Clinical outcomes of modelling mammography screening strategies
Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.
by Martin J. Yaffe, Nicole Mittmann, Pablo Lee, Anna N.A. Tosteson, Amy Trentham-Dietz, Oguzhan Alagoz and Natasha K. Stout
Organized provincial breast cancer screening programs in Canada typically include a mechanism to invite eligible women to attend at recommended intervals, standardized reporting, quality assurance, monitoring of outcomes, and a link between the screening process and subsequent imaging to assess suspicious screen-detected findings. However, the age range and frequency of population screening have been subjects of debate, and implementation of screening varies across the country.
This analysis employs a validated microsimulation model of breast cancer, adapted to the Canadian context,Note 1 to predict health outcomes associated with different digital mammography screening strategies (including No Screening) across different age ranges. The model estimates the benefits, harms, limitations, and use of resources for each strategy.
A computer simulation modelling approach was used to examine the health benefits and costs of digital mammography for 11 screening strategies, which vary by age of starting and ceasing screening and frequency of examinations, compared with No Screening (Text table 1). They represent screening strategies currently used in various Canadian jurisdictions and recommendations of the U.S. Preventive Services Task ForceNote 2,Note 3 and the Canadian Task Force on Preventive Health Care.Note 4 Because of interest in the impact of screening at ages 40 to 49, annual screening of this age group was also modelled.
The health outcomes are breast cancer mortality reductions and life-years (LY) saved, compared with No Screening. Expected resource use (number of mammograms, diagnostic work-up of positive findings from screening, therapeutic procedures, and management per screening strategy) was also modelled.
The modelNote 5,Note 6 has been used to study the efficacy and cost-effectiveness of breast cancer screening.Note 7, Note 8, Note 9 Its adaptation to the Canadian context is described in a companion paper.Note 1 The framework for the outcome analysis is the University of Wisconsin Breast Cancer Epidemiology Simulation Model,developed under the U.S. National Cancer Institute-funded Cancer Intervention and Surveillance Modeling Network (CISNET) program. The model was Canadianized under a grant provided by The Canadian Breast Cancer Foundation.
For each screening strategy, the model predicted age-specific incidence and mortality. Based on life tables, the number of LY lost to breast cancer detected at each age was estimated.Note 10
The calculations pertain to a single birth cohort―women born in 1960. This allows estimation of age-specific breast cancer outcomes, such as incidence and mortality, independent of cross-sectional population-based effects associated with year of birth. Published data were used to describe the accuracy of cancer detection and the efficacy of treatment in the model.Note 9, Note 11
Clinical input parameters
Upon tumour detection, all women received baseline treatment (for example, surgery with or without radiation). Adjuvant treatment was assigned based on age, stage and hormone receptor status.
The percentages of women with ER+/ER- (estrogen receptor positive/estrogen receptor negative) disease were assigned by the model, based on U.S. SEER data.Note 12, Note 13 Assignment of HER2 (human epidermal growth factor receptor 2) status was added.Note 14
Treatment effectiveness was implemented by a "cure"/"no cure" model. A woman who is "cured" will not die of breast cancer; a woman who is not "cured" may die of breast cancer or non-breast cancer causes. Baseline cure fractions by stage at detection were determined by model calibrations. These baseline cure fractions were modified by the presence of adjuvant treatment using observed hazard reductions from clinical trials (Text table 2).Note 11
For each woman, the model recorded age at detection of a breast cancer, its "stage" (in situ, local invasive, regional involvement, or distant metastasis), age at which she died, and whether the death was attributable to breast cancer or another cause. Calculations were made for a birth cohort of 2,000,000 women for each screening strategy. Results are expressed as age- and stage-specific incidence and mortality per 1,000 women in the cohort who were alive at age 40 (abbreviated to "per 1,000"). Mortality reductions attributable to screening were obtained by comparing the number of deaths due to breast cancer for each screening strategy with the corresponding number for No Screening at ages 40 to 99. The number of recalls following screening and the number of biopsies that were negative for breast cancer were calculated, as well as the number of screening examinations and the number of women required to be screened per life saved and per LY saved.
To estimate the number of LYs lost to breast cancer, the age at which a woman died because of breast cancer as predicted by the model was compared with the mean age at which she would be expected to die based on Canadian life tables.Note 10
The model predicts that with No Screening, in a 1960 birth cohort of 2,000,000 women, 37,075 (1.9% of the cohort or 12% of those who developed breast cancer) would die of breast cancer, and 259,342 (13% of the cohort or 87% of those who developed breast cancer) would die of other causes (data not shown). On average, each woman diagnosed with breast cancer after age 40 who died of breast cancer lost 19.1 LYs. In the absence of screening, the number of LYs lost to breast cancer per 1,000 women alive at age 40 was 360.
