The following are members of the Dutch MRI Screening (MRISC) Study Group: Carina C.M. Bartels, MD (Erasmus MC-Daniel den Hoed Cancer Center, Rotterdam); A. Peter E. Besnard, MD (Netherlands Cancer Institute, Amsterdam); Nicoline Hoogerbrugge, MD, PhD (University Medical Center Nijmegen); Sybren Meijer, MD, PhD (VU University Medical Center, Amsterdam); Jan C. Oosterwijk, MD, PhD (University Medical Center, University of Groningen); Caroline Seynaeve, MD, PhD (Erasmus MC-Daniel den Hoed Cancer Center, Rotterdam); Madeleine M. A. Tilanus-Linthorst, MD (Erasmus MC-Daniel den Hoed Cancer Center, Rotterdam); Rob A. E. M. Tollenaar, MD, PhD (Leiden University Medical Center, Leiden).
Within the Dutch MRI Screening (MRISC) study, a Dutch multicenter screening study for hereditary breast cancer, the authors investigated whether previously reported increased diagnostic accuracy of magnetic resonance imaging (MRI) compared with mammography would be maintained during subsequent screening rounds.
From November 1999 to October 2003, 1909 eligible women were included in the study. Screening parameters and tumor characteristics of different rounds were calculated and compared. The authors defined 3 different types of imaging screening rounds: first round in women never screened by imaging before, first round in women screened by imaging (mainly mammography) before, and subsequent rounds.
The difference in sensitivity for invasive cancers between mammography and MRI was largest in the first round of women previously screened with mammography (20.0 vs. 93.3%; P = .003), but also in subsequent rounds, there was a significant difference in favor of MRI (29.4 vs. 76.5%; P = .02). The difference in false-positive rate between mammography and MRI was also largest in the first round of women previously screened with mammography (5.5 vs. 14.0%; P<.001), and it remained significant in subsequent rounds (4.6 vs. 8.2%; P<.001). Screen-detected tumors were smaller and more often lymph node negative than symptomatic tumors in age-matched control patients, but no major differences in tumor stage were found between tumors detected at subsequent rounds compared with those in the first round.
In western countries, women with a familial or genetic predisposition for breast cancer often opt for either prophylactic surgery or intensive surveillance,1, 2 which frequently starts between the ages of 25–35 years. Screening in this group usually consists of mammography and clinical breast examination (CBE),3, 4 but the effectiveness of breast cancer screening with these modalities is questionable, especially in carriers of a BRCA1/2 germline mutation.2, 4 Recently, retrospective and prospective magnetic resonance imaging (MRI) screening studies were started to investigate whether MRI is a better screening method in this high risk group of women.5, 6 The study design7 and the first results of the Dutch multicenter prospective study for MRI screening in women with a familial or genetic predisposition (MRISC study) have been reported.8 Participants were screened for breast cancer by a 6-month complete breast examination (CBE) in addition to yearly mammography and MRI examinations, which were independently evaluated. We found that screening facilitated breast cancer diagnosis at an earlier tumor stage in these study participants than in symptomatic age-matched controls and that MRI was more sensitive, but less specific, than mammography. Overall, the sensitivity of mammography for invasive cancers was 33% and that of MRI 80%. The false-positive rate was 5% for mammography and 10% for MRI. However, it is known that screening parameters and tumor characteristics can differ among different rounds of a screening program, especially when the first round is compared with subsequent rounds, looking for prevalent and incident tumors, respectively. In the first round, the effect of length-time bias, caused by slowly growing cancers that have a longer detectable preclinical phase, is higher than in subsequent rounds.9 This may result in a higher detection rate and a higher percentage of large tumors at the first screening round in comparison with subsequent rounds.10, 11 On the other hand, the availability of previous examinations at subsequent imaging rounds can decrease the recall rate in comparison with the first round.12 It has to be noted that in the MRISC8 study and other MRI screening studies,13–16 MRI was added to the existing screening program of mammography and CBE in a significant number of the high-risk women. This means that results of the first screening round in these studies were frequently based on comparing a first MRI with a subsequent mammography. It could be that mammographically occult cancers were preferentially detected by MRI in the first screening round of women previously screened with mammography, which might exaggerate the difference in performance between MRI and mammography. Therefore, it is important to investigate whether this initially large difference between mammography and MRI is maintained in subsequent screening rounds. This analysis was performed by comparing the percentage of positive tests and the positive predictive value (PPV) of the 3 different screening modalities between each of the different rounds. Furthermore, we compared the detection rate and screening parameters of mammography and MRI from the first imaging round of women who had never been screened by mammography with those of the first imaging round of women who had been screened by mammography and with those of subsequent imaging rounds. Finally, tumor characteristics of cancers detected during first and subsequent rounds were compared.
