|Home | About | Journals | Submit | Contact Us | Français|
We investigated the additional contribution of mammography to screening accuracy in BRCA1/2 mutation carriers screened with MRI at different ages using individual patient data from six high-risk screening trials.
Sensitivity and specificity of MRI, mammography and the combination of these tests were compared stratified for BRCA mutation and age using generalised linear mixed models with random effect for studies. Number of screens needed (NSN) for additional mammography-only detected cancer was estimated.
In BRCA1/2 mutation carriers of all ages (BRCA1=1219 and BRCA2=732), adding mammography to MRI did not significantly increase screening sensitivity (increased by 3.9% in BRCA1 and 12.6% in BRCA2 mutation carriers, P>0.05). However, in women with BRCA2 mutation younger than 40 years, one-third of breast cancers were detected by mammography only. Number of screens needed for mammography to detect one breast cancer not detected by MRI was much higher for BRCA1 compared with BRCA2 mutation carriers at initial and repeat screening.
Additional screening sensitivity from mammography above that from MRI is limited in BRCA1 mutation carriers, whereas mammography contributes to screening sensitivity in BRCA2 mutation carriers, especially those 40 years. The evidence from our work highlights that a differential screening schedule by BRCA status is worth considering.
Women with a BRCA1 or BRCA2 mutation have limited choices to prevent mortality resulting from their 40–80% lifetime risk for breast cancer (Chen and Parmigiani, 2007). Screening with yearly MRI from age 25 years onwards, and additional mammography from age 30 years is recommended in international guidelines (Mann et al, 2008; Sardanelli et al, 2010; Zonderland et al, 2012; NICE, 2013), and is estimated to be slightly less effective than preventive mastectomy (Kurian et al, 2010; Heemskerk-Gerritsen et al, 2013). Several prospective high-risk screening studies have evaluated both MRI and mammography (Lord et al, 2007; Warner et al, 2008) as a screening strategy in high-risk women to improve screening sensitivity. In the absence of randomised controlled trials for MRI screening, these screening studies build on evidence that early detection of breast cancer may confer benefit as shown for mammography in population screening (Glasziou and Houssami, 2011). The combination of mammography and MRI screening of BRCA1/2 carriers in most guidelines, from the age of 30 or 40 years (Mann et al, 2008; Sardanelli et al, 2010; Zonderland et al, 2012; NICE, 2013), is based on the enhanced sensitivity shown through this strategy (Lord et al, 2007; Warner et al, 2008), despite arguments around limitations of mammography. These include that mammography is relatively sensitive in fatty breasts (generally in older women) but less sensitive in young women who frequently have dense breasts. In addition, screening with mammography could lead to the induction of breast cancer by X-rays at younger ages (Jansen-van der Weide et al, 2010). Proper repair of DNA double-strand breaks that are caused by low-dose X-rays is impaired at any age in both BRCA1 and BRCA2 mutations carriers (Powell and Kachnic, 2003). This makes BRCA1 and BRCA2 mutation carriers more susceptible than non-carriers, possibly also at older ages, to the cumulative effect of yearly mammograms. Given these potential disadvantages of mammography, it is important to balance the potential benefits and harms of mammography screening in BRCA1/2 mutation carriers. Hence, substantial early detection of breast cancer by mammography is needed to outweigh the potential harm of cancer induction (Jansen-van der Weide et al, 2010) in BRCA1/2 mutation carriers.
We performed an individual patient data (IPD) meta-analysis from six prospective MRI screening studies to determine if mammography screening in BRCA1/2 mutation carriers in addition to MRI improves screening accuracy, and whether this effect differs between BRCA1 and BRCA2 gene mutation carriers or by different age groups.
An IPD meta-analysis was conducted by pooling individual data from relevant prospective MRI screening studies (Phi et al, 2014). Studies were eligible if mammography and MRI breast cancer sensitivity and specificity were compared in women with a BRCA1/2 mutation. After searching PubMed, 12 studies met the eligibility requirements and were sought to contribute to the data (Phi et al, 2014). Six of these provided IPD data (Leach et al, 2005; Rijnsburger et al, 2010; Trop et al, 2010; Sardanelli et al, 2011; Passaperuma et al, 2012; Riedl et al, 2015), and were included in this meta-analysis; the reasons for non-inclusion of some studies have been reported in our earlier work (Phi et al, 2014). Included studies were assessed in terms of reporting quality, and were qualified as high quality (Phi et al, 2014). The data were assembled and cross-checked with the original publications; inclusion criteria for analyses were women with a BRCA1/2 mutation, screened annually with both mammography and MRI. Breast cancer diagnosis was confirmed by pathology and the absence of breast cancer at 1 year follow-up (Phi et al, 2014). A summary of the included studies was reported previously (http://jco.ascopubs.org/content/33/4/349/T1.large.jpg)
Primary outcome was sensitivity and specificity of mammography and MRI separately, as well as combined. Analyses were stratified for mutation type (BRCA1 or BRCA2) and age in years at screening (40 years and younger, between 41 and 50 years, over 50 years).
