Performance of Screening Breast MRI across Women with Different Elevated Breast Cancer Risk Indications

Introduction

Breast cancer screening with mammography is a widely accepted practice and has been shown to reduce breast cancer mortality in the general population by 30%–40% (1,2). For specific women at high risk for breast cancer, evidence-based guidelines also recommend adjunctive screening with MRI, which demonstrates sensitivities of 71%–100% in select populations (3). The 2007 American Cancer Society guidelines and 2017 National Comprehensive Cancer Network Clinical Practice Guidelines in Oncology recommend screening MRI for the following high-risk populations: (a) women who are BRCA mutation carriers and their first degree, untested relatives; (b) women with Li-Fraumeni and other high-risk predisposition syndromes; (c) women who received radiation to the chest between the ages of 10–30 years; and (d) women with approximately 20%–25% or greater lifetime risk of breast cancer based on risk models heavily reliant on family history (eg, BRCAPRO) (3–5). The American Cancer Society consensus group concluded there was insufficient evidence to recommend for or against screening MRI for women with a personal history of breast cancer, lobular carcinoma in situ, atypical lobular hyperplasia, or atypical ductal hyperplasia despite the known higher risk of breast cancer in these populations (3,6,7).

Since the 2007 American Cancer Society guidelines were released, clinical evidence has grown supporting the performance of screening breast MRI in women with a personal history of breast cancer or high-risk lesion (8–14). Consequently, the 2017 National Comprehensive Cancer Network guidelines and the 2018 American College of Radiology recommendations added their support to consideration of annual screening breast MRI in women with a history of atypical ductal hyperplasia, atypical lobular hyperplasia, or lobular carcinoma in situ (4,5,15). The American College of Radiology also now recommends screening MRI for women with a personal history of breast cancer with dense breast tissue or a cancer diagnosis before age 50 years (15). As serious consideration is being given to expand MRI screening to women with a personal history of breast cancer or high-risk lesion, and as other groups such as the American Cancer Society consider updating their own guidelines, limitations of the literature to date should be noted. To our knowledge, no single cohort study robustly compared performance across groups, and thus prior studies do not allow for direct comparison of examination performance between women with a personal history of breast cancer and those with a history of high-risk lesion.

Given this background, the purpose of our study was to evaluate screening breast MRI performance across women with different elevated breast cancer risk indications, including women with a personal history of breast cancer or high-risk lesion.

Materials and Methods

Examination Selection

This Health Insurance Portability and Accountability Act–compliant, exploratory, retrospective case review study was approved by the institutional review board. The requirement to obtain informed consent was waived. From January 1, 2011, through December 31, 2014, 5215 screening breast MRI examinations were performed at our academic medical center. Forty-five examinations with atypical screening indications (eg, angiosarcoma history) were excluded. Thus, our final cohort included 5170 screening MRI examinations from 2637 patients.

Data Sources

Demographic characteristics, screening indication, American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) assessment, and breast pathology were extracted from our electronic medical record, cross-referencing pathologic findings in our health care system’s cancer registry of six eastern Massachusetts medical centers. These data are carefully curated to audit and ensure screening program quality. Mammographic density was recorded from the most contemporaneous mammogram within 24 months of the MRI.

MRI Technique

MRI examinations were typically performed on menstrual cycle days 7–14. Most examinations (5068 of 5170, 98%) were performed with a 1.5-T magnet; the remaining examinations were performed with a 3.0-T magnet. A dedicated breast coil was used, and the scanner and coil were from the same manufacturer (GE Healthcare, Chicago, Ill). The protocol included axial T1-weighted, T2-weighted fat-suppressed, and T1-weighted fat-suppressed sequences centered 1.5, 3, 4.5, and 7 minutes after injection of 20 mL gadopentetate dimeglumine (Magnevist; Bayer Healthcare Pharmaceuticals, Wayne, NJ). Postprocessing included T1-weighted subtracted, T1-weighted maximum intensity projection, T1-weighted fat-suppressed and subtracted sagittal reconstructed, and kinetic analysis images (CADstream; Merge Healthcare, Chicago, Ill). Most examinations (4902 of 5170, 95%) were interpreted by one of 13 breast imaging fellowship-trained, or comparably skilled, radiologists with 1–10 years of breast MRI experience. The radiologist received the ordering provider’s history and could access the patient’s electronic medical record.

