| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Evidence-based Practice |
1 From the Department of Radiology (J.M.L., D.B.K., P.M.M., E.F.H., G.S.G.), Institute for Technology Assessment (J.M.L., P.M.M., E.F.H., G.S.G.), Division of Hematology/Oncology, Department of Medicine (P.D.R.), Massachusetts General Hospital, 101 Merrimac St, 10th Floor, Boston, MA 02114; and Department of Health Policy and Management, Harvard School of Public Health, Boston, Mass (M.C.W., G.S.G.). Received February 1, 2007; revision requested April 4; revision received May 21; accepted June 14; final version accepted August 8. Supported in part by National Institutes of Health grant R25 CA92203. J.M.L. supported in part by a GE–Association of University Radiologists Research Award. Address correspondence to J.M.L. (e-mail: jlee45{at}partners.org).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
Materials and Methods: A microsimulation model was developed to compare three annual screening strategies versus clinical surveillance: (a) mammography, (b) MR imaging, and (c) combined MR imaging and mammography. Input parameters were obtained from the published medical literature, existing databases, and expert opinion. The model was calibrated to targets from the Surveillance Epidemiology and End Results database (1975–1980) compiled during a period prior to the onset of widespread mammographic screening. Sensitivity analysis was performed to evaluate the effect of uncertainty in parameter estimates.
Results: With clinical surveillance, the estimated median diameter of invasive breast cancers at presentation was 2.6 cm. Average life expectancy was 71.15 years. With annual screening with mammography, MR imaging, or combined mammography and MR imaging, median invasive tumor diameters at diagnosis decreased to 1.9, 1.3, and 1.1 cm, respectively. Annual screening with mammography, MR imaging, or combined mammography and MR imaging increased average life expectancy by 0.80 year, 1.10 years, and 1.38 years, respectively, and decreased relative mortality from breast cancer (16.8%, 17.2%, and 22.0%, respectively). Program sensitivity was greater than 50% only with MR imaging screening strategies. The majority of women undergoing screening had one or more false-positive screening examinations (53.8%, 80.2%, and 84.0% for mammography, MR imaging, and combined mammography and MR imaging, respectively). Many women also underwent one or more biopsies for benign disease (11.3%, 26.3%, and 30.3%, respectively). Results were sensitive to BRCA1 penetrance estimates and to MR imaging sensitivity in the detection of ductal carcinoma in situ.
Conclusion: Annual screening with combined mammography and MR imaging provides BRCA1 mutation carriers with the greatest life expectancy gain and breast cancer mortality reduction. However, an important trade-off of this strategy is an increased rate of false-positive screening results and biopsies performed for benign disease.
Supplemental material: http://radiology.rsnajnls.org/cgi/content/full/246/3/763/DC1
© RSNA, 2008
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
Mammography is a fast, noninvasive, and relatively inexpensive screening test that has been shown to decrease breast cancer mortality in the general population (7,8). Although the sensitivity of mammography in the general population is relatively high (83%–95%) (9), the sensitivity of mammography in BRCA gene mutation carriers is substantially lower, in the range of 33%–56% (10–12). This is thought to be related to multiple factors such as the younger age at screening and increased breast density in these women, as well as to pathologic and imaging characteristics of breast cancers in this population.
Compared with mammography, magnetic resonance (MR) imaging has higher sensitivity for the detection of breast cancer, particularly in women at increased risk (13–21), and may have a role in screening. The main limitations of breast MR imaging are its cost and comparatively lower specificity. The possible gains in sensitivity combined with the added benefits of early detection with MR imaging may come with acceptable risks and costs in select populations of women in whom disease prevalence is increased compared with that in the general population. The trade-offs related to screening with MR imaging are most likely to be reasonable for those women who carry high-risk susceptibility gene mutations. However, MR imaging screening strategies—whether alone, in combination, or alternating with mammography—have yet to be fully evaluated or standardized.
As with mammography, early detection with MR imaging presumably decreases breast cancer mortality, although results of no studies to date have confirmed this. It is unlikely that a randomized controlled trial will be performed to assess whether screening with MR imaging can decrease breast cancer mortality because of the length of follow-up that would be required to accrue a sufficient number of deaths of breast cancer in the various patient groups, the large number of patients required to demonstrate a significant difference between groups, and the associated expense that would be incurred. This provides an ideal opportunity to apply decision analysis and disease modeling methods to evaluate the potential effectiveness of screening with breast MR imaging.
