Professional Documents
Culture Documents
J Med Screen
0(0) 1–9
The effect of mammography screening ! The Author(s) 2018
Reprints and permissions:
regimen on incidence-based breast sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0969141318780152
cancer mortality journals.sagepub.com/home/msc
Abstract
Objectives: Incidence-based mortality quantifies the distribution of cancer deaths and life-years lost, according to age
at detection. We investigated the temporal distribution of the disease burden, and the effect of starting and stopping
ages and interval between screening mammography examinations, on incidence-based mortality.
Methods: Incidence-based mortality was estimated using an established breast cancer simulation model, adapted and
validated to simulate breast cancer incidence, screening performance, and delivery of therapies in Canada. Ten strategies
were examined, with varying starting age (40 or 50), stopping age (69 or 74), and interval (1, 2, 3 years), and
“No Screening.” Life-years lost were computed as the difference between model predicted time of breast cancer
death and that estimated from life tables.
Results: Without screening, 70% of the burden in terms of breast cancer deaths extends between ages 45 and 75.
The mean of the distribution of ages of detection of breast cancers that will be fatal in an unscreened population is
61.8 years, while the mean age of detection weighted by the number of life-years lost is 55, a downward shift of
6.8 years. Similarly, the mean age of detection for the distribution of life-years gained through screening is lower than
that for breast cancer deaths averted.
Conclusion: Incidence-based mortality predictions from modeling elucidate the age dependence of the breast cancer
burden and can provide guidance for optimizing the timing of screening regimens to achieve maximal impact. Of the
regimens studied, the greatest lifesaving effect was achieved with annual screening beginning at age 40.
Keywords
Breast cancer screening, mammography, incidence-based mortality, mammography screening regimens, quality-adjusted
life-years, age to begin screening, screening interval
Date received: 14 September 2017; accepted: 10 May 2018
5
Introduction Applied Research in Cancer Control, Department of Pharmacology,
University of Toronto, Toronto, Canada
In developing strategies for reducing the burden of 6
Office of Research, Cancer Care Ontario, Toronto, Canada
7
death and morbidity associated with breast cancer, it Department of Population Health Sciences and Carbone Cancer Center,
University of Wisconsin-Madison, Madison, USA
is useful to understand the origin of that burden, 8
Department of Industrial and Systems Engineering, University of
namely: At what age are breast cancers that ultimately Wisconsin-Madison, Madison, USA
9
The Dartmouth Institute for Health Policy and Clinical Practice, Geisel
School of Medicine at Dartmouth, Lebanon, USA
1 10
Physical Sciences Program, Sunnybrook Research Institute, Department of Population Medicine, Harvard Medical School and
Toronto, Canada Harvard Pilgrim Health Care Institute, Boston, USA
2
Departments of Medical Biophysics and Medical Imaging, University of
Toronto, Toronto, Canada Corresponding author:
3
Ontario Institute for Cancer Research, Toronto, Canada Martin J Yaffe, Sunnybrook Research Institute, Room S657, 2075 Bayview
4
Health Outcomes and PharmacoEconomic (HOPE) Research Centre, Ave, Toronto, ON M4N 3M5, Canada.