Screening Indication Definitions

Screening indication was taken as the highest risk factor given by the ordering provider, according to the following hierarchy: BRCA mutation carrier or history of chest radiation (BRCA/RT group) > personal history of breast cancer (PH group) > history of high-risk lesion (HRL group) > family history of breast cancer (FH group). Patients screened because of a family history of breast cancer had none of the four preceding risk factors. In some instances, the ordering provider gave additional patient history that included other risk factors. However, as is clinical practice, the factor associated with greatest risk was considered the indication for examination. The Claus model lifetime breast cancer risk was calculated for patients with a family history of breast cancer, per American College of Radiology Imaging Network 6666 protocol methods (16,17). Non-BRCA mutation carriers either had tested negative or were untested. Patients screened because of a history of high-risk lesions had atypical ductal hyperplasia, lobular neoplasia, flat epithelial atypia, and/or other atypia.

Performance Metrics

“Initial BI-RADS” assessments were taken from the MRI reports (18). The electronic medical record was reviewed in consensus while blinded to screening indication (D.A.S., with 8 years of breast imaging experience, and K.S.B., a resident) to determine “final BI-RADS” assessments in the following instances. Initial BI-RADS category 0 examinations (n = 126) were assigned a final BI-RADS category based on BI-RADS category 0 resolution with additional imaging within 90 days. Four patients initially assigned BI-RADS category 4 were assigned final BI-RADS assessments of category 3 (n = 2) or category 2 (n = 2) based on subsequent recommendations after review of prior images or multidisciplinary discussion.

An examination was classified as true positive if cancer was diagnosed within 1 year after an initial BI-RADS category 0, 3, 4, or 5 assessment and false positive if no cancer was found. An examination was classified as true negative if no cancer was diagnosed within 1 year after an initial BI-RADS category 1 or 2 assessment and false negative if cancer was identified. Cancer was defined as invasive carcinoma or ductal carcinoma in situ (DCIS) within the breast. For patients with multiple pathologic conditions, the highest severity pathology (ie, invasive carcinoma > DCIS) was recorded.

We used the BI-RADS Atlas, fifth edition, performance metrics, including cancer detection rate (CDR), positive predictive value of positive screening result (PPV₁), positive predictive value of biopsy recommendation (PPV₂), positive predictive value of biopsies performed (PPV₃), sensitivity, and specificity (19). A single examination with an unresolved initial BI-RADS category 0 was included in PPV₁ calculation. It was excluded from PPV₂, PPV₃, and biopsy recommendation rate calculations, which used the final BI-RADS assessments.

Statistical Analysis

To compare performance across screening indications, patients imaged for BRCA and radiation therapy indications were combined (BRCA/RT group) and considered the reference standard, given clear screening recommendations for these groups (3–5). Unadjusted performance metrics were estimated for each indication group by using generalized estimating equations with a logit link and independent correlation structure to account for multiple examinations from one patient, fitting a unique model with robust sandwich standard errors within each group. For examinations performed for a family history of breast cancer, the CDR was compared with the Pearson χ² test between patients with and patients without 20% or greater Claus lifetime breast cancer risk.

For each performance metric multivariable, logistic regression generalized estimating equation models were fit and adjusted for age, screening indication, available prior comparison MRI, mammographic density (dense vs not dense), examination year, and a binary indicator for women with multiple risk factors. Examinations with missing breast density information (n = 94) were excluded. We report each model’s adjusted odds ratios (AORs), 95% confidence intervals (CIs), and P values for the screening indication coefficients. To assess multicollinearity of model covariates, the variance inflation factor was estimated for all model coefficients. This is an observational, retrospective study and tests were not powered to assess equivalency of MRI performance metrics across indications. A type I error of 5% was used for estimation of all CIs. P < .05 was considered indicative of a statistically significant difference. All data were analyzed with statistical software (R version 3.5.1; R Foundation for Statistical Computing, Vienna, Austria, 2018). The packages “rms” and “geepack” were used in this analysis.

