential “underuse.” For average risk women, we considered receipt of a screening MRI to be evidence of potential “overuse.” As a secondary analysis, to look at trends in MRI use over time, we conducted an “exam level” analysis, where the outcome was the proportion of mammograms with an MRI within 15 months of their mammogram.

Statistical Analysis

We compared lifetime risk of breast cancer by patient race, ethnicity and level of education. We then analyzed associations between use of screening MRI and these patient characteristics separately for women with lifetime risk of breast cancer <20% and ≥20%. Relative risks (RR), robust standard errors and 95% confidence intervals (CIs) were estimated with modified Poisson regression using generalized estimating equations (GEE).²⁰ We present two sets of models; the first adjusting only for registry and the second adding adjustment for age at first mammogram during the study period, race/ ethnicity, level of education and year of first mammogram during the study period. Because of small numbers of non-Hispanic black and Hispanic women (and a similar direction of association in these groups), the adjusted models compared non-Hispanic white vs. non-white (non-Hispanic black, Hispanic and “other” race/ ethnicity combined). To account for the number of mammograms observed in the BCSC data during the study period for each woman (women who received more mammograms had more opportunity to have a MRI), regression models included the number of mammograms as an offset term. For the secondary analysis examining trends in use over time, we used a modified Poisson regression model adjusted for BCSC registry applied to the exam-level data. Generalized estimating equations were used to address correlation among an individual woman's exams. All analyses were performed using SAS software version 9.3 (SAS institute), and a 2-sided p<0.05 was considered statistically significant.

Results

Description of the Sample

Our sample included 316,172 women with a screening mammogram during the study period; of these, 1,047 (0.3%) women received a screening MRI (Table 1). The majority of women in the sample were age 40-59 years at the time of the first mammogram included in the study. The majority of women (68.7%) described themselves as non-Hispanic white, 7.6% as non-Hispanic black, 6.6% as Hispanic and 17.1% as “other” race/ ethnicity (81.4% of this “other” group was Asian). While the majority of women had at least a college degree, 20.9% had a high school (or equivalent) degree or less, and 25.5% had some college or technical school beyond high school. High-risk women comprised 1.4% of the sample. Overall, only 6.2% (269 of 4,358) of women with ≥20% lifetime risk of breast cancer were observed to receive a screening MRI exam, and few (0.3% or 777 of 311, 332) women with <20% risk were observed to receive a screening MRI exam. Of the 1,047 women who received a breast MRIs in our sample, 74.3% were women with <20% risk.

Table 1.

Demographic characteristics.

Characteristic	N	%
Number of women	316,172	100
Number of women with breast MRI exam	1,047	0.33
Number of mammograms	617,723
Age at first exam during the study period
35-39	14,025	4.4
40-49	114,206	36.1
50-59	113,327	35.8
60-69	74,614	23.6
Race
Non-Hispanic white	197,430	68.7
Non-Hispanic black	21,778	7.6
Hispanic	19,098	6.6
Other	49,193	17.1
Level of Education
High school graduate, GED, or less	54,534	20.9
Some college or technical school	66,485	25.5
College graduate or higher	139,970	53.6
Lifetime risk of breast cancer (Gail)
>= 20%	4,358	1.4
< 20%	311,332	98.6
Use of breast MRI
>= 20%	269	6.2
< 20%	777	0.3

NOTE: Race/ ethnicity was missing for 28,673; level of education for 55,183, and lifetime risk of breast cancer for 482.

Lifetime Risk of Breast Cancer by Sociodemographic Characteristics

A higher proportion of non-Hispanic white women in this sample had a lifetime risk of breast cancer ≥20% than non-Hispanic black or Hispanic women (1.9%, 0.1% and 0.6% respectively). Women with at least some college were also slightly more likely to have a risk of breast cancer ≥20% compared to women who had a high school education or less (1.6% vs. 1.3%).

Use of MRI by Women with a Lifetime Risk <20%

Among those at average risk, non-Hispanic white women were 69% more likely than non-Hispanic black women (RR: 1.69; 95% CI 1.11 – 2.57) to receive screening MRI (Table 2). There were no differences in MRI use between Hispanic or other non-white women and non-Hispanic black women. After adjustment for race, level of education, age at first exam in study period, and exam year in addition to registry, non-Hispanic white women were 62% more likely than non-white women to receive a screening MRI (RR 1.62; 95% CI 1.32-1.98). Average risk women who had at least a college education were 132% more likely than women with a high school education or less (RR 2.32; 95% CI 1.76 – 3.06 after adjustment) and 43% more likely than those with some college to receive a screening MRI (for women with a high school education or less relative to those with at least a college education: RR 1.43; 95% CI 1.04 – 1.96).