We have developed a computer model of breast cancer natural history that simulates individual lifetimes and incorporates the effects of mammographic and MR imaging screening. The purpose of our study was to project the intermediate and long-term clinical outcomes of MR imaging screening for breast cancer in women with BRCA1 gene mutations.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
We developed a single-cohort Markov model of breast cancer that was analyzed as a first-order Monte Carlo simulation. The model was populated with 500 000 25-year-old women who were asymptomatic BRCA1 mutation carriers and consisted of three linked modules: (a) breast cancer development and detection, (b) treatment and follow-up, and (c) screening (Fig 1). Model development and analysis of model output was performed by one author (J.M.L.) in consultation with the remaining coauthors. Input parameters were identified through a critical review of the published medical literature and publicly available databases. Coauthors with clinical expertise in breast imaging (D.B.K., J.M.L.) and oncology (P.D.R.) provided guidance regarding model assumptions, model structure, and input parameter estimates. Coauthors with expertise in decision analysis (G.S.G., M.C.W., P.M.M.) provided guidance regarding model construction, calibration, and interpretation of results.
|
Breast Cancer Development and Detection
Individual women entered the breast cancer development and detection module at the beginning of the simulation. During each cycle of the model, a woman could remain healthy, develop undiagnosed breast cancer, or die of causes unrelated to breast cancer (see Competing Mortality section below). The base-case estimates of age-specific annual incidence of breast cancer for BRCA1 gene carriers were obtained from a meta-analysis of pooled pedigree data from 22 studies involving 8139 index patients (4).
Key assumptions regarding breast cancer onset and progression included the following (see also Appendix E1, http://radiology.rsnajnls.org/cgi/content/full/246/3/763/DC1):
1. Ductal carcinoma in situ (DCIS) is a nonobligate precursor of invasive breast cancer (22,23).
2. The natural history of DCIS was considered to include lesions that stabilize in capacity for growth and invasion but not lesions capable of regression.
3. Tumors were considered to follow a logistic growth model with lognormal distribution of the growth parameter (24).
4. Growth rates were drawn from a lognormal distribution at the time of onset and were independent of age.
Model parameters were obtained from the published literature or existing databases maintained by the Surveillance Epidemiology and End Results (SEER) Program (25) and the Breast Cancer Surveillance Consortium (26). The Breast Cancer Surveillance Consortium approved our request to receive aggregated deidentified data created by its Statistical Coordinating Center. These data provided the input parameters referenced in Tables E1–E3 (http://radiology.rsnajnls.org/cgi/content/full/246/3/763/DC1). Additional key input parameters are shown in these tables. If required input parameters were not available, their values were estimated through model calibration (Appendix E1, http://radiology.rsnajnls.org/cgi/content/full/246/3/763/DC1). Women with undiagnosed breast cancer could present with signs and symptoms of breast cancer or remain asymptomatic, with progression of disease during each cycle spent in the undiagnosed breast cancer health state.
The clinical detection of invasive breast cancer was based on tumor diameter. The clinical detection of DCIS was based on (a) the proportion of DCIS that can be palpable (enabling detection at physical examination) and (b) tumor diameter. Nonpalpable DCIS was assumed to be clinically undetectable at any size. In the model, some cancers were never detected, and women could die with undiagnosed tumors. Within a cycle, women with undiagnosed breast cancer could present with signs and symptoms of breast cancer. The likelihood of lymph node involvement and distant metastases was based on primary tumor diameter at diagnosis (26). Estrogen receptor status was assigned during staging (27,28).
Treatment and Follow-up
Once diagnosed, a woman with breast cancer underwent staging of the disease and treatment. She then entered the treatment and follow-up module, in which her condition could remain stable, she could die of breast cancer, or she could die of other causes in each cycle. On the basis of study results (6,27,29–31) that demonstrated that women with BRCA mutations, stage for stage, have similar prognoses and outcomes as women with sporadic cancers, the probability of each event in the treatment and follow-up module was based on breast cancer stage at diagnosis, estrogen receptor status, and age at diagnosis. Key input parameters are shown in Table E4 (http://radiology.rsnajnls.org/cgi/content/full/246/3/763/DC1). All women who died of breast cancer were first diagnosed with breast cancer, because it is unlikely that a woman would have widely disseminated metastatic disease at presentation and die of breast cancer within a single 3-month cycle.