Results

Descriptive Characteristics

The screening MRI cohort consisted of 5170 screening examinations from 2637 patients. Table 1 summarizes cohort characteristics. The average patient age was 52 years, ranging from 48 years in patients with a family history of breast cancer to 54 years in those with a personal history of breast cancer. Most examinations (93%) were performed in non-Hispanic, white women (n = 4778). The most common screening indication was personal history of breast cancer (2835, 55%), followed by family history (1314, 25%), BRCA mutation or radiation therapy (607, 12%), and high-risk lesion (414, 8%). Most patients in the HRL group had one high-risk lesion (305, 74%), with atypical ductal hyperplasia most common (228, 55%) followed by lobular carcinoma in situ (151, 37%). Most examinations (4217, 82%) were performed for one risk factor. In the PH group, 348 examinations (12%) had multiple risk factors, most commonly high-risk lesion. In the HRL group, 230 examinations (56%) were performed in women who also had a family history of breast cancer. In the BRCA/RT group, 563 examinations were performed for BRCA mutation, 42 for prior radiation therapy, and two for both BRCA mutation and prior radiation therapy. Most examinations in the entire cohort (4509, 87%) had a prior comparison breast MRI, ranging from 75% (n = 990) in the FH group to 95% (n = 2685) in the PH group. Of 2637 patients, 48% (n = 1270) had a single examination during our study period, 22% (n = 570) had two examinations, and 30% (n = 797) had three or more examinations.

A Claus lifetime breast cancer risk was estimated for 758 of 781 patients with a family history as their only risk factor. A Claus risk was not estimated for two patients missing detailed family history data, two patients older than 79 years, and 19 patients with no first- or second-degree relatives with breast cancer. The Claus model is only applicable to women up to age 79 years with a history of breast cancer in first- and/or second-degree relatives. For 84 patients, 131 relatives were conservatively assigned the lowest-risk category due to missing age at diagnosis. Among 215 patients with at least two affected relatives, 373 second-degree relatives were excluded due to missing bloodline information. Of 1314 examinations performed because of a family history, 1282 (98%) were performed in patients for whom a Claus risk could be calculated. Most of these examinations (1050 of 1282, 82%) were performed in women with a Claus lifetime breast cancer risk of less than 20%.

Performance Metrics according to Screening Indication

Initial review of the 5170 screening MRI reports resulted in 4751 (92%) negative assessments and 419 (8%) positive assessments. Subsequent review after assignment of a final BI-RADS category resulted in 4971 (96%) examinations with no biopsy recommended and 199 (4%) with biopsy recommended. Of the 199 final BI-RADS category 4 and 5 biopsy recommendations, 185 (93%) were performed. Nine biopsies were cancelled due to nonvisualization of the target lesion, and short-interval follow-up was recommended. Three examinations showed suspicious extramammary findings that were further evaluated with cross-sectional imaging, and two were followed up clinically (a superficial lesion) or with imaging rather than biopsied per surgeon’s and patient’s discretion.

Screening MRI performance metrics are shown in Table 2. The CDR was highest in the BRCA/RT group at 26 per 1000 examinations (16 of 607 examinations; 95% CI: 16, 43 per 1000 examinations) and lowest in the FH group, at 8 per 1000 examinations (10 of 1314 examinations; 95% CI: 4, 14 per 1000 examinations). There was no evidence of a difference in CDR between examinations performed in the FH group in patients with and patients without a 20% or greater Claus lifetime breast cancer risk (P = .28).

PPV₁ and PPV₂ were highest for the BRCA/RT group and lowest for the FH group. PPV₃ was highest in the BRCA/RT and PH groups at 41% (15 of 37 examinations [95% CI: 26%, 56%] and 29 of 71 examinations [95% CI: 31%, 52%], respectively) and lowest in the FH group at 14% (nine of 63 examinations [95% CI: 8%, 25%]). Representative images from a false-positive screening examination in a patient with a family history of breast cancer are shown in Figure 1. Sensitivity and specificity were highest in the PH group (88% [35 of 40 examinations; 95% CI: 73%, 95%] and 95% [2648 of 2795 examinations; 95% CI: 94%, 96%], respectively).