Sensitivity was defined as the number of breast cancers detected by a screening modality (MRI or mammography, or the combination) from the total number of breast cancers diagnosed during the study course. Specificity of a screening modality was defined as the number classified as true negative by the test from the total number of true-negative plus false-positive results.
A true positive was defined as a positive screening result (BI-RADS 0, 3, 4, 5) followed by a pathology-proven breast cancer. A false positive was defined as a positive screening result (BI-RADS 0, 3, 4, 5) not followed by a pathology-proven breast cancer within 1 year of follow-up. A true negative was defined as a negative screening result (BI-RADS 1, 2) not followed by pathology-proven breast cancer within 1 year of follow-up. A false-negative case was defined as a negative screening result (BI-RADS 1, 2) followed by a pathology-proven breast cancer within 1 year of follow-up.
Stratified by BRCA status and age group, descriptive statistics of the characteristics of the women and their breast cancer were provided. Breast cancer incidence was calculated per 1000 woman-years. The related 95% confidence intervals (CIs) were computed, assuming the incidence follows a Poisson distribution. To compare differences between groups in proportion of DCIS, invasive tumour size and grade, χ2 tests or Fisher's exact tests were applied.
To estimate the sensitivity and the specificity of the screening modalities, repeated screening results were summarised to form binomial counts for each woman. For each woman, the number of true-positive and true-negative screens per modality, and the number of total screening visits with or without breast cancer detected were counted. In this way, binomial counts per modality were calculated and analysed, taking into account that each woman was her own control. As the dependent variable was assumed to follow a binomial distribution, a generalised linear mixed model with logit link function was applied, and the binomial proportions were modelled as a function of modality and BRCA status and conducted separately for sensitivity and specificity. Studies were entered as random-effect variables and study heterogeneities were assumed to depend on modality. The analyses were conducted separately for each age group. To test the differences between the sensitivities and specificities for the three modalities, Wald tests were applied, where the hypothesis was that the difference between the two proportions under study was 0.
The number of mammographic screens that would have been needed (NSN) to detect one breast cancer that was missed by MRI was calculated, and stratified according to BRCA mutation, age group and screening round (first or subsequent round). All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA). P-values <0.05 were considered statistically significant.
The analyses were based on 1951 BRCA1/2 mutation carriers with 6085 woman-years of follow-up (Table 1). There was no significant difference in cancer risk between BRCA1 and BRCA2 mutation carriers.
Five breast cancers were diagnosed before the age of 30 in BRCA1 mutation carriers, and none in BRCA2 mutation carriers. The proportion of DCIS differed between BRCA groups in age groups older than 40 years, as shown in Table 1.
In BRCA1 mutation carriers, there were no statistically significant differences in sensitivity and specificity between mammography and MRI combined compared with MRI alone. Sensitivity of the combination was higher compared with that of MRI alone in all age groups (age 40: 86.8% (63.1–96.2) vs 77.5% (57–90), P=0.441; age 41–50: 94.1% (74.5–98.9) vs 93.1% (70.8–98.7), P=0.895; age >50: 89.3% (71.3–96.6) vs 89.1% (54.8–98.2), P=0.986). Combining mammography and MRI decreased specificity compared with MRI screening alone in all age groups (age 40 years: 81% (73.9–86.5) vs 84.3% (78.7–88.7), P=0.409; age 41–50 years: 77.2% (70.5–82.8) vs 82.9% (77.9–87), P=0.135; age >50 years: 87.4% (79.3–92.6) vs 89.9% (82.6–94.3), P=0.566). Further results are shown in Table 2.
In BRCA2 carriers, there were no significant differences in sensitivity or specificity between combined mammography and MRI and MRI alone in all age groups. Sensitivity of the combination was higher compared with that of MRI alone in all age groups (age 40 years: 87.2% (56.1–97.3) vs 52.7% (27.2–76.8), P=0.075; age 41–50 years: 91.2% (70.4–97.9) vs 86.4% (58.2–96.7), P=0.646; age >50 years: 94.1% (67.5–99.2) vs 85% (43.7–97.7), P=0.474). Combining mammography and MRI decreased specificity compared with MRI screening alone in all age groups (age 40 years: 75.3% (66.6–82.4) vs 80.2% (72.9–85.8), P=0.351; age 41–50 years: 80% (73.3–85.3) vs 86% (81.1–89.8), P=0.105; age >50 years: 88.6% (80.7–93.6) vs 91.1% (84–95.2), P=0.565). Further results are shown in Table 2.