Sunnybrook Research Institute, Toronto, Canada Email: martin.yaffe@sri.utoronto.ca
2 Journal of Medical Screening 0(0)
lead to death, detected? and How can we intervene most on the basis of data derived from the meta-analysis of
effectively to reduce that burden? One metric for this the chemotherapy trials conducted by the Early Breast
burden is incidence-based mortality (IBM),1,2 where Cancer Trialists’ Collaborative Group.11–15 Patients are
the number of deaths due to breast cancer or either cured or not cured (die from breast cancer) and, in
the number of life-years lost (a measure of premature the model, are also exposed to competing causes of
mortality) is expressed versus the age at which the dis- death. For those not cured, the cancers continue to
ease surfaces clinically or is detected by screening. grow until they reach a specified size that is used as a
We have recently completed an evaluation of the surrogate for the point of death. Women without breast
effects of different screening scenarios, i.e. the ages to cancer eventually die from other causes, again according
begin and discontinue mammography screening and to life-table-weighted random number selection. The
the interval between screens, on health outcomes, model reflects the performance of modern digital mam-
cost-effectiveness, and cost–utility.3–5 We used a well- mography, and includes use of trastuzumab as part of
established breast cancer model, developed using the therapeutic regimen,3 with survival improvements
US data and adapted to the Canadian context. The based on the report by Howlader et al.16 In addition,
model allows the impact of screening to be quantified we updated the sensitivities and specificities for screen-
in terms of lives saved (premature breast cancer deaths ing mammography, based on the data from the NCI’s
averted) as well as life-years and quality-adjusted life- Breast Cancer Surveillance Consortium (BCSC) in the
years (QALY) saved. US6 and from the Screening Mammography Program of
The model also allows the prediction, within a birth British Columbia (SMPBC).17
cohort, of the age at which each breast cancer would be The model incorporates four major components: (1)
detected by screening or through clinical finding, and initiation, growth, and progression of a breast cancer,
the age at which each woman in the cohort would die including its characteristics; (2) detection clinically or
from breast cancer or some other cause. In designing through screening; (3) response to therapy; and (4)
an intervention program, it is helpful to understand the death of cohort members by causes other than breast
distribution of disease burden, i.e. IBM, and how an cancer. These are more fully described in publications
intervention may affect that distribution. In an analo- on the University of Wisconsin CISNET Model.6–10
gous manner, the reduction of breast cancer-related Inputs are either applied directly in the model, or are
deaths or the increase in life-years attributable to used indirectly to calibrate model outputs, such as stage-
screening can be assessed in relation to the age at specific incidence and mortality, to empirical data. Some
which cancer was or could have been detected. key aspects of the model and input variables, as well as
identification of the source of data are summarized in
Table 1. Data are either applied directly in the model
Methods (D in Table 1), or used indirectly (I) in an iterative cal-
We utilized a version of the Wisconsin Cancer ibration to match model outputs such as incidence, stage
Intervention and Surveillance Modeling Network distribution, or mortality to empirical data.6–10,19
(CISNET) breast cancer model6–10 which we had pre- We validated the model’s fit to the Canadian popu-
viously tailored to the Canadian population.3 lation in three ways. First, we demonstrated the ability
The model is a discrete event simulation where a of the modified model to predict age-specific breast
large birth cohort (in this case two million women) cancer incidence in Canada in the absence of screening.
moves through life. Events are triggered by random Second, we estimated the ratios of detection of ductal
number selections with weightings for initiation of a carcinoma in situ (DCIS) to invasive cancer and found
breast cancer, its growth rate, hormone and HER2 that these ratios are comparable with those observed in
receptor status, detection by screening or clinically, the SMPBC.17 Finally, we compared the overall stage
etc., all calibrated such that average values match distribution of detected cancers predicted by the model
empirical data for these variables. Once a breast with that observed in the SMPBC and found that the
cancer is initiated in the model, its growth follows a agreement was good.3
Gompertz-type function, with various sizes used as sur- In this study, we applied the model to a hypothetical
rogates for progression of the cancers from in situ, to 1960 birth cohort of two million women, to estimate
locally invasive, regional, or distant metastases. For for each woman who will develop breast cancer the
example, for treatment, the model uses a cure/no cure age at which the cancer will either surface clinically
approach, with a cure rate that has been calibrated to or be detected by screening, and the age of death if a
match 10-year survival. The model also incorporates breast cancer death ultimately occurs. Utilizing
components reflecting cancers with limited malignant Canadian life table data,23 we have also estimated the
potential.8 The model begins with baseline survival age at which each woman would be expected to die if
data in the absence of treatment and modifies survival she did not die of breast cancer. The difference between
Yaffe et al. 3
Table 1. Summary of inputs to the model. Sources of the input data are also indicated.