MATERIALS AND METHODS
The design and methods of the MRISC study have been previously described.7, 8 In brief, from November 1999 until October 2003, 1952 women with a genetic breast cancer risk from 6 different centers in the Netherlands were included. All participating women gave written informed consent. Inclusion criteria were a cumulative lifetime risk of breast cancer from ≥15% because of a genetic (358 proven gene mutation carriers) or familial predisposition according to the modified tables of Claus,17 an age at entry between 25 and 70 years, and no evident symptoms suspicious for breast cancer or previous breast cancer. From the 1952 women initially included, 1909 women were eligible for this study's analyses, with a median follow-up of 2.9 years. In this group of 1909 women, 50 breast cancers were detected, including 6 cases of ductal carcinoma in situ (DCIS). The overall detection rate was 9.5 per 1000 women years, and for invasive cancers it was 8.4 per 1000 women years.
Participating women were screened twice a year by CBE in addition to mammography and MRI once a year, which were independently evaluated. The mammography and MRI were scored in a standardized way according to the Breast Imaging Reporting and Data System (BI-RADS).18, 19 We defined a BI-RADS Score 3 (“probably benign finding”), Score 0 (“need additional imaging evaluation”), Score 4 (“suspicious abnormality”), or Score 5 (“highly suggestive of malignancy”) as positive, because, in these cases, additional examination was indicated.
In general, screening rounds occurred every 6 months in view of biannual CBE. Imaging screening rounds by MRI and mammography occurred yearly. For every general screening round, the number of CBEs, mammographies, and MRIs performed were counted. For each of the 3 screening modalities, the percentage of positive tests, the breast cancer detection rate per 1000 tests. and the positive predictive value (PPV) were calculated per screening round. The PPV was calculated as the number of positive test results in women who ultimately appeared to have cancer divided by the total number of positive tests.
To compare the detection rate of invasive cancers and the screening parameters of both mammography and MRI in the first and subsequent imaging rounds, we distinguished 3 types of imaging screening rounds,
1First imaging round in women never screened before by neither mammography nor MRI (referred to as “without prior mammography”),
2First imaging round in women already screened before the start of the study by mammography (referred to as “with prior mammography”), and
For comparison of screening parameters between these 3 types of screening rounds, we used only the screening rounds that included the results of both mammography and MRI. Therefore, numbers do not necessarily total 1909. The analyses for comparison of true positive rate and false-positive rate between the first imaging round without prior mammography, the first imaging round with prior mammography, and subsequent rounds were based on only invasive cancers, because there was a difference between the sensitivity of MRI for invasive breast cancers and DCIS8 and because the numbers of DCIS were too small for a separate statistical analysis. For subsequent rounds, receiver operating characteristic (ROC) curves were constructed for mammography and MRI.
Characteristics of tumors detected during the different imaging screening rounds were described and compared. To determine whether breast cancers detected at or in the interval after a subsequent screening round were diagnosed at a more favorable stage, tumor characteristics of this subgroup of breast cancers were separately compared with those from 2 symptomatic control groups, earlier described and used in our previous article.8 The first control group was derived from all breast cancers diagnosed in 1998 in the Netherlands, where data were obtained from the National Cancer Registry. Subjects were matched for age at diagnosis with patients from the MRISC study group (in 5-year categories). The second group comprised unselected patients, who were diagnosed in Leiden and Rotterdam with primary breast cancer between 1996 and 2002 and who were participating in a prospective study (“PROSPECT STUDY”) to investigate the prevalence of gene mutations. From this series of consecutive patients, we selected nonscreened patients who were between 25 and 60 years old with a cumulative lifetime risk of ≥15% for breast cancer due to a family history, which was routinely recorded in this database.