In BRCA1 carriers overall, adding mammography to MRI screening increased sensitivity by roughly 4–92.5% (Table 2) (P=0.553). In the 40 years age group, the addition of mammography increased sensitivity by 9.3% (Table 2). Without mammography, 3 of 46 (6.5%) breast cancers, including 2 DCIS, would not have been detected (Table 3) in this subgroup. In the 41–50 years group, additional mammography increased sensitivity by only 1% (Table 2), detecting 1 DCIS (2.7%) (Table 3). Similarly, in the >50 years age group, mammography detected one additional cancer (3.4% of cancers) (Table 3).
In BRCA2 carriers, adding mammography to MRI screening increased sensitivity by 12.6–92.7% (Table 2) (P=0.154). In the 40 age group, additional mammography increased sensitivity by 34.5% (Table 2). Without mammography, 6 of 18 cancers (33.3%), including 2 DCIS, would not have been detected in this young age group (Table 3). In women aged 41–50 years, adding mammography nonsignificantly increased sensitivity by nearly 5% (Table 2) and detected 3 cancers, including 1 DCIS, which were not detected by MRI (8.1% of cancers). In the >50 years age group, screening sensitivity increased nonsignificantly by ~9% (Table 2), and mammography detected two cancers (11.8%) that were not detected by MRI, including 1 DCIS.
For the first screening round, the NSN for mammography to detect one breast cancer not detected by MRI was 527 for women with a BRCA1 mutation and 94 for women with a BRCA2 mutation for all ages (Table 4). For subsequent screening rounds, the NSN for mammography to detect an additional breast cancer for women with a BRCA1 mutation (717 screens) was roughly three times that for women with a BRCA2 mutation (231 screens).
This IPD meta-analysis has identified differences in the contribution of mammography to screening high-risk women according to age and mutation status. Adding mammography to MRI screening in BRCA1 mutation carriers leads to a very modest increase in sensitivity of 3.9% among 112 breast cancers (P=0.553), and a small decrease in specificity (by 4%, P=0.154). One invasive cancer and 2 DCIS (6.5%) of the 46 BRCA1 breast cancers detected before the age of 40 years, and only 1 DCIS and 1 invasive cancer <1cm (3%) in a total of 66 BRCA1 breast cancers would not have been detected at that screen after the age of 40 years. The percentage of early-stage (DCIS or <1cm invasive) cancers detected with both MRI and mammography screening of 36.6% (41 out of 112) would decrease by 3.6% (37 out of 112) if mammography was not be performed. Using combined MRI and mammography, 63.4% of the detected cancers were invasive and >1cm, with 0.9% of these detected by mammography only. To detect one breast cancer missed by MRI, we estimated that 527 screens for the first screening round and 717 screens for subsequent rounds with mammography would be needed.
The contribution of mammography above MRI to screening sensitivity in the 72 BRCA2 mutation carriers was 12.6% (P>0.05). Additional mammography in BRCA2 mutation carriers also decreased the specificity. Without mammography one-third of breast cancers would not have been detected in BRCA2 mutation carriers aged 40 years and younger, but this proportion was 9.3% in those older than 40 years. We estimate that the percentage of BRCA2 cancers detected at very early stage (DCIS or invasive <1cm) with combined MRI and mammography screening of 54.2% (39 out of 72) would decrease to 47.2% (34 out of 72) without mammography. Only 94 screens at first round and 231 screens at subsequent rounds of mammography screening are needed to detect a breast cancer missed by MRI. Without mammography, four advanced-stage cancers (4 out of 72 cancers, 5.6%) would have been missed in BRCA2 carriers. An advantage of mammography over MRI has been the ability to detect DCIS by visualising microcalcifications. The proportion of DCIS is larger for women with a BRCA2 mutation than for women with a BRCA1 mutation, thus differences in histology distributions in BRCA-associated breast cancers may account for our findings (Heijnsdijk et al, 2012). There might also be BRCA mutation-specific differences in tumour phenotypes that also contribute to differences in screen detection. The modest additional value of digital-only mammography to current MRI screening of BRCA1 mutation carriers was recently shown in a retrospective study (Obdeijn et al, 2014). Only 2 (2%) DCIS of 94 breast cancers were detected by mammography alone, none in women aged below 40 years and no invasive cancers. Importantly, in this retrospective study with recent data MRI screening detected 67% of the breast cancers detected as DCIS or <1 cm, considerably more than the 41–44% published for the Dutch, UK and Canadian studies of our IPD meta-analyses (Rijnsburger et al, 2010; Passaperuma et al, 2012; Evans et al, 2014) or 36.6% of this IPD meta-analysis.