Direct or
indirect
Category Characteristic input Description Source (ref number)
these ages gives the number of years lost due to prema- occurred, its “stage” at diagnosis (in situ, local inva-
ture breast cancer death and is a conventional measure sive, regional involvement, or distant metastasis), the
of premature mortality.24 The model was run for 10 age at death, and whether the death was attributable to
active screening regimens and a no screening scenario breast cancer or another cause. Results were expressed
(Table 2). These regimens represent screening strategies as age- and stage-specific incidence and mortality per
currently being used in different jurisdictions within 100,000 women alive at age 40. In some examples,
Canada, as well as recommendations from the US results were expressed per 100,000 in the initial
Preventive Services Task Force and The Canadian cohort when so indicated in the text. The breast
Task Force on Preventive Health Care.25–27 cancer mortality reductions were calculated as illustrat-
The model recorded the history of each woman by ed in Figure 1. Screened women were “followed” from
indicating the age at detection of a breast cancer if this the age of the first screen until 15 years after the
4 Journal of Medical Screening 0(0)
termination of screening for each regimen. Unscreened due to breast cancer. Here, follow-up on deaths is from
women were followed over the same time period as for age 40 to age 89 (i.e. 15 years after the age at which
that regimen. screening would end) for both the screened and
In the example in Figure 1, the arrow at left symbol- unscreened women. When this difference is summed
izes detection of cancer at age Ab, at a screen for a over the cohort for all ages at death in the follow-up
group of women screened annually from age 40 to period, the number of life-years lost to breast cancer,
74. In this case, the cancer was successfully treated detectable at age Ab with screening or Ac without, can
and the woman lives to her normal life expectancy, be estimated.
i.e. the expected time of death from all causes other To quantify the limitation or harms associated with
than breast cancer, at age Ae using Canadian life mammography screening we also estimated the number
tables.23 The arrow at age, Ac illustrates a more of abnormal recalls following screening, and the number
advanced cancer that surfaces clinically and results in of biopsy procedures that were negative for breast cancer,
a premature breast cancer death at age Ad. The differ- as well as the numbers of screens and women required to
ence between the actual and expected times of death, be screened per life saved and per life-year saved. These
i.e. (AeAd), provides an estimate of the life-years lost have been reported previously.4 In the model, benefits
and harms were quantified in a common “currency” by
Table 2. Screening scenarios modeled, predicted breast cancer expressing both in terms of quality-adjusted years of life
deaths per 100,000 women alive at age 40 and percent mortality lost or gained. Age-specific utility and disutility weighting
reduction relative to No Screening. Full adherence with screen- factors were obtained from published work,28–30 with
ing is assumed. adjustment for health states following the method of
Breast cancer Mortality Tosteson et al.31
Scenario deaths/100,000 women reduction (%) It is recognized that some cancers will be overde-
tected, i.e. detected at screening when they otherwise
No Screening 1940 0
would not have become clinically apparent before the
Annual 40–74 930 53.4
individual had died due to some cause other than breast
Annual 40–69 1030 50.2
Annual 50–74 1100 49.2 cancer. The probability of overdetection is greatest for
Annual 50–69 1200 45.5 screening of older women and for DCIS. Treatment of
Annual 40–49, 1150 42.0 these overdetected (often referred to as overdiagnosed)
Biennial 50–74 cancers can be considered a harm. This effect will be
Annual 40–49, 1240 38.7 included in future analysis because of the very large dis-
Biennial 50–69 parity in estimates of this phenomenon at this time.
Biennial 50–74 1330 35.9 Finally, because the model generates histories
Biennial 50–69 1420 32.3 including cancer stages for individual women, the
Triennial 50–74 1460 27.9
incidence-based paradigm can also be used to consider
Triennial 50–69 1540 24.6
the characteristics and outcomes associated with
premature
br. cancer br ca death
detected expected
through screening no screening life-years lost time
of death
life expectancy
Ab A c Ad Ae
Age Birth 40 50 60 70 80 90
Calendar Year 1960 2000 2050 2060
Figure 1. Theoretical effect of earlier detection through screening on years of life saved. In this example breast cancer is detected
clinically at age Ac and breast cancer death occurs at age Ad. Screening provides earlier detection at age Ab and woman lives to her
normal life expectancy at Ae with difference in life-years of Ae Ad. In our model, effects of screening are followed to 15 years past
time at which screening is terminated.