Competing Mortality
We assumed that women with undiagnosed breast cancer died of non–breast cancer causes. Non–breast cancer mortality was age dependent and was derived from United States life tables (32). Because the majority of BRCA gene mutation carriers identified to date are white, mortality for white women was used and adjusted for age and sex, with mortality from breast cancer subtracted. An additional adjustment was made for increased ovarian cancer mortality beyond that of the general population among BRCA gene mutation carriers by using age-specific incidence estimates for BRCA gene mutation carriers (4) and age-specific relative survival rates from the SEER database (25).
Screening
The model was used to evaluate three screening strategies relative to a strategy of clinical surveillance (no imaging-based screening): (a) annual mammography, (b) annual MR imaging, and (c) annual combined mammography and MR imaging. All screening strategies began at age 25 years. The annual screening mammography strategy was based on the recommendations of the Cancer Genetics Studies Consortium (33) and the National Comprehensive Cancer Network (34). The combined annual mammography and MR imaging screening strategy was based on the complementary depiction abilities of the two modalities, as described in ongoing screening studies of women at increased hereditary risk of cancer (13–15,17). Because the prevalence of DCIS has been reported to be lower in cancers detected in women with BRCA1 mutations (35), screening with MR imaging only has been advocated as an alternative strategy (18). Complete compliance with annual screening was assumed. Key input parameters are shown in Tables E5–E7 and Appendix E1 (http://radiology.rsnajnls.org/cgi/content/full/246/3/763/DC1).
Outcomes
The primary outcomes of the analysis were the life expectancy gain with the three annual screening strategies versus that with clinical surveillance and the breast cancer mortality reduction obtained with annual screening. Intermediate outcomes evaluated were as follows: cumulative incidence of breast cancer, mean age at diagnosis, median diameter of invasive cancers, percentage of cancers 2 cm or smaller in diameter, and stage distribution of cancers detected with each screening strategy. Program sensitivity, defined as the number of screening-detected cancers divided by the total number of diagnosed cancers, was calculated for each screening strategy. The diagnostic consequences of screening were also evaluated. These included the following: the percentage of women with one or more false-positive screening test result in their lifetimes, the percentage of women with one or more false-positive biopsy result for benign disease, and the frequency distribution of these false-positive test results.
Sensitivity Analysis
We analyzed the model as a Markov Monte Carlo simulation to examine first-order uncertainty, which characterizes the random variability in individual outcomes conditional on underlying parameter values. We examined the effect of second-order uncertainty, which characterizes the imprecision of knowledge regarding parameter values, by performing sensitivity analysis over the plausible clinical range for the following input parameters: penetrance of BRCA gene mutations, age-specific risk of cancer, and diagnostic test performance of screening modalities. In evaluating the sensitivity of screening, we evaluated the effect of adjusting detection thresholds as well as sensitivity for varying lesion sizes. In evaluating the specificity of screening, we examined the reported range of specificity for mammography (93%–99%), MR imaging (81%–94%), and combined mammography and MR imaging (75%–94%) (13–15,17). Although Leach et al (13) reported no increase in specificity between initial and subsequent screening examinations, other investigators (14,15) have reported such an increase. We therefore performed an additional sensitivity analysis, assuming a 5% increase in specificity for subsequent screenings for each modality.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
|
|
When compared with clinical surveillance, annual mammography depicted a greater percentage of DCIS, but the percentage of local (node-negative) stage cancers was unchanged (Fig 2). Compared with annual mammography, annual MR imaging depicted more local stage cancers but not as many cases of DCIS. Annual MR imaging did reveal more cases of DCIS than clinical surveillance. Because mammography is more sensitive than MR imaging for depicting DCIS and MR imaging is more sensitive than mammography for depicting invasive cancers, annual combined mammography and MR imaging depicted the most early stage cancers, with a corresponding decrease in regional and distant stage cancers.
|
Diagnostic Consequences of Screening
Although annual mammography alone provided life expectancy gain and breast cancer mortality reduction, the decreased sensitivity of mammography in the BRCA1 population limited the number of screening-detected cancers. Only 40% of diagnosed cancers were detected with mammographic screening; the remaining 60% of diagnosed cancers were found clinically as interval cancers. It was only when screening strategies including MR imaging were implemented that more than 50% of diagnosed cancers were detected with screening.
