HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text Submit a response Services Email this article to a friend Alert me to new issues of the journal

© RSNA, 2008

Breast Cancer Screening in BRCA1 Mutation Carriers: Effectiveness of MR Imaging–Markov Monte Carlo Decision Analysis

Appendix E1

Natural History (Figure)

Overview of breast cancer natural history. Ovals indicate health states. Arrows indicate the state in which patients begin each cycle and point to the state into which the patient enters during a Markov cycle. Parameters governing transitions are adjacent to arrows in the figure. For example, four parameters (age, preclinical phase duration, incidence multiplier, and proportion DCIS vs invasive) govern the transition from healthy to either undiagnosed DCIS or undiagnosed invasive cancer. Although not included in the schematic, women can die from causes other than breast cancer in the healthy, undiagnosed DCIS, or undiagnosed invasive cancer states during each Markov cycle.

In the Breast Cancer Development and Detection module, the values for seven unobservable parameters were estimated by means of calibration (Tables E1–E7). Two of these parameters were related to the onset of cancer and were simultaneously adjusted during model calibration. The preclinical phase was defined as the number of years between the onset of cancer and the time of diagnosis, and was initially estimated on the basis of the medical literature (24). Because we assumed that not every cancer that developed would be detected, the age-specific incidence estimates (4) used to determine the onset of cancer were adjusted by an incidence multiplier. Five additional parameters were related to tumor progression:

1. The probability that a new cancer would be an invasive cancer at the outset (as opposed to beginning as DCIS).
2. The annual rate of progression from DCIS to invasive cancer.
3. The proportion of nonpalpable DCIS (not detectable at physical examination).
4. The annual stabilization rate for nonpalpable DCIS (cessation of growth and capacity for invasion).
5. The mean of the lognormal distribution of the tumor growth parameter. (The standard deviation [0.68] of the lognormal distribution was obtained from the medical literature [24]).

Table E1. Key Input Parameters for Breast Cancer Development and Detection Module: Sensitivity of Physical Examination by Tumor Size (43,44)

Table E2. Key Input Parameters for Breast Cancer Development and Detection Module: Frequency of Node Positivity and Distant Metastases by Primary Tumor Size—Clinical Detection

Table E3. Key Input Parameters for Breast Cancer Development and Detection Module: Parameters Derived by Calibration

Table E4. Key Input Parameters for Treatment and Follow-up Module: Breast Cancer Mortality by Stage

Note.—Effect of tamoxifen (on estrogen receptor-positive tumors): 31% mortality reduction (38).

Table E5. Key Input Parameters for Screening Module: Sensitivity of Screening by Modality and Primary Tumor Size in BRCA Mutation Carriers (13)

Table E6. Key Input Parameters for Screening Module: Specificity of Screening by Modality (13)

Table E7. Key Input Parameters for Screening Module: Frequency of Node Positivity and Distant Metastases by Primary Tumor Size—Screening Detection (26)

Staging of invasive breast cancer was determined by primary tumor diameter at diagnosis combined with the presence or absence of lymph node and distant metastases. Logistic regression was performed on data provided by the Breast Cancer Surveillance Consortium to estimate the frequency of nodal and distant metastases on the basis of primary tumor diameter at diagnosis and method of cancer detection (clinical presentation vs screening), which provided the input parameters in Tables E2 and E7.

Screening

The estimates for the diagnostic test performance of mammography, MR imaging, and combined mammography and MR imaging were obtained from a multicenter trial of women at increased familial risk for breast cancer (13). Because the estimate of sensitivity of the combination of mammography and MR imaging used in this study was derived from a clinical trial in which patients underwent screening with both tests, an assumption of conditional independence was not necessary.

The diagnosis of cancer included a three-stage testing sequence of screening, diagnostic work-up, and biopsy. Women with positive screening test results underwent further diagnostic work-up. Women whose diagnostic work-up results were negative or benign were tracked as having had a false-positive screening examination. Women whose diagnostic work-up results were suspicious subsequently underwent biopsy to establish a final diagnosis of either malignant or benign disease. We assumed that women with cancer who had a positive screening test result also had a positive diagnostic work-up, leading to a recommendation for biopsy. In women without cancer, the probability of a biopsy recommendation after diagnostic work-up was assumed to be conditionally independent of the initial screening test (mammography, MR imaging, or combined mammography and MR imaging). If the biopsy results demonstrated benign disease, the woman was tracked as having had a false-positive screening examination, a false-positive diagnostic work-up, and a false-positive biopsy recommendation. Women with negative screening test results underwent no further intervention until the next screening event. If a cancer was missed on a screening test, cancer progression continued until the next screening event or until the cancer presented clinically as an interval cancer. Once breast cancer was diagnosed at biopsy after either clinical presentation or screening detection, women entered the Treatment and Follow-up module.