Yaffe et al. 5
cancers detected at specific times. We examined the the IBM demonstrates a prevalence “spike” at the first
predictions at three time points: that of the first year of screening (ages 40 or 50). This phenomenon is
screen, one year prior to that, and one year later. unrelated to the time at which a woman will die of
breast cancer, and reflects the fact that those deaths
Results are attributed to the pool of cancers that had been
previously undetected in the cohort, but which are
The model predicts that mortality begins to drop short- found earlier due to the lead time provided by screen-
ly after the onset of screening at age 40 or 50, with the ing. This spike is followed by a sharp drop in IBM.
number of breast cancer deaths per 100,000 women An example of the life-years lost, versus the age at
alive at age 40 dropping from 1940 for unscreened which the cancer surfaces or is detected by screening, is
women to 1100 for a cohort screened annually between shown in Figure 3 for different screening regimens. For
ages 50 and 74 and to 940 for annual screening from the no screening case, these curves express the burden
ages 40 to 74 (Table 2). Mortality increases relative to of breast cancer in terms of loss of life. The distribu-
annual screening about two years after a transition tions of breast cancer deaths according to the age at
from annual to biennial screening at age 50, but which cancer is detected (regardless of the age at which
remains well below that for an unscreened population.4 the death actually occurs) and life-years lost (which
Presumably these fairly rapid changes are due to earlier does depend on when death occurs) are summarized
or later treatment, respectively, of more advanced or in Table 3. Also shown in Table 3 are the number of
rapidly growing cancers occurring because of changes
in detectability related to screening. The mortality 1200
reduction begins to diminish shortly after the cessation
1000
of screening, but persists through the remainder of life,
becoming minimal by approximately age 89, 15 years 800
QALY lost
60
50
Number of deaths
40
30
20
10
0
20 30 40 50 60 70 80 90 100
Age at diagnosis
Figure 2. IBM. Number of breast cancer deaths (from initial birth cohort of 100,000) versus age at which cancer was detected for
No Screening for annual screening from 40 to 74 and for annual and biennial screening from 50 to 74 , .
6 Journal of Medical Screening 0(0)
0.04
Cancer Society;2 however, their calculations are based
0.03 on SEER data, which cover an undefined mix of
gained
85–89
64.7
287
190
3.7
0.8
0.7
0.1
that there is some mortality advantage of earlier detec-
10
3
0
tion even for cancers that are more advanced.
The model gives a measure of premature deaths,
80–84
97.3
649
436
5.5 where the mean age of detection with respect to the
1.8
1.6
0.3
10
43
1
number of eventual breast cancer deaths in unscreened
women is 61.8 years. When the mortality burden is
75–79
1170
807
275
7.1
3.3
3.0
40
4
2
mature deaths, Table 3 and a comparison of Figures 2
and 3 show that the distribution shifts to lower ages,
70–74
1849
1330
148
104
958
8.4
5.2
4.9
11
2838
2016
14.2
10.8
10.4
250
176
19
13
3446
2439
13.6
12.8
12.7
240
167
18
16
5605
4320
3022
14.0
15.8
15.9
246
170
20
5389
4165
2789
11.4
15.2
15.3
14.2
201
133
4792
3764
2232
13.5
13.8
Values do not add to 100% because of deaths due to cancer diagnosed before age 35 and after age 89.
8.7
9.6
90
15
3397
2717
1212
9.6
4.4
96
10
42
and 74.4
8
1778
NA
NA
NA
NA
3.2
6.2
6.5
55
receptors, no attempt has been made to apply further Institute as part of the CISNET. Model input data on the performance of
screening mammography was also supported by the National Cancer
characterization, e.g. by molecular subtype. Institute-funded BCSC grant number UC2CA148577 and contract number
Furthermore, the progression from in situ to invasive HHSN261201100031C. For a full description of U.S. data sources, please
disease has been based strictly on the modeled diameter see: http://www.breastscreening.cancer.gov/work/acknowledgement.html. To
discuss use of model run data, please contact Dr Yaffe.
of the tumor, and this has resulted in some discrepan-
cies between modeled and actual ratios of DCIS and
References
invasive cancers detected by screening. Finally, in the
1. Chu KC, Miller BA, Feuer EJ, et al. A method for partitioning cancer
model, the growth rate of cancers was not treated as mortality trends by factors associated with diagnosis: an application to
being age dependent, although there is some evidence female breast cancer. J Clin Epidemiol 1994; 47: 1451–1461.
that premenopausal cancers may grow more quickly.32 2. Oeffinger KC, Fontham ETH, Etzioni R, et al. Breast cancer screening for
women at average risk 2015 Guideline update from the American Cancer
Society. JAMA 2015; 314: 1599–1614.