Although screening with MR imaging resulted in increased life expectancy and breast cancer mortality reduction, it was associated with a high rate of false-positive test results (Table 3). With MR imaging screening strategies, 80.2%–84.0% of women undergoing screening had one or more false-positive screening examinations during their lives. Of these women, 35.0%–42.0% were recalled for further evaluation at least four times during their lives. In addition, more than 25% of women who underwent MR imaging screening also underwent biopsies that ultimately revealed benign breast disease.
|
Sensitivity analysis of diagnostic test performance indicated that the breast cancer mortality estimates were sensitive to the diagnostic test performance of MR imaging. If the MR imaging sensitivity for depicting DCIS lesions larger than 10 mm in diameter decreased from the base-case estimate of 50% to 47%, breast cancer mortality reduction from annual mammography and annual MR imaging screening became equal. At this threshold level, the life expectancy with annual MR imaging screening (72.25 years) remained greater than that with annual mammography (71.95 years). Annual combined mammography and MR imaging consistently resulted in greater life expectancy gain and larger breast cancer mortality reduction than all other annual screening strategies.
Sensitivity analysis of the specificity of screening indicated the range of magnitude for false-positive test results. As the specificity of MR imaging was increased from 81% to 97%, the percentage of women with one or more false-positive screening MR imaging examinations in their lifetimes decreased from 80% to 32%. The percentage of women who underwent one or more biopsies that demonstrated benign disease decreased from 26% to 5%. A 5% increase in specificity for subsequent screening examinations provided a more modest decrease in false-positive screening results (from 80% to 75%) and biopsies for benign disease (from 26% to 21%). Similar results were seen for the annual combined mammography and MR imaging strategy. As the specificity of the combined strategy increased from 77% to 94%, the percentage of women with one or more false-positive screening MR imaging examinations in their lifetimes decreased from 84% to 49%. The percentage of women who underwent one or more biopsies that demonstrated benign disease decreased from 30% to 10%. A 5% increase in specificity for subsequent combined screening examinations likewise provided a more modest decrease in false-positive screening results (from 84% to 80%) and biopsies for benign disease (from 30% to 26%).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
Estimates of intermediate outcomes obtained from our model approximate those in the medical literature. The reported sensitivity of mammography for screening women with high risk ranges from 33% to 56% (10–15,17,20). Our model estimate of program sensitivity (the number of screening-detected cancers divided by the total number of diagnosed cancers) for mammography is 39.4%. Our model estimate of the percentage of invasive cancers 2 cm in diameter or smaller that are detected with mammography and MR imaging (72.6%) corresponds with 65% in a previously published report (14). Our model estimate of the percentage of node-positive cancers identified with mammography and MR imaging (18.6%) falls within the reported range of 13%–37% (13–15). Similarly, our model estimate of the percentage of DCIS identified with mammographic screening (19.4%) falls within the range in published reports (12%–27%) (13–15,17).
Although annual MR imaging screening resulted in greater average life expectancy and breast cancer mortality reduction than annual mammography, results of sensitivity analysis indicated that mortality reduction was dependent on MR imaging test performance. MR imaging sensitivity for depicting DCIS lesions larger than 10 mm in diameter needed to remain higher than 47% for the breast cancer mortality reduction from annual MR imaging screening to exceed that of annual mammography. In the base case, mammography was more sensitive in depicting DCIS but failed to reveal many invasive cancers. MR imaging failed to reveal some cases of DCIS but was able to depict more cases of invasive cancer at an earlier stage, providing a larger increase in average life expectancy. A slight decrease in the MR imaging sensitivity for depicting DCIS would result in some women with DCIS that might have been detected with mammography ultimately dying of invasive breast cancer when screened only with MR imaging.
It may seem surprising at first that at this threshold level of MR imaging sensitivity, the breast cancer mortality reduction from annual MR imaging and annual mammography equalized, while annual MR imaging continued to provide additional life expectancy gains. However, even though we identified a condition in which a similar number of women died of breast cancer causes, many of the women who underwent annual MR imaging instead of annual mammography obtained substantial gains in survival from having their cancers diagnosed at an earlier stage, which is reflected in the increased life expectancy with annual MR imaging screening.
The reported sensitivity of MR imaging in the detection of DCIS varies in the published medical literature, possibly because of differences among institutions in terms of parameters like imaging protocol and interpretation criterion and variability in interpretation thresholds among individual radiologists, as well as because of the heterogeneity of the biology and imaging appearance of DCIS lesions. Our analysis has identified a threshold level that affects projected outcomes. However, it is equally important to note that our analysis indicated that annual combined mammography and MR imaging screening consistently resulted in greater life expectancy gain and larger breast cancer mortality reduction than all other annual screening strategies.