Table 2.

Use of screening MRI by women with a Gail lifetime risk < 20%.

Characteristic	RR*	95% CI	RR**	95% CI
Race
Non-Hispanic white	1.69	(0.11, 2.57)	1.62	(1.32, 1.98)
Non-Hispanic black	1.00	-- --
Hispanic	0.95	(0.56, 1.60)
Other	0.88	(0.56, 1.39)
Non-white			1.00	-- --
Level of Education
High school graduate, GED, or less	1.00	-- --	1.00	-- --
Some college or technical school	1.63	(1.19, 2.23)	1.43	(1.04, 1.96)
College graduate or higher	3.04	(2.33, 3.96)	2.32	(1.76, 3.06)

^*Adjusted for registry.

^**Adjusted for race, level of education, age at first exam, exam year and registry.

Use of Screening MRI by Women with a Lifetime Risk ≥20%

Among women with a lifetime risk of breast cancer of ≥20%, there was no statistically significant difference in use of screening MRI by race or ethnicity (Table 3). High-risk women who had a high school education or less were 60% less likely to receive screening MRI than those who had graduated from college (RR 0.40; 95% CI 0.25 – 0.63 after adjustment). There was also a similar trend for women with some college but who had not graduated (RR 0.81; 95% CI 0.60 – 1.09 after adjustment).

Table 3.

Use of Screening MRI among women with a Gail risk >= 20%

Characteristic	RR*	95% CI	RR**	95% CI
Race
Non-Hispanic white	1.00	-- --	1.00	-- --
Non-Hispanic black	--	-- --	-	-- --
Hispanic	0.87	(0.36, 2.11)	-	-- --
Other	0.72	(0.39, 1.33)	0.73	(0.44, 1.21)
Level of Education
High school graduate, GED, or less	0.34	(0.22, 0.55)	0.40	(0.25, 0.63)
Some college or technical school	0.76	(0.56, 1.02)	0.81	(0.60, 1.09)
College graduate or higher	1.00	-- --	1.00	-- --

^*Adjusted for registry.

^**Adjusted for race, level of education, age at first exam, exam year, and registry.

Use of Screening MRI Over Time

As a secondary analysis, performed at the “exam level” instead of the “woman level ”to look at trends in MRI use over time, we found a significant increase in the use of screening MRI from 2008 onward (Table 4). In 2011, the RR of receiving a screening MRI relative to 2007 was 1.25 (95% CI 1.05 – 1.48).

Table 4.

Trends in MRI use over time.

Year	Adjusted Relative Risk	95% CI		P-value
2007	--
2008	0.88	0.76	1.03	0.0001
2009	0.95	0.81	1.10
2010	1.31	0.97	1.31
2011	1.25	1.05	1.48

* Adjusted for registry

Discussion

The uptake of screening breast MRI into US clinical practice has the potential to increase population-based health disparities by race, ethnicity and SES. Absence of MRI screening in high-risk, disadvantaged women may be associated with delayed diagnosis, and use of screening MRI in excess of guidelines by average risk, more resourced women may divert health care resources from more appropriate use on a population level, and expose individual women to associated harms, including false positive test results, unnecessary diagnostic work-up, and benign biopsies. Our work shows that there has been an increase in the use of screening MRI between 2007 to 2011. Our work confirms that there is broad underuse of MRI by high-risk women, but goes beyond prior work to demonstrate that among high-risk women, women with lower educational attainment are less likely to receive screening MRI than women with at least a college education.^5,6 Our work also found evidence for use of screening MRI in excess of guidelines by average risk women, as this group received the majority of screening MRIs in the study population. Our results further suggest that average risk women with at least a college education were more likely to receive a screening MRI in excess of guidelines than women with less education and that non-Hispanic white women were more likely to receive a screening MRI in excess of guidelines than non-white women. Consistent with prior work, we found that white women and women with more education were, on average, at higher risk of breast cancer, perhaps because of differences in reproductive and lifestyle risk factors.²¹ Importantly, our analyses of MRI use account for differences in risk. Further work should examine potential causes of the observed disparities which may include patient preference, provider-level variation in recommending and ordering screening MRI, patient-provider communication, or financial barriers to care.