3. Yaffe MJ, Mittmann N, Lee P, et al. Modelling mammography screening
Conclusion for breast cancer in the Canadian context: modification and testing of a
microsimulation model. Health Rep 2015; 26: 3–8.
This work illustrates how presentation of the results
4. Yaffe MJ, Mittmann N, Lee P, et al. Clinical outcomes of modelling
from simulation in an incidence-based manner can be mammography screening strategies. Health Rep 2015; 26: 9–15.
helpful in understanding the age distribution of the dis- 5. Mittmann NM, Stout NK, Lee P, et al. Total cost-effectiveness of mam-
mography screening strategies. Health Rep 2015; 26: 16–25.
ease burden of breast cancer. This approach provides
6. Stout NK, Lee SJ, Schechter CB, et al. Benefits, harms and costs for breast
insights that may be useful in optimizing the design of cancer screening following US implementation of digital mammography.
an intervention program aimed at reducing that J Natl Cancer Inst 2014; 106: dju092.
7. Batina NG, Trentham-Dietz A, Gangnon RE, et al. Variation in tumor
burden. Consideration of the burden and benefit in natural history contributes to racial disparities in breast cancer stage at
terms of QALYs rather than lives lost or saved incor- diagnosis. Breast Cancer Res Treat 2013; 138: 519–528.
porates both benefits and harms into the analysis. The 8. Fryback DG, Stout NK, Rosenberg MA, et al. The Wisconsin Breast
Cancer Epidemiology Simulation Model. J Natl Cancer Inst Monogr
use of QALYs or life-years shifts both the burden of 2006; 36: 37–47.
disease and the impact of screening downward mark- 9. Alagoz O, Ergun MA, Cevik M, et al. The University of Wisconsin Breast
edly in terms of the age at diagnosis. The model esti- Cancer Epidemiology Simulation Model: an update. Med Decis Making
2018; 38: 99S–111S.
mates that approximately 24% of the QALYs lost to 10. National Cancer Institute, Cancer Intervention and Surveillance Modeling
breast cancer (compared with 14% of deaths) are a Network. www.cisnet.gov (accessed 26 May 2018).
result of cancers that are detectable between ages 40 11. Early Breast Cancer Trialists’ Collaborative Group. Systemic treatment of
early breast cancer by hormonal, cytotoxic, or immune therapy. 133 rand-
and 49, and 23% of the QALYS gained by annual omised trials involving 31,000 recurrences and 24,000 deaths among 75,000
screening between ages 40 and 74 are attributable to women. Part I. Lancet 1992; 339: 1–15.
cancers detected in women in their 40s. As models 12. Early Breast Cancer Trialists’ Collaborative Group. Systemic treatment of
early breast cancer by hormonal, cytotoxic, or immune therapy. 133 rand-
become more sophisticated, incorporating such fea- omised trials involving 31,000 recurrences and 24,000 deaths among 75,000
tures as cancer subtypes and more detailed data on women. Part II. Lancet 1992; 339: 71–85.
response to therapy, the incidence-based approach 13. Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). Effects of
chemotherapy and hormonal therapy for early breast cancer on recurrence
may provide guidance for the design of stratified inter- and 15-year survival: an overview of the randomised trials. Lancet 2005;
ventions by identifying opportunities for maxi- 365: 1687–1717.
mal impact. 14. Clarke M, Coates AS, Darby SC, et al. Early Breast Cancer Trialists’
Collaborative Group (EBCTCG). Adjuvant chemotherapy in oestrogen-
receptor-poor breast cancer: patient-level meta-analysis of randomised
Acknowledgments trials. Lancet 2008; 371: 29–40.
We are grateful to Statistics Canada and to Dr Michael Wolfson at the 15. Peto R, Davies C, Godwin J, et al. Early Breast Cancer Trialists’
University of Ottawa for supplying key incidence data. The content is solely Collaborative Group (EBCTCG). Comparisons between different poly-
the responsibility of the authors and does not necessarily represent the official chemotherapy regimens for early breast cancer: meta-analyses of long-
views of Statistics Canada, the National Cancer Institute or the National term outcome among 100,000 women in 123 randomised trials. Lancet
Institutes of Health. We thank the participating women, mammography facil- 2012; 379: 432–444.
ities, and radiologists for the data they have provided for this study. A list of 16. Howlader N, Altekruse SF, Li CI, et al. US incidence of breast cancer sub-
the BCSC investigators and procedures for requesting BCSC data for research types defined by joint hormone receptor and HER2 status. J Natl Cancer
purposes are provided at: http://breastscreening.cancer.gov/. Inst 2014; 106: dju055.