Differences in percentage of positive tests, detection rate, and PPV between general screening rounds were tested by a chi-square test (linear-by-linear association). Differences in results of screening parameters among the 3 types of imaging screening rounds (the first round without prior mammography, the first round with prior mammography, and subsequent rounds) were tested by a Pearson chi-square test, and differences between mammography and MRI were tested by an MC Nemar chi-square test. The area under the curve (AUC) of the ROC curves (a measure for discriminative capacity of a diagnostic test) for mammography and MRI were compared and tested by a z-test. Differences in tumor characteristics among the 3 types of screening rounds were tested by a Pearson chi-square test; differences in these characteristics of screen-detected tumors during subsequent rounds versus those of the symptomatic control groups were tested by a Fisher exact test. A two-sided P value of <.05 was considered to indicate statistical significance. All statistical analyses were performed by SPSS (version 9.0, SPSS Inc, Chicago, IL).
In Table 1 the number of tests, percentage of positive tests, breast cancer detection rate, and the positive predictive value (PPV) per general screening round are presented for all 3 different screening modalities. In view of the screening scheme, nearly all women had a CBE at the first screening round, whereas the first mammography and MRI were especially performed during either the first or the second general screening round with an interval of 6 months between rounds. At the first round, the percentage of positive tests was 3.0, 6.7, and 12.9% for CBE, mammography, and MRI, respectively. For CBE, there was no significant change in percentage of positive tests, detection rate, and PPV during subsequent rounds. At the first 2 general rounds (i.e., 1 imaging round) the percentage of positive tests by MRI (12.9 and 11.3%) was at least twice as high compared with mammography (6.7% and 4.8%). However, during subsequent rounds, the percentage of positive tests by MRI decreased significantly (P<.001) from 12.9 to 6.9%, whereas for mammography, no significant differences during follow-up were observed (P = .16). With respect to the breast cancer detection rate, there was a statistically nonsignificant trend toward a decrease for mammography and a statistically significant (P = .009) decrease for MRI, especially from the fourth general round. The PPV did not show a significant decrease during subsequent rounds for either MRI (P = .30) or mammography (P = .29).
Table 1. Number of Tests and All Detected Breast Cancers (Invasive + DCIS) per General Screening Round
No. of general round*
Positive Tests n (%)
Detected BC n (per 1000 tests)
Positive Tests n (%)
Detected BC n (per 1000 tests)
Positive Tests n (%)
Detected bc n (per 1000 tests)
DCIS: ductal carcinoma in situ; CBE: clinical breast examination, MRI: magnetice resonance imaging; BC: breast cancer; and PPV: positive predictive value.
* The interval between 2 general screening rounds is 6 months in view of biannual CBE. At the first and subsequent general screening rounds, women can be screened only by CBE without an imaging technique. Not always did women receive all 3 screen modalities together, thus, numbers do not necessarily add to 1909.
5th + next rounds
In Table 2, the screening parameters of mammography and MRI for invasive cancers in the 3 types of imaging screening rounds (first round of women without prior mammograph, first round of women with prior mammography, and subsequent rounds) are presented. As described in Materials and Methods for this analysis, we selected only rounds in which both mammography and MRI were performed. As a consequence, apart from 6 DCIS, 5 invasive breast cancers were excluded from the analysis (see footnote Table 2). Of 1723 evaluable women, 303 (18%) women were never previously screened by mammography and MRI before the study, and 1420 (82%) women had already been screened by mammography before entry in the study.
Table 2. Screening Parameters for First and Subsequent Imaging Rounds of 39 Evaluable Invasive Breast Cancers*
No. of Tests
No. of Positive Tests n (%)
No. of True Positives
No. of False Negatives
Detection Rate per 1000 Tests
True Positive Rate, Sensitivity, %
False Positive Rate 100-Specificity, %
Based on 39 of 44 invasive breast cancers (6 ductal carcinomas in situ [DCIS] were excluded). Reasons for omitting 5 cases: in 3 cases no magnetic resonance imaging (MRI) or mammography was performed, because of pregnancy (n = 2; 1 in the interval 16 months after imaging, 35 mm, microinvasive lymph node) or refusing MRI (n = 1). In the fourth case, a tumor was detected by an additional mammography, after a screening mammography classified as BI-RADS “0” but at a location different from the first lesion. The fifth case was detected at a screening visit consisting of only a complete clinical breast examination (CBE). Percentage positive tests indicates percentage of tests with a positive result (number of positive tests divided by number of tests); True positives, number of positive tests in women who appeared to have cancer; False positives, number of positive tests in women who do not appeared to have cancer; False negatives, number of negative tests in women who appeared to have cancer; Detection rate, number of tests per 1000 tests that detected a cancer (the number true positives divided by the number of tests); PPV, percentage of positive tests that were true positives (the number of true positives divided by the number of positive tests); True positive rate, percentage of cancers that had a positive test result (the number of true positives divided by the total number of cancers); False positive rate, percentage positive test results in women who did not appear to have cancer (the number of false negatives divided by the total number of tests in women who did not appear to have cancer).