Figure 1:

Download as PowerPoint

Results after application of a multivariable logistic regression model to adjust for multiple possible confounders are shown in Table 3. The BRCA/RT group was used as the reference group because screening is clearly recommended for these risk factors. After adjustment, there was no difference in CDR between the PH (AOR = 0.60; 95% CI: 0.30, 1.19; P = .14) and HRL (AOR = 0.52; 95% CI: 0.20, 1.34; P = .18) groups compared with the BRCA/RT group, but there was a lower CDR in the FH group (AOR = 0.33; 95% CI: 0.13, 0.82; P = .02). All positive predictive value metrics were lower for the FH group (AOR for PPV₃ = 0.35; 95% CI: 0.12, 0.99, P = .048) compared with the BRCA/RT group, and we found no difference in metrics for the PH (AOR for PPV₃ = 0.80, 95% CI: 0.33, 1.93, P = .63) or HRL (AOR for PPV₃ = 0.54, 95% CI: 0.14, 2.08, P = .37) groups. After adjustment, there was insufficient evidence to suggest a significant difference in sensitivity or specificity between the BRCA/RT group and the groups screened for other indications. There was no evidence that the multiple risk factor variable was associated with performance in any of the models (P > .05 for all). For all model covariates, across all models, the variance inflation factor was less than 3, which suggests that there was not a significant amount of multicollinearity between variables in a single model. To keep models congruent across performance metrics, variables were not removed from our prespecified models.

Characteristics of Diagnosed Cancers

The characteristics of screening-detected cancers are shown in Table 4. Representative images of a screening-detected cancer in patients in the PH and HRL groups are shown in Figures 2 and 3, respectively. DCIS accounted for 24% (16 of 67) of all cancers diagnosed with MRI. Just over half of the cancers (51%; 34 of 67) were minimal cancers (DCIS or invasive cancer ≤1 cm). Almost all screening-detected invasive cancers (96%; 49 of 51) were stage I. Slightly more than half of the detected cancers (52%; 35 of 67) were found in the PH group. There were 13 interval cancers: three (23%) in the BRCA/RT group, five (38%) in the PH group, two (15%) in the HRL group, and three (23%) in the FH group. Five interval cancers were detected in women who presented with clinical symptoms, four were incidentally found with prophylactic mastectomies, and four were detected at screening mammography performed within 9 months of MRI.

Figure 2:

Download as PowerPoint

Figure 3:

Download as PowerPoint

Discussion

We found no evidence of a difference in screening breast MRI performance among women with a BRCA mutation or history of chest radiation, women with a personal history of breast cancer, and women with a personal history of high-risk lesion. Regression analysis did not demonstrate a difference in cancer detection rate (CDR) and positive predictive value for biopsies performed for women with a personal history of breast cancer and those with high-risk lesions compared with BRCA mutation and/or prior radiation therapy. Most cancers (52%, 35 of 67) were detected in patients with a personal history of breast cancer, who comprised more than half (55%, 2835 of 5170) of all examinations. Screening breast MRI should be strongly considered for patients with a personal history of breast cancer, as screening more patients at elevated risk will result in more cancers detected overall. Worse test performance was found in women with only family history of breast cancer, with lower CDR and positive predictive values compared with those with BRCA mutation or previous chest radiation. For patients with a family history of breast cancer, the positive predictive value of biopsy recommendation and positive predictive value for biopsies performed fell below Breast Imaging Reporting and Data System MRI screening benchmarks (19). With regression analysis, we failed to find evidence of a significant difference in sensitivity or specificity between BRCA mutation or history of chest radiation and the other screening indications. Therefore, we focused on CDR and positive predictive value to compare performance between indications. Posthoc Claus breast cancer risk assessment revealed 82% of examinations performed because of a family history of breast cancer occurred in women with less than 20% lifetime risk.

Since the release of the 2007 American Cancer Society guidelines, multiple publications have shown favorable screening MRI performance in women with a personal history of breast cancer or high-risk lesion. Lehman et al (8) and Azari-Kleinman et al (9) showed screening MRI has a similar CDR (17 vs 18 per 1000) and higher PPV₃ (19%–25% vs 14%–15%) in patients with a personal history compared with those screened for genetic risk or family history. Schwartz et al (12) and Friedlander et al (13) reported CDRs of 12–16 per 1000 and PPV₃ values of 20%–24% for women with a history of high-risk lesion. Although these studies suggest MRI screening benefits individual high-risk populations, performance comparison across multiple screening indications is limited.