An important trade-off related to screening with MR imaging is the high rate of false-positive test results. Our study provides a quantitative point estimate and range of the frequency of false-positive screening results. When MR imaging screening strategies were applied, the majority of women undergoing screening (80.2%–84.0%) had at least one false-positive screening examination during their lives. More than a third were recalled for further evaluation multiple (four or more) times. In addition, more than 25% of women who underwent MR imaging screening also underwent biopsies that ultimately revealed benign breast disease. Sensitivity analysis indicated a modest decrease in the magnitude of false-positive test results with a 5% gain in specificity for subsequent examinations. However, if specificity at the high end of what is reported in the medical literature can be achieved, the projected magnitude of false-positive test results is substantially decreased.
Our model estimates of the benefits of screening BRCA1 gene mutation carriers with breast MR imaging are more conservative than those of a published cost-effectiveness analysis. Plevritis et al (36) estimated that annual combined mammography and MR imaging screening of BRCA1 gene mutation carriers yielded a higher life expectancy gain (2.1 years vs our estimate of 1.38 years), had a higher program sensitivity (85% vs our estimate of 69.9%), and would reveal smaller cancers (77% of detected cancers 2 cm in size or smaller vs our estimate of 72.6%). Similarly, Plevritis et al estimated the breast cancer mortality reduction due to annual combined mammography and MR imaging screening to be 38%, versus our estimate of 22.0%.
Our model explicitly included the natural history of DCIS, a structural assumption that likely accounts for much of the lower estimated benefit. DCIS has a favorable prognosis compared with that of invasive cancer, with the mortality rate of DCIS being approximately one-tenth of the mortality rate for node-negative invasive cancers (37). In contrast, Plevritis et al (36) did not model the natural history of DCIS. In their model, DCIS was embedded in the same stage as localized small invasive tumors, with an implicit assumption that both groups of tumors would have a similar, favorable prognosis. Attributing the life expectancy gain associated with the detection of small invasive cancers to screening-detected DCIS (some cases of which may not progress and which has a lower observed mortality rate) provides a higher estimate of life expectancy gain than if the two types of cancer were evaluated separately.
A second assumption that may have differed between the models is that of the diagnostic test performance of combined mammography and MR imaging. Our estimate of the sensitivity of the combination of mammography and MR imaging was derived from data in a clinical trial in which patients underwent screening with both tests (13), so an assumption of conditional independence is unnecessary. Diagnostic performance in the Plevritis et al (36) model was based on assumptions of a size-dependent threshold mechanism for tumor detection. Whether combined test results were assumed to be conditionally independent was not specified.
A potential limitation of our study was the extrapolation of data from the general population to the BRCA1 mutation carrier population in developing the natural history model of breast cancer. Whenever possible, parameter estimates specific to BRCA mutation carriers were used. However, some of the available data came from small, very selected patient samples. In those instances, we chose to base input parameter estimates on sources with larger samples of sporadic breast cancers in the general population, such as the large databases maintained by SEER (25) and the Breast Cancer Surveillance Consortium (26) or meta-analyses of randomized trials (38).
Future refinement of the model will include the incorporation of additional clinical screening strategies such as alternating mammography and MR imaging every 6 months, as well as strategies combining screening with prophylactic bilateral salpingo-oophorectomy, which decreases breast cancer risk by approximately 50% (39,40), and evaluating the cost-effectiveness of mammography and MR imaging screening in BRCA gene mutation carriers.
In summary, our evaluation of annual screening strategies for asymptomatic BRCA1 mutation carriers indicated that screening with annual mammography and MR imaging provides the greatest life expectancy gain and breast cancer mortality reduction. However, an important trade-off of this strategy is a high rate of false-positive screening test results and subsequent biopsies for benign disease.
|
ADVANCES IN KNOWLEDGE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
|
IMPLICATION FOR PATIENT CARE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
|
FOOTNOTES |
|---|
Abbreviations: DCIS = ductal carcinoma in situ • SEER = Surveillance Epidemiology and End Results
See Materials and Methods for pertinent disclosures.
Author contributions: Guarantor of integrity of entire study, J.M.L.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, J.M.L., P.D.R.; statistical analysis, J.M.L., E.F.H., M.C.W., G.S.G.; and manuscript editing, J.M.L., P.M.M., E.F.H., P.D.R., M.C.W., G.S.G.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