First imaging round, without prior mammography
First imaging round, with prior mammography
Subsequent imaging rounds
First imaging round, without prior mammography
First imaging round, with prior mammography
Subsequent imaging rounds
For mammography, the detection rate of invasive cancers was 16.5 per 1000 tests in the first imaging round in women without prior mammography, which was significantly (P = .003) higher than in the first screening round in women with prior mammography (2.1 per 1000 tests) and in subsequent rounds (2.1 per 1000 tests). For MRI, the detection rate of invasive cancers was also highest in the first round in women without prior mammography (13.2 per 1000 tests), but it was also high in the first round in women with prior mammography (9.9 per 1000 tests) compared with the detection rate in subsequent rounds (5.3 per 1000 tests) resulting in a significant trend. Similar results for mammography and MRI were found for the PPV (Table 2). There was no large difference between the true-positive rate (sensitivity) of mammography (5 of 7) and MRI (4 of 7) for invasive breast cancers in the first round in women without prior mammography (71.4 vs. 57.1%). The difference in sensitivity between mammography and MRI was largest in the first imaging round with prior mammography, i.e., 20.0% (3/15) vs. 93.3% (14 of 15) (P = .003). Moreover, also in subsequent rounds, this difference remained significant for invasive tumors (29.4 vs. 76.5%; P = .02). The difference in false-positive rate (1-specificity) between mammography and MRI was largest in the first imaging round in women with prior mammography (5.5 vs. 14.0%; P<.001) in favor of mammography. For subsequent rounds, the false-positive rate also remained lower for mammography than for MRI (4.6 vs. 8.2%; P<.001).
The area under the curve (AUC) in the receiver operating characteristic (ROC) curves for subsequent rounds was 0.665 for mammography and 0.850 for MRI (Fig. 1). The difference between the areas was 0.185 (95% confidence interval [CI] 0.003–0.367); P<.05).
The characteristics of 45 evaluable tumors (the 6 DCIS are separately indicated) detected in the 3 different kinds of imaging screening rounds (including 4 interval cancers) are presented in Table 3 in comparison with those of the 2 symptomatic control groups. In total, 22 prevalent invasive breast cancers were found during the first screening round with (n = 15) or without (n = 7) prior mammography, whereas 17 invasive incident cancers were detected during subsequent screening rounds. Nine (41%) of these 22 prevalent invasive cancers were 1 cm or smaller, which was not significantly different in comparison with 53% (9 of 17) of the incident cancers (cancers found during subsequent screening periods) (P = .52). The node-negativity rate (including isolated tumor cells) was 86% (19 of 22) for prevalent cancers and 69% (11 of 16) for incident cancers (P = .24). Of the tumors detected during the first round 23% (5 of 22) had a high differentiation grade, in contrast to 47% (8 of 17) in the incident tumors (P = .10).
Table 3. Characteristics of Detected Breast Cancers by Imaging Screening Round (39 Invasive BC + 6 DCIS)
1st Imaging Round Without Prior Mammography
1st Imaging Round With Prior Mammography
Control Group 1, NCR
Control Group 2, Prospective Study
BC: breast cancer; DCIS: ductal carcinoma in situ; NCR: National Cancer Registry.
T 1-2 cm
T 2 cm
Micrometastasis 0.2–2.0 mm
When comparing characteristics of screen-detected tumors with those of the control groups, especially the size of the incident tumors were more frequently (53%) ≤1 cm in comparison with both symptomatic control groups (14.0% (P<.001) and 12.5% (P = .005), respectively). Tumors found during subsequent rounds were also, although not significantly, less likely (31.3% i.e., 5 of 16) to be lymph node positive (including micrometastasis) than those in both symptomatic control groups, where the lymph node positivity rate was 52.4% (P = .09) and 56.4% (P = .09), respectively.