While many studies have documented disparities in the use of mammography,^12,22,23 this work is one of the first studies to directly examine risk-based differences in the use of breast cancer screening with MRI by race, ethnicity and education. Other studies have found that high-risk women have low overall rates of adherence with the ACS guideline recommendation for screening MRI,²⁴ most perhaps because it wasn’t recommended by a health provider, women didn’t realize that they were high risk, claustrophobia, costs, or time constraints.²⁵ Our findings of disparities in MRI use by high-risk women are supported by earlier work that suggests women screened with MRI were more likely than women screened by mammography alone to be white.⁵ A study using self-reported data from the 2010 National Health Interview Survey found that 2% of women, independent of risk, report having a MRI in the prior two years, most commonly for diagnostic evaluation.²⁶ That paper did not report differences in screening MRI use by race, ethnicity or education, and differs from our results in finding that non-Hispanic black women were more likely than white women to receive MRI for any indication (i.e., screening or diagnostic).

Our results highlight the importance of educating all women about their breast cancer risk, so that high-risk women can receive recommended screening and that average risk women can avoid unnecessary test use. While clinicians frequently do not perform breast cancer risk assessment,²⁷ more advantaged women may be more likely to discuss their risk of breast cancer with a physician or perceive their cancer risk to be higher due to family history.²⁸ As we learn more about risk-based screening, we will need to develop systems-based solutions and tools to help providers and patients assess risk to promote informed decision-making. With the dissemination of advanced imaging technology for breast cancer screening, a challenge is to get the right test to the right woman. For example, income inequality in the local area where a woman lives may affect the diffusion of new technology, like MRI, and contribute to disparities in use.²⁹ Our results inform technology use at a population level, a timely goal given the emphasis on comparative effectiveness in national health care reform. If an increasing proportion of the ~39 million mammography exams performed each year are supplemented by MRI, the associated costs will be substantial. In 2013, a screening mammogram cost approximately $150 while the range for MRI was $880 - $1627.³⁰ The increased complexity of risk-based screening protocols places greater demands on health care systems to implement screening guidelines.³¹ Tools for informed decision-making should be accessible to all patients and payers should be required to pay for risk-appropriate screening.

Our study has several limitations. We used the Gail model to assess breast cancer risk. While the ACS guidelines recommend that risk assessment is done using a family history focused model (i.e., Tyrer-Cuzick, Claus models),¹ the Gail model is more widely accessible to patients and providers and may be an important first step to risk assessment in primary care. Even among the family history based models, there are important inconsistencies in the assessment of individual women.³² Some women with lifetime breast cancer risk scores <20% from the Gail model may have an indication for screening MRI that was not captured in our data (e.g., BRCA mutation, first degree relative with a BRCA mutation, prior chest radiation treatment). It is unlikely, however, that these risk factors account for a substantial portion of the women with <20% risk undergoing screening MRI. The relatively small number of non-Hispanic black and Hispanic women in our sample limits our ability to examine these groups separately in the adjusted models. Our analysis is limited to exams performed in facilities that participate in the BCSC. Women enter and leave the registry based on where they receive their breast imaging. Most women were not under observation for the entire study period. In addition, some of the facilities only participated in the BCSC for part of the study period. Some women who received a screening mammogram at a facility that participates in the BCSC may have received a screening MRI at a non-participating facility. By limiting the analysis to facilities that performed both mammography and MRI, we have minimized the potential source of underestimation of MRI use. However, we are not able to estimate the actual prevalence of MRI use. For this reason, our analysis is primarily focused on relative risk estimates that are unbiased even with incomplete capture of MRIs .³³ Although guidelines recommend annual screening, we cannot examine this as we had limited longitudinal data on individual women. Despite these limitations, our sample is one of the few that can examine risk-based disparities in the use of screening MRI in community practice using self-reported information about race, ethnicity and level of education and record-based indication for exam that allows us to include MRI exams done for screening purposes.

Optimal breast cancer screening and prevention will only occur if advanced screening technologies are translated into clinical practice in evidence-based, equitable ways. The broad dissemination of electronic health records offers opportunities for more explicit risk calculation and presentation in clinical practice. Systems-based solutions to risk assessment as well as tools for informed decision-making that are accessible to diverse populations are needed to ensure that the dissemination of advanced imaging does not contribute to health disparities or put added strain on limited health care resources.