Health Reports
Comparing Canada’s OncoSim-Breast model with the United States’ Cancer Intervention and Surveillance Modeling Network (CISNET) breast cancer models

Abstract

Background

The OncoSim-Breast model, developed by the Canadian Partnership Against Cancer and Statistics Canada, represents breast cancer-related events in the Canadian female population. This study aimed to compare OncoSim-Breast with recent results from the United States’ National Cancer Institute’s Cancer Intervention and Surveillance Modeling Network (CISNET) breast cancer models. The primary focus was on the impact of extending breast cancer screening to women aged 40 to 49.

Data and methods

The OncoSim-Breast model used Canadian demographics, competing mortality, and test performance, while the CISNET models used comparable United States data to analyze 10 different mammography screening scenarios. Lifetime outcomes were calculated for a cohort of 40-year-old women born in 1980, assuming perfect adherence to digital mammography screening. OncoSim-Breast’s estimates were compared with the median and range of estimates from the five CISNET models. The primary outcomes were breast cancer deaths averted and life years gained per 1,000 40-year-old women.

Results

OncoSim-Breast projected that starting screening at age 40 would lead to 1.7 breast cancer deaths averted and 53 life years gained per 1,000 women, compared with starting screening at age 50. CISNET models projected a median of 1.3 breast cancer deaths averted (range 0.8 to 3.2) and 43 life years gained (range 31 to 103) per 1,000 women for the same scenario. Secondary outcomes estimated by OncoSim-Breast and CISNET models were similarly consistent and comparable.

Interpretation

This study demonstrates that OncoSim-Breast’s estimates of the impact of starting breast cancer screening earlier align with those from CISNET models.

Keywords

cross-model validation, simulation, screening, age, death, quality-adjusted life year, mammography

Authors

Oguzhan Alagoz is with the Department of Industrial and Systems Engineering at the University of Wisconsin–Madison. Claude Nadeau and Rochelle Garner are with the Health Analysis and Modelling Division at Statistics Canada. Jean HE Yong is with the Canadian Partnership Against Cancer. Andrew Coldman is with Cancer Control Research at BC Cancer. Amy Trentham-Dietz is with the Department of Population Health Sciences and the Carbone Cancer Center at the University of Wisconsin–Madison.

 

What is already known on this subject?

• The guidelines from the Canadian Task Force on Preventive Health Care (CTFPHC), which recommend mammography screening every two to three years for average-risk women aged 50 to 74, are undergoing revisions because of questions about the optimal starting age for screening.
• The OncoSim-Breast model, developed by the Canadian Partnership Against Cancer, in collaboration with Statistics Canada, contributed to the latest draft breast cancer screening guidelines by the CTFPHC.
• The five breast cancer models from the United States’ National Cancer Institute’s Cancer Intervention and Surveillance Modeling Network (CISNET) informed the United States Preventive Services Task Force breast cancer screening recommendations in 2009, 2016, and 2024.

What does this study add?

• This study compared estimated outcomes from OncoSim-Breast with those from the CISNET breast cancer models.
• OncoSim-Breast projected that starting screening at age 40 would lead to 1.7 breast cancer deaths averted and 53 life years gained per 1,000 women, compared with starting screening at age 50. CISNET models projected a median of 1.3 breast cancer deaths averted (range 0.8 to 3.2) and 43 life years gained (range 31 to 103) per 1,000 women for the same scenario. 
• OncoSim-Breast estimated that starting mammography screening at age 40 or 45, compared with age 50, had a similar impact on breast cancer outcomes as in the CISNET breast cancer models.