It could be argued that at the time the studies forming our IPD analyses were conducted, radiologists might not have had extensive experience with breast MRI screening. Most likely, both a learning curve, as expected for any new screening modality, and improved techniques explain the relatively improved MRI sensitivities in more recent studies. A learning curve for MRI screening accuracy in high-risk women was evident for the Canadian study, in particular for DCIS detection (Warner et al, 2011). However, in a previous report in this study population (Phi et al, 2014), the sensitivity of each of MRI and mammography fluctuated over the years, and heterogeneity was evident across different studies possibly masking any potential effect of timeframe (Phi et al, 2014). A cohort study from the Netherlands showed that digital mammography had higher sensitivity compared with studies reporting film mammography (and a transition to digital) (Obdeijn et al, 2014). However, in the Italian HIBCRIT-1 Study, transition from film screen to digital mammography (resulting in screening with roughly equal mix of film screen and digital) did not increase mammography sensitivity in high-risk women (Sardanelli et al, 2011). Newer mammography technologies such as tomosynthesis (3D mammography), which have better screening sensitivity than standard mammography (Houssami et al, 2014), have not yet been compared with MRI screening of BRCA carriers. This lacking evidence in high-risk screening is worthy of research effort but would still imply increased ionising radiation from tomosynthesis (Svahn et al, 2014).
In contrast to benefits of possible earlier breast cancer detection, there are also possible harmful effects of additional mammography as outlined in the Introduction. Two-fold increase in breast cancers in BRCA1/2 mutation carriers after exposure to 4 or more radiographs, compared with non-exposure, was significant below age 30 years (HR=1.9 (95% CI: 1.2–3.0), but not at 30–39 years (Pijpe et al, 2012). Two other studies did not demonstrate tumour induction in BRCA1/2 mutation carriers by screening mammography or low-dose contralateral irradiation from breast-conserving treatment (Pierce et al, 2000; Narod et al, 2006). However, this may have been because of modest follow-up time in these studies, with consideration that latency time for radiation-induced breast cancer is 10–15 years (Travis et al, 2005; Jansen-van der Weide et al, 2010).
From two meta-analyses based on retrospective studies, the estimated cumulative risk of breast cancer by the age of 70 years vary from 57% (95% CI: 47–66%) to 65% (95% CI: 44–78) in women with a BRCA1 mutation and from 45% (95% CI: 31–56%) to 49% (95% CI: 40–57) in women with a BRCA2 mutation (Antoniou et al, 2003; Chen and Parmigiani, 2007). In this IPD meta-analysis, we combined IPD from six prospective studies, making this the largest analysis in the world of prospectively collected screening data on BRCA1/2 mutation carriers, although numbers are modest in some subgroups. We did not observe a significant difference in the risk of breast cancer between BRCA1 and BRCA2 mutation carriers given a relatively small sample of breast cancers in the IPD data set. Although data from six studies could not be included, this only resulted in ~716 women with BRCA1/2 mutations (36 breast cancers) not being included in the IPD (Kuhl et al, 2005, 2010; Lehman et al, 2005, 2007; Hagen et al, 2007; Weinstein et al, 2009). As these studies showed generally similar results for the added value of mammography to MRI, we would not expect their non-inclusion to have substantially altered our estimates.
This work differs from our recent report using the same IPD data (Phi et al, 2014) because the present analyses focus on screening outcomes by BRCA status and age group to determine mammography's contribution. Based on our findings, the additional detection from mammography in BRCA1 mutation carriers who receive MRI screening is minimal, and might not outweigh potential disadvantages (potential cancer induction by radiation, false-positive results). It may be reasonable, on the basis of this collective evidence, to consider potential omission of mammography screening in BRCA1 mutation carriers or to open discussion on its potential omission given its limited contribution. In BRCA2 mutation carriers, the contribution of mammography above MRI is more evident. Different screening recommendations for these two groups of women defined by BRCA mutation status should be considered on the basis of the evidence we report, factoring the estimated contribution of mammography and its potential harms.
No specific project funding was received. N Houssami receives a National Breast Cancer Foundation (NBCF Australia) Breast Cancer Research Leader Fellowship.
This work is published under the standard license to publish agreement. After 12 months the work will become freely available and the license terms will switch to a Creative Commons Attribution-NonCommercial-Share Alike 4.0 Unported License.
The following authors reported consulting or advisory roles: Thomas H Helbich (Siemens; Philips); Francesco Sardanelli (Bayer HealthCare including honoraria; Bracco Imaging including funding; IMS-Giotto including funding); Edwin R van den Heuvel (MSD). The other authors declare no conflict of interest.