17. BC Cancer Agency. Screening Mammography Program 2012 Annual
report, Vancouver, BC, Canada, www.bccancer.bc.ca/screening/
Declaration of conflicting interests Documents/SMP_Report-Annual Report2012.pdf (2012, accessed 9
The author(s) declared no potential conflicts of interest with respect to the February 2017).
research, authorship, and/or publication of this article. 18. Statistics Canada. Table 103-0550 New cases of primary cancer (based on
the February 2014 CCR tabulation file), by cancer type, age group and sex,
Canada, provinces and territories, annual. CANSIM (database), http://
Funding
www5.statcan.gc.ca/cansim/a26?lang¼eng&id¼1030550 (2015, accessed
The author(s) disclosed receipt of the following financial support for the 31 July 2015).
research, authorship and/or publication of this article: This work was sup- 19. University of Wisconsin. CISNET readers guide. Core profile documenta-
ported by a contract from The Canadian Breast Cancer Foundation. The tion. Madison: University of Wisconsin, https://cisnet.flexkb.net/mp/pub/
University of Wisconsin breast cancer simulation model used in this analysis CISNET_ModelProfile_BREAST_UWISC_001_07232013_58567.pdf
was also supported by grant number U01 CA152958 from the National Cancer (2013, accessed 27 April 2018).
Yaffe et al. 9
20. Holford TR. Analysing the temporal effects of age, period and cohort. 27. Canadian Task Force on Preventive Health Care. Recommendations on
Stat Methods Med Res 1992; 1: 317–337. screening for breast cancer in average-risk women aged 40–74 years.
21. Ries LAG and Eisner MP. Cancer of the female breast. In: Ries LAG, CMAJ 2011; 183: 1991–2001.
Young JL, Keel GE, et al. (eds) Cancer survival among adults: U.S. SEER 28. Fleishman JA. Demographic and clinical variations in health status. Report
program, 1988–2001, patient and tumor characteristics (NIH pub. No. 07- #15. Agency for healthcare research and quality, http://meps.ahrq.gov/
6215). Bethesda, MD: National Cancer Institute, 2007. mepsweb/data_files/publications/mr15/mr15.shtml (accessed 28
22. Breast Cancer Surveillance Consortium. BCSC Mammography Data. May 2018).
National Cancer Institute, http://breastscreening.cancer.gov/statistics/ 29. de Haes JCJM. The impact of a breast cancer screening programme on
mammography_data.html (2014, accessed 27 April 2018). quality-adjusted life-years. Int J Cancer 1991; 49: 538–544.
23. Statistics Canada. Life tables, Canada, provinces and territories, https:// 30. Hanmer J, Lawrence WF, Anderson JP, et al. Report of nationally repre-
www.statcan.gc.ca/pub/84-537-x/84-537-x2017001-eng.html (2011, sentative values for the non-institutionalized U.S. adult population for
accessed 28 May 2018). seven health-related quality of life scores. Med Decis Making 2006;
24. Ries LAG, Melbert D, Krapcho M, et al. (eds) SEER cancer statistics 24: 391–400.
review, 1975–2005. Bethesda, MD: National Cancer Institute, http://seer. 31. Tosteson ANA, Stout NK, Fryback DG, et al. Cost-effectiveness of digital
cancer.gov/csr/1975_2005/ (2008, accessed 28 May 2018). mammography breast cancer screening: results from ACRIN DMIST. Ann
25. US Preventive Services Task Force. Screening for breast cancer: U.S. Intern Med 2008; 148: 1–10.
Preventive Services Task Force recommendation statement. Ann Intern 32. Peer PG, van Dijck JA, Hendriks JH, et al. Age-dependent growth rate of
Med 2009; 151: 716–726. primary breast cancer. Cancer 1993; 71: 3547–3551.
26. Nelson HD, Tyne K, Naik A, et al. Screening for breast cancer: an update
for the U.S. Preventive Services Task Force. Ann Intern Med 2009;
151: 727–737.