Introduction

Breast cancer is the leading cause of cancer and the second leading cause of cancer death in women in Canada and the United States.^Note 1, ^Note 2 Mammography screening is an important approach to reducing deaths from breast cancer and the only screening modality supported by randomized controlled trial (RCT) evidence for widespread population use.^Note 3 While screening offers benefits, such as increased life expectancy and reduced breast cancer mortality, it also carries risks, including false-positive mammograms and overdiagnosis. Additionally, resources allocated to population-based screening must be carefully balanced against the need for symptomatic detection, particularly in health systems with limited resources. Thus, selecting an appropriate screening strategy requires careful consideration of these trade-offs. Given concurrent changes in breast cancer incidence patterns, screening test performance, and advancements in therapies, it is also essential to routinely evaluate the impact of different screening strategies on breast cancer outcomes. Breast cancer screening guidelines generated by the Canadian Task Force on Preventive Health Care (CTFPHC) in 2018 recommend that average-risk women aged 50 to 74 receive mammography screening every two to three years.^Note 4 The guidelines by the CTFPHC are undergoing revisions because of questions about the optimal starting age for screening.^Note 5  

RCT s would provide high-quality data and evidence to inform cancer screening guidelines. However, relying solely on RCT data for cancer screening has limitations, because large cohorts and long-term follow-up are often required to obtain reliable estimates of screening effectiveness.^Note 6 Additionally, RCT s are not always designed to address all relevant screening questions, such as the optimal age to start and stop screening, or the most effective screening interval. In the absence of comprehensive RCT data, mathematical modelling has been used increasingly to guide the development of cancer screening. For instance, Cancer Intervention and Surveillance Modeling Network (CISNET) models have informed the United States Preventive Services Task Force (USPSTF) screening recommendations for colorectal, lung, and breast cancers.^Note 7, ^Note 8, ^Note 9 More recently, the OncoSim-Breast model contributed to the latest draft of breast cancer screening guidelines by the CTFPHC.^Note 5

OncoSim-Breast was developed by the Canadian Partnership Against Cancer, in collaboration with Statistics Canada, and successfully represents the population-level breast cancer burden in Canada.^Note 10 In a validation experiment, OncoSim-Breast simulated the screening strategies of the United Kingdom Age trial, a randomized trial that compared annual screening among women aged 40 to 49 with usual care in the United Kingdom, and OncoSim-Breast’s projections of incidence and mortality were found to be similar to those observed in this trial.^Note 10 However, the model has never been cross-validated against other established breast cancer simulation models. This is noted as an important aspect of model validation by the report on validation from the joint task force of International Society for Pharmacoeconomics and Outcomes Research and the Society for Medical Decision Making.^Note 11 The purpose of this study is to compare estimated outcomes from OncoSim-Breast with those from the CISNET breast cancer models that informed the 2024 USPSTF breast cancer screening recommendations.^Note 12 This comparison provides a valuable benchmark for OncoSim-Breast, given the extensive validation of the CISNET breast cancer models in the United States context.^Note 13, ^Note 14 It should be noted that, to the best of the authors’ knowledge, OncoSim-Breast is the only model of this type representing Canadian breast cancer epidemiology. Therefore, the CISNET breast cancer models are considered the best alternative for comparison because they have undergone extensive validation. Furthermore, identifying areas where the models disagree is helpful, potentially uncovering biases or limitations in OncoSim-Breast, which could prompt further refinements. Moreover, successful cross-model validation enhances confidence in the findings of OncoSim-Breast and supports its credibility for use by policy makers and stakeholders.

Data and methods

Overview of the OncoSim-Breast model

The OncoSim-Breast model has been described in detail in a previous publication by Yong et al. (2022).10 Briefly, OncoSim-Breast simulates Canadian individuals, considering their sex, province or territory, and breast cancer-related risk factors (BRCA1 or BRCA2 gene mutation, family history of breast cancer, and exposure to postmenopausal hormone therapy). The model has a modular structure that includes a detailed natural history component representing both in situ (i.e., ductal carcinoma in situ) and invasive breast cancers, and a survival component representing province-specific breast cancer survival outcomes by subtype and cancer stage. The natural history model uses a structure similar to that of the University of Wisconsin Breast Cancer Epidemiology Simulation Model, one of the six CISNET breast cancer models.^Note 15 OncoSim-Breast was calibrated against age- and year-specific cancer incidence as reported by the National Cancer Incidence Reporting System (1969 to 1991) and the Canadian Cancer Registry (1992 to 2020). OncoSim-Breast also accounts for the dynamic demographics in Canada, including immigration and emigration trends and projections.