The stage-specific incidence rates of breast cancer provided by the model for No Screening can be compared with the incidence rates for annual screening at ages 50 to 74 (Figure 1). Figure 2 shows deaths due to breast cancer per 100,000 by age for No Screening, for biennial screening at ages 50 to 74, and for annual screening at ages 40 to 49 followed by biennial screening until age 74. The effect of screening on mortality reduction persists for some time after the age when screening ceases, but mortality eventually rises toward the unscreened rate.
For each screening strategy, Table 1 reports mortality reductions based on deaths that occurred between the age when screening began and 15 years after screening was discontinued. This shows the effect when screening ceases many years before life expectancy. It also emulates calculations that would be done in a randomized trial with a finite follow-up period, in this case, 15 years after the screening intervention.
Table 1 also summarizes the screening outcomes of each strategy, with a focus on efficiency. The maximum number of screening examinations that a woman could receive during her lifetime is given for each strategy, along with the number of screening examinations that must be conducted per breast cancer death averted, and the number of women who must be screened per death averted and per LY gained.
In general, the amount of mortality reduction was associated with the total number of lifetime screens (Figure 3). However, for the same number of screens, more frequent screening after age 50 appears to have a greater impact than beginning screening earlier.
When the number of LYs saved is considered, the story changes—a greater impact was achieved by screening women in their 40s than by reducing the interval between screens (Figure 4). The model also predicts that outcomes for the hybrid strategies, in which the screening interval is lengthened from annual to biennial at an age approximating menopause (in this case, 50), were superior to those for biennial or triennial screening, but not as good as those for annual screening.
Table 2 presents the number of screening examinations for each strategy, the number of recalls for non-invasive imaging whose outcome is negative (no cancer), and the number of negative breast biopsies arising from the lifetime screening experience for each strategy per 1,000 women alive at age 40. Greater intensity or longer duration of screening increased the probability that a woman would be recalled after a suspicious screen for further imaging that turned out to be negative, and the probability of undergoing a biopsy that was negative for breast cancer. These biopsies are performed after both a suspicious (positive) screening result and a positive result of additional non-invasive imaging. The calculation incorporated British Columbia data on the positive yield of biopsiesNote 15: 16.6% for women in their 40s; 33.7% for those in their 50s; 49.2% for those in their 60s; and 54.7% for those in their 70s. Based on these data, from the number of cancers detected, the number of biopsies that would be performed on women who did not have cancer was estimated. The numbers were lowest for biennial and triennial screening at ages 50 to 69 (144 and 141 per 1,000, respectively) and highest for annual screening at ages 40 to 74 (308 per 1,000).
As expected, in the absence of screening (Figure 1), the observed incidence of in situ cancer is very low, with most cancers being detected by the woman herself or clinically when they are invasive and either local or regional (nodal metastases). A comparison of the No Screening predictions with those for screening—in this case, annual at ages 50 to 74—shows that screening results in a downward stage shift, with more in situ and localized invasive cancers detected and less disease discovered at more advanced stages during the period when screening is performed. Incidence returns to the non-screened level fairly rapidly once screening is discontinued, at age 74 in this example.
On the initial screen at age 40 or 50, a "spike" in cancer detection is evident. These "prevalent" cancers, which were occult in previous years when screening did not occur, will include a number that are larger and more advanced than those detected on subsequent screens. Some may be slow-growing cancers that are unlikely to be lethal, but others will be lethal and represent the potential for reducing mortality had they been detected earlier by screening. This possibility is supported by the model. For annual screening beginning at age 40, the model predicts that 11.2% of the women with prevalent invasive cancers, detected at age 40, will die of breast cancer. This is similar to the value of 9.4% for women whose cancers were detected in the next screen at age 41—that is, mortality due to the prevalent cancers is about the same as that associated with incident cancers. Similarly, if screening starts at age 50, 10.9% of those with breast cancers detected in the prevalence screen at age 50 will die of breast cancer, and 8.3% of those with incident cancers detected in the next screen at age 51 will die.