While the possibility that mammographic examinations may induce a cancer in women with BRCA gene mutations is a real concern, the likelihood of this is very low (41,42). Because of the rarity of this occurrence, we made a simplifying assumption to omit the induction of cancer by mammographic screening.

Model Calibration

Model calibration involves adjusting input parameters so that the model reproduces reported results. Model parameters with uncertain values are varied over a plausible range to identify parameter estimates that best produce expected output on the basis of data from published sources. A grid search across seven-parameter space was followed by adjustment of individual parameter values to identify parameter sets that minimized the total sum of squared errors between model output and preselected calibration targets. Acceptance windows of ±5% were applied to each calibration target to further select the best-fitting parameter set.

We required that candidate parameter sets meet primary calibration targets before evaluating goodness of fit for secondary calibration targets. We chose the following primary calibration targets: (a) the percentage of cancers ≤2 cm in diameter (33.9%) and (b) the percentage of cancers manifesting clinically as DCIS (4.6%). These targets were identified from data in the SEER database (25) from 1975 to 1980, a time frame before the onset of mammographic screening. We decided that information from the prescreening era in SEER’s large, geographically diverse database was a suitable source for calibration targets because the vast majority of cancers in BRCA gene mutation carriers and in the general population are ductal carcinomas.

Two secondary calibration targets were also selected: (a) the cumulative incidence of breast cancer for the simulated cohort and (b) the proportion of cancers presenting as DCIS when a strategy of mammographic screening is applied. On the basis of penetrance estimates from the literature (4), we chose a cumulative incidence target of 65% by age 70. Although the proportion of cancers presenting as DCIS in the general population in 2005 was 21.7% (5), reports of screening mammography of BRCA gene mutation carriers vary widely. Because the numbers of BRCA gene mutation carriers screened to date are still relatively small (10–12), we applied an acceptance window of 15%–30% for all cancers in the model diagnosed with screening mammography.

For a cohort of 25-year-old asymptomatic BRCA1 gene mutation carriers, the model estimated that 63.1% undergoing clinical surveillance (no imaging) developed breast cancer during their lifetimes, approximating published estimates (4). The model also produced two primary outputs (percentage of cancers ≤2 cm in diameter and percentage of DCIS among detected cancers) within 5% of calibration target values. When annual mammography screening was applied, the proportion of DCIS identified at screening mammography (19%) was within the preset window of 15%–30%.

References

41. Narod SA, Lubinski J, Ghadirian P, et al. Screening mammography and risk of breast cancer in BRCA1 and BRCA2 mutation carriers: a case-control study. Lancet Oncol 2006;7:402-406.

42. Law J, Faulkner K. Cancers detected and induced, and associated risk and benefit, in a breast screening programme. Br J Radiol 2001;74:1121-1127.

43. Fryback DG, Stout NK, Rosenberg MA, Trentham-Dietz A, Kuruchittham V, Remington PL. The Wisconsin Breast Cancer Epidemiology Simulation Model. J Natl Cancer Inst Monogr 2006;36:37-47.

44. Stout NK. Quantifying the benefits and risks of screening mammography for women, researchers and policy makers: an analysis using a simulation model. Madison, Wis: University of Wisconsin, 2004.

45. Dunphy FR, Spitzer G, Fornoff JE, et al. Factors predicting long-term survival for metastatic breast cancer patients treated with high-dose chemotherapy and bone marrow support. Cancer 1994;73:2157-2167.

46. Yeatman TJ. The natural history of locally advanced primary breast carcinoma and metastatic disease. Surg Oncol Clin N Am 1995;4:569-589.

47. Sickles EA, Miglioretti DL, Ballard-Barbash R, et al. Performance benchmarks for diagnostic mammography. Radiology 2005;235:775-790.

This Article Abstract Full Text Submit a response Services Email this article to a friend Alert me to new issues of the journal