Overview of the Cancer Intervention and Surveillance Modeling Network breast cancer models and the modelling work for the 2024 United States Preventive Services Task Force recommendations

Since 2000, the National Cancer Institute in the United States has funded six independent research groups to establish the CISNET Breast Working Group, a collaborative modelling network of breast cancer researchers.^Note 6 While these six models have been developed independently, they collaborate to address crucial cancer control questions. Models use common high-quality data for parameter inputs, including observed trends in screening and treatment uptake within the population over time, performance of screening, treatment effectiveness, and other-cause mortality.^Note 6 The CISNET breast cancer models have been validated extensively; therefore, they provide a useful resource for comparing emerging models.^Note 16

Recently, the CISNET breast cancer models were commissioned to provide modelling evidence for the USPSTF’s 2024 breast cancer screening recommendations.^Note 17 The models completed a large number of simulation experiments related to mammography screening according to start and stop age, frequency, and modality, as well as strategies tailored to the level of breast cancer risk and comorbidity burden. These findings have been published and were used in the USPSTF’s 2024 recommendations for breast cancer screening.^Note 7, ^Note 12, ^Note 17 These scenarios and results in the USPSTF report provide the basis for the simulations and comparisons included in the present study.

Description of the simulation experiments

The CISNET models use United States-derived estimates and OncoSim-Breast uses Canada-derived estimates for key input parameters, including underlying breast cancer incidence, other-cause mortality (life expectancy), performance of screening mammography, and breast cancer survival after diagnosis. Therefore, a perfect alignment is nearly impossible, and model estimates are not expected to be equal. A total of 10 screening scenarios in the recent evidence provided by CISNET to the USPSTF were selected, and OncoSim-Breast was adjusted to replicate these scenarios as closely as possible. No methodological modifications related to natural history, screening performance, or survival after diagnosis were made to replicate CISNET analyses. Instead, the following main assumptions were made to adjust OncoSim-Breast for these simulations:

• A hypothetical cohort of Canadian women born in 1980 was simulated.
• Women with and without a family history of breast cancer followed the same screening strategies.
• Women with BRCA1 and BRCA2 genetic mutations or a personal history of breast cancer were excluded from the simulation.
• All simulated women adhered perfectly to the recommended screening strategy.
• No screening occurred prior to age 40.
• Digital mammography was used as the screening modality. 
• Sensitivity of digital mammography, specificity, survival after diagnosis, and other-cause mortality were used as per the default values in OncoSim-Breast version 3.6.3.9.

OncoSim-Breast has a detailed method for considering emigration and immigration within the simulated population. Among the options, scenarios were used that assumed that women emigrating out of Canada followed the same screening strategy after emigration, and their outcomes were also collected accordingly, regardless of where they lived. By contrast, the CISNET breast cancer models do not simulate any emigration or immigration scenarios explicitly. All the experiments in this study were conducted using the OncoSim-Breast model version 3.6.3.9.

A total of 10 screening scenarios were simulated: 1 with no screening; 6 scenarios involving annual and biennial strategies that started screening at age 40, 45, or 50 and continued to age 74; and 3 hybrid strategies that started with annual screening at age 40 or 45, then switched to biennial screening at age 50 or 55.

Analyses

All outcomes were calculated over the lifetime, starting at age 40 (the youngest age when screening starts among all strategies), regardless of when screening started for each strategy, to capture the lifetime effect of screening. The primary outcome was breast cancer deaths averted and life years (LYs) gained with screening. Secondary outcomes included quality-adjusted life years (QALYs) gained with screening and additional resource needs (number of mammograms) and harms (false-positive recalls, and benign biopsies after a false-positive recall). Harm-to-benefit ratios that were used in the recent CISNET modelling for the USPSTF 2024 analysis were also computed.^Note 7 The projections of the OncoSim-Breast model were compared against those from scenarios that considered digital mammography as the screening modality from the CISNET modelling report. The comparisons between OncoSim-Breast and the CISNET models were conducted deterministically.