For screening that begins around age 40, the model shows a reduction of mortality for all screening strategies compared with No Screening, with the reduction starting about two years after the initiation of screening (Figure 2). The sharp drop in mortality suggests advantage in earlier detection, even for more advanced cancers. This can be inferred from the model, in that the natural history of the cancers in the screened and unscreened cohorts is the same until the point when screening begins. Treatment of some of the cancers detected on the first screen results in some women being alive two years later, while their counterparts in the unscreened cohort have died. Given the very high survival for early-stage breast cancer, this implies that the cancers in both groups must have been fairly advanced to produce such a pronounced effect at two years. The mortality reduction begins to diminish shortly after screening ceases, becoming minimal around age 89, 15 years after screening has ended.
A strategy that has been suggested is to screen annually at ages 40 to 49 (or slightly older), and then biennially until age 74. When this strategy is compared with biennial screening at ages 50 to 74 (now operational in several provinces), the model predicts 1.8 fewer breast-cancer-related deaths and 54 LYs saved per 1,000 women alive at age 40. Given that approximately 230,000 women in Canada turn 40 each year, annual screening at ages 40 to 49 translates into approximately 414 premature deaths averted and 12,420 LYs saved each year.
For the same cohort and the two screening strategies, multiplying differences in values in Table 2 by the same factor (230) predicts that a total of 2,232,150 additional screening examinations would be performed over the cohort's collective lifetime. There would be 226,090 additional recalls for further imaging examinations where cancer was not found and 25,990 negative biopsies. However, Table 2 slightly underestimates the specificity achieved in Canadian screening, which is closer to 93%. If this were the case, the number of additional recalls with no cancer found would decrease to 210,060, and the number of additional negative biopsies, to 24,170. Both of these factors scale directly with (1-Sp), and therefore, even small improvements in specificity would yield substantial reductions in recall examinations and negative biopsies. For this reason, digital breast tomosynthesis has attracted considerable interest as a possible replacement for mammography, as early studies have shown up to a 30% improvement in Sp.Note 16
The model predicts that a much lower number of women must be screened to avert a breast cancer death than reported in the U.S. and Canadian Task Force Recommendations.Note 2, Note 4 For example, for annual screening at ages 40 to 49, the model predicts that 526 women would have to be screened per breast cancer death averted; the Canadian Task Force reported the number as 2,108.Note 4 This difference occurs, in part, because the Recommendations refer to the number of women required to be invited into a randomized trial of screening rather than the number actually required to be screened to achieve that benefit.
Strengths and limitations
A strength of this analysis is that the model does not contain explicit assumptions about the mortality reduction provided by screening.
Although all six CISNET models agree in their predictions of cancer incidence and qualitative conclusions about the value of screening and treatment, some variability is evident in the quantitative estimates of benefit associated with screening and treatment. The University of Wisconsin Breast Cancer Simulation model tends to be one of the more "optimistic" in terms of the screening benefit,Note 7, Note 9 but the impact of the difference is not large.
In the model, the growth rate of cancers was not assigned an age dependence, despite some indication that premenopausal cancers grow more quickly.
The analysis assumed full compliance with screening and treatment by all eligible women. To reflect the reality of partial participation, the cohort would simply be split according to the percentage participating in a particular strategy or not receiving screening. The resulting incidence, deaths or LYs lost would be calculated as a weighted combination of the outcomes for the two groups.
According to the model, more frequent screening detected more breast cancers, resulted in fewer breast cancer deaths and years of life lost, and was associated with an increased ratio of in situ to invasive cancers detected. At the same time, screening triggers recalls for further imaging examinations where cancer is not found and negative biopsies. The number of negative biopsies was lowest for biennial and triennial screening at ages 50 to 69 and highest for annual screening at ages 40 to 74.
The work performed to create this report was supported by a contract from The Canadian Breast Cancer Foundation, Ontario Region. The authors are grateful to Statistics Canada and to Dr. Michael Wolfson at the University of Ottawa for supplying key incidence data. The University of Wisconsin breast cancer simulation model used in this analysis was supported by grant number U01 CA152958 from the National Cancer Institute through the Cancer Intervention and Surveillance Modeling Network (CISNET). Model input data on the performance of screening mammography were provided by the National Cancer Institute-funded Breast Cancer Surveillance Consortium (BCSC) grant number UC2CA148577 and contract number HHSN261201100031C. The content of this study does not necessarily represent the official views of the National Cancer Institute or the National Institutes of Health. The collection of BCSC cancer data was supported by several state public health departments and cancer registries throughout the United States. A full description of these sources is available at: http://www.breastscreening.cancer.gov/work/acknowledgement.html. The authors thank the participating women, mammography facilities, and radiologists for the data they provided for this study. A list of the BCSC investigators and procedures for requesting BCSC data for research purposes are provided at: http://breastscreening.cancer.gov/