Vreemann et al (20) recently addressed this limitation, reporting screening MRI performance metrics according to indication in 2463 women screened with 8818 MRI examinations. As in our study, they found a higher CDR (21 per 1000 screening examinations) in BRCA mutation carriers and a moderate CDR (12 per 1000 screening examinations) in patients with a personal history of breast cancer. The lowest CDR was found in patients with a family history of breast cancer (6 per 1000 screening examinations). They did not directly assess performance in patients with high-risk lesion. We evaluated performance for 414 studies performed for high-risk lesions. Vreemann et al noted that the PPV₃ was substantially lower in women with a family history of breast cancer (12%) compared with other indications. We similarly found the lowest PPV₃ in women with a family history of breast cancer (14%). The study by Vreemann et al spanned 10 years (January 2003 to January 2014), during which imaging protocols changed significantly. They could not assess the influence of changing protocols and experience throughout their study. Our study, with a shorter duration (4 years), had one standard protocol. We used regression analyses to adjust for multiple potential confounders, including examination year to account for evolving experience.

Our limitations include missing data from the retrospective review. Ninety-four examinations were missing information about mammographic density. Our FH group lacked consistent information about ordering providers’ risk assessment methods. We further explored variable risk within the FH group with Claus risk assessments. However, these assessments may underestimate risk given incomplete family history data for some patients. We lacked additional risk factor details to run more sophisticated risk assessment models, like BRCAPRO and Tyer-Cuzick (21). These models may have enabled more effective risk stratification of our FH group, warranting further future investigation. Our single academic institution study may not generalize to community practices.

For examinations with multiple risk factors, we did not investigate how multiple indications interact with each other with respect to cancer detection. Instead, a single hierarchy-based indication was used, which may oversimplify the relationship between multiple risk factors and MRI screening performance. Indication was defined by the highest risk factor, as described in the literature. BRCA mutation carriers have a 45%–85% breast cancer lifetime risk (22,23). Women with a history of mantle radiation as youth have a 20%–25% breast cancer risk by age 45 years (24,25). Breast cancer survivors have recurrence rates of 19.3% in the first 10 years after diagnosis (7,26). Women with lobular neoplasia have a lifetime risk of 10%–20%, with more moderate increased risk for atypical ductal hyperplasia or other high-risk lesions (6). Risk assessment of women with a family history of breast cancer revealed 11%–13% lifetime risk, depending on risk model used (27). Our use of a single indication per examination is consistent with current clinical practice and guideline recommendations, which identify individual risk factors for which screening breast MRI is recommended (3,4).

In conclusion, the results of our study support consideration of screening breast MRI for women with a personal history of breast cancer, including ductal carcinoma in situ, or high-risk lesion. Our results also suggest that better risk assessment strategies are needed for patients with a family history of breast cancer to address worse screening breast MRI performance in women screened solely for this risk factor. Radiologists can play an important role performing rigorous, standardized risk assessment before MRI screening to improve examination performance and foster greater, more appropriate, examination utilization.

Disclosures of Conflicts of Interest: D.A.S. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: received a research grant from GE Healthcare. Other relationships: disclosed no relevant relationships. K.S.B. disclosed no relevant relationships. S.F.M. disclosed no relevant relationships. G.M.R. disclosed no relevant relationships. C.E. disclosed no relevant relationships. Z.G. disclosed no relevant relationships. K.S.H. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: receives honoraria from Focal Therapeutics; is a founder of and has financial interest in CRA Health (formerly Hughes RiskApps); Dr. Hughes’s interests were reviewed and are managed by Massachusetts General Hospital and Partners Health Care in accordance with their conflict of interest policies. Other relationships: disclosed no relevant relationships. C.D.L. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: is a consultant for GE Healthcare; institution received money for grants/grants pending with GE Healthcare. Other relationships: disclosed no relevant relationships.

Author Contributions

Author contributions: Guarantors of integrity of entire study, D.A.S., K.S.B., S.F.M.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, D.A.S., K.S.B., C.D.L.; clinical studies, D.A.S., K.S.B., G.M.R., C.E., K.S.H., C.D.L.; statistical analysis, D.A.S., K.S.B., S.F.M., Z.G.; and manuscript editing, D.A.S., K.S.B., S.F.M., G.M.R., C.E., K.S.H., C.D.L.