In various countries, breast cancer screening for young women with a familial or genetic predisposition is a realistic option to reduce breast cancer mortality.3 Because of the low sensitivity of mammography in this group, other venues to improve early detection are explored. For this purpose, the value of MRI as a screening method is being investigated,5–8, 13–16 for instance, in our large Dutch MRI screening study, the MRISC.8 This study concluded that MRI was more sensitive than mammography, but less specific, and that tumor characteristics from women in the study were more favorable when compared with those of age-matched symptomatic controls.8 However, data on screening parameters and tumor characteristics can differ between different screening rounds, especially between the first and subsequent rounds. Therefore, apart from overall results, it is also important to report results separately for the different rounds in a screening program. In the MRISC study as well as in other MRI screening studies,13–16 MRI (and in some studies also ultrasound) was added to the already existing screening program using mammography and CBE. This means that, within this study, the result of a first MRI is compared with the result of a mammography in women with a prior mammography before start of the study. Therefore, it may be that in the first screening round in the group of women with prior mammography, mammographically occult carcinomas were preferentially detected by MRI, thus exaggerating differences in sensitivity between MRI and mammography. However, in subsequent rounds, all women had both a mammography and MRI in a prior round, resulting in less imbalance. Therefore, although the number of tumors detected in the MRISC study was still relatively small, in this detailed study we separately analyzed screening results in these different screening rounds.
We found that, for all 3 screening modalities, the positive test rates and PPV were reasonably stable over time. Only for MRI, there was a significant trend of a decreasing positive test rate over time. Apart from a higher rate of prevalent cancers, possible reasons for this latter observation are the availability of a previous MRI for comparison and more experience of radiologists in interpretation of MRI in subsequent rounds, resulting in less false-positive tests.
For mammography as well as MRI, the detection rate was indeed highest in the first round in women without a prior mammography. In contrast to mammography, for MRI the detection rate in the first imaging round in women with prior mammography was also higher than in subsequent rounds. This may be caused by the preferential detection of mammographically occult cancers by MRI in this round. Warner et al.15 found only a slightly higher overall detection rate by MRI in the first round (46.6 per 1000 tests) compared with the second round (36.8 per 1000 tests).
Because of a small absolute number of cancers (n = 7) in the first imaging round in women without prior mammography, it is not possible to draw hard conclusions about differences in sensitivity of mammography and MRI in this subgroup of women. The difference in sensitivity between mammography and MRI was most obvious in the first imaging round in women with a prior mammography (20.0 vs. 93.3%; P = .001). Therefore, the inclusion of the large subgroup with prior mammography before start of the study probably resulted in overestimation of the difference in sensitivity between mammography and MRI. Indeed, in subsequent screening rounds, the difference in sensitivity (29.4% vs. 76.5%) decreased, but it remained significantly (P = .02) in favor of MRI. This means that also in subsequent rounds, MRI is a more sensitive method than mammography for early detection of breast cancer. Warner et al.15 found also a higher difference in sensitivity in the first round (38% vs. 85%) compared with the second round (43% vs. 71%) (Table 4), for all breast cancers combined, although they made no differentiation between a first round in women without and with a prior mammography. They also found a slight decrease in sensitivity of MRI and a similar sensitivity of mammography in subsequent rounds. Although the percentage of women screened before their MRI study was not given in their reports,15, 20 MRI was also added to the existing screening program consisting of mammography and CBE in probably a significant number of women. Regarding the British Royal College of Radiologists MARIBS study, Leach et al.16 found no decrease in sensitivity of MRI between the first and subsequent rounds (75% vs. 80%, respectively), while the sensitivity of mammography also remained unchanged (40% vs. 40%). However, the differences among these 3 studies are probably not significant.
Table 4. Comparisons of Sensitivity and Specificity of First and Subsequent Imaging Screening Rounds in 3 Different Studies
As expected, for mammography, the specificity was lower in the first imaging round in women without a prior mammography than in subsequent rounds, but in both first imaging rounds together (with and without prior imaging), specificity was comparable to those in subsequent rounds (94% vs. 95%) (Table 4). For MRI, the specificity was lower in both first imaging rounds together than in subsequent rounds (87% vs. 92%) (Table 4). In the already mentioned Canadian (Warner et al.15) study, the same trend was seen; the specificity of MRI increased from 93% in the first round to 97% and 99% in the second and third round, respectively.15 In the British study this trend was absent and specificity of MRI was comparable for the first and subsequent rounds (82 vs. 81%, respectively).16 The findings in the MRISC and Canadian study are in line with findings from the Dutch nationwide breast cancer screening program for women aged 50–75 years21 and the screening programs in the United States.22, 23 The specificity of mammography in those studies was also lower in the first screening round than in subsequent rounds. A possible explanation of these observations is the absence of previous images during the first round for comparison,12 but for MRI, it is also possible that radiologists involved in these studies gained experience in interpretation of this detection method during subsequent rounds, thus lowering false-positive rates.
For sensitivity, we saw the highest contrast between mammography and MRI in the first imaging round of women with prior mammography (94.5% vs. 86.0%; P<.001), but there was also a significant difference between both modalities in subsequent rounds (95.4% vs. 91.8%; P<.001). Thus, also the difference in specificity between mammography and MRI regarding the overall results may be an overestimation, because in the first round in the group of women with prior mammography, the mammography can be compared with the previous one, whereas this is not possible for the MRI.
Because results from subsequent rounds are most predictive for long-term effects, we generated ROC curves of these rounds separately. Comparison of the AUC, a measure for the discriminating capacity of a diagnostic test (including sensitivity and specificity), showed that also in subsequent rounds, MRI could better discriminate between benign and malignant cases than mammography (P = .046).
The ultimate goal of screening is to reduce the stage of breast cancer at the time when the cancer is first detected in order to reduce the breast cancer mortality. Smaller tumors in incident rounds as compared with the prevalent round are expected in screening programs through the detection of more large slowly growing tumors in the prevalent round (length-time bias).9 In contrast with this theory, it has been reported that, in many screening studies, tumors detected in subsequent rounds are not smaller than tumors detected in the first round or that the differences are very small.24 In our study, the characteristics of prevalent tumors found in the first round without prior mammography were suggestive of this length-time bias sampling: the majority (85.7%) was Bloom and Richardson Grade 1, all were node-negative, even though there were fewer (28.6%) small tumors (≤1 cm) than in the 2 other subgroups. However, these differences in prevalent tumor characteristics were not significant when compared with those of the cancers detected during subsequent rounds. Characteristics of invasive cancers found overall,8 and also in subsequent rounds, were more favorable with respect to tumor size, nodal status, and differentiation grade compared with tumors found in symptomatic control patient groups.
In conclusion, it is reassuring that also in subsequent rounds, a higher sensitivity and a better (P = .046) discriminating capacity (based on ROC AUCs) of MRI in comparison with mammography was found along with a favorable tumor stage compared with age-matched symptomatic controls. As results of these subsequent screening rounds are most predictive for long-term effects, we expect that this screening program will contribute to a decrease of breast cancer mortality in these high-risk women.
The authors thank all participants and collaborators within the MRISC study for their contribution to this study. From Erasmus MC, Rotterdam: L. Aronson, P. Bos, S. van Dooren, A.N. van Geel, E.J. Meijers-Heijboer, M. Menke, A.J. Rijnsburger, A. Tibben, and D. Urich. From Leiden University Medical Center, Leiden: C. van Asperen, A. Nieborg, V. T. H. B. M. Smit, and M. N. J. M. Wasser. From Netherlands Cancer Institute, Amsterdam: R. Kaas, W. Koops, H. Peterse, M. Piek-den Hartog, A. Schlief, and M. van de Vijver. From University Medical Center, University of Groningen: C. Dorbritz, T. van Echten, S. van Hoof, A. M. van der Vliet, and J. de Vries. From University Medical Center Nijmegen: J.O. Barentsz, L. V. A. M. Beex, H. Brunner, J. H. C. L. Hendriks†, M. Hogenkamp, R. Holland, M. Stoutjesdijk, A. L. M. Verbeek, M. Verhoeven, and T. Wobbes. From VU University Medical Center, Amsterdam: I. Groot, P. A. M. van Leeuwen, F. Menko, and A. Taets van Amerongen.