All results were presented using two reference screening strategies: no screening and biennial screening from ages 50 to 74 (B50-74), which is the current practice in Canada. The latter was used to compare the projections from OncoSim-Breast and the CISNET models regarding the impact of extending screening from ages 50 to 74 to include ages 40 to 49. 

The CISNET modelling report for the USPSTF included results from each of the five CISNET breast models for screening with digital mammography.^Note 12 To assess whether OncoSim-Breast’s projections were within the range of the CISNET models, outcomes were reported from the median CISNET model (referred to as CISNET-median), as well as from the five CISNET models estimating the minimum value (referred to as CISNET-min) and the maximum value (referred to as CISNET-max) of the outcome under consideration. The range of outcomes for the CISNET models was determined by identifying the lowest (CISNET-min) and the highest (CISNET-max) estimates among their results.

Results

Benefits of screening

When biennial screening from ages 50 to 74 (B50-74)—the current screening strategy recommended in Canada—is used as the reference strategy, all three benefit metrics of screening (breast cancer deaths averted, LYs gained, and QALYs gained) estimated by OncoSim-Breast were comparable to the median values estimated by the CISNET models and were within the range of the values reported by the CISNET breast models (Table 1 and Chart 1). For example, OncoSim-Breast estimated that biennial screening from ages 40 to 74 (B40-74) would reduce breast cancer deaths by 1.7 per 1,000 women, relative to the B50-74 strategy. By comparison, the CISNET models estimated a median reduction of 1.3 breast cancer deaths (range: 0.8 to 3.2) with the B40-74 strategy, relative to B50-74. Furthermore, OncoSim-Breast estimated that B40-74 would lead to a gain of 53 LYs per 1,000 women, relative to B50-74. By comparison, the CISNET models estimated a median gain of 43 LYs (range: 31 to 103) with the B40-74 strategy, relative to B50-74 (Table 1 and Chart 1).  



Description of Chart 1



Data table for Chart 1
Table summary
This table displays the results of Data table for Chart 1. The information is grouped by Screening strategies (appearing as row headers), Breast cancer deaths averted compared with biennial screening for ages 50 to 74, Life years gained compared with biennial screening for ages 50 to 74, Quality-adjusted life years gained compared with biennial screening for ages 50 to 74, OncoSim-Breast, CISNET-median, Number, Error bar, Low and High, calculated using per 1,000 women units of measure (appearing as column headers).

Screening strategies	Breast cancer deaths averted compared with biennial screening for ages 50 to 74				Life years gained compared with biennial screening for ages 50 to 74				Quality-adjusted life years gained compared with biennial screening for ages 50 to 74
	OncoSim-Breast	CISNET-median			OncoSim-Breast	CISNET-median			OncoSim-Breast	CISNET-median
		Number	Error bar			Number	Error bar			Number	Error bar
			Low	High			Low	High			Low	High
	per 1,000 women
B50-74	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
B45-74	0.8	0.6	0.3	1.6	28.8	22.7	15.2	61.5	21.0	16.3	9.9	47.1
B40-74	1.7	1.3	0.8	3.2	53.0	43.2	31.4	102.7	29.8	31.4	19.5	77.1
A45-49, B50-74	1.5	1.1	0.9	2.6	43.4	31.0	29.5	78.3	37.4	20.7	19.8	58.4
A45-54, B55-74	1.7	1.0	0.8	2.2	52.2	39.7	30.5	76.2	35.5	23.7	20.1	44.8
A40-49, B50-74	2.4	1.8	1.4	3.8	73.9	59.2	49.7	123.2	50.0	37.1	32.1	89.8
A50-74	2.8	2.0	0.9	2.3	47.9	38.6	16.4	47.2	27.1	24.2	7.8	30.6
A45-74	4.2	2.7	1.9	4.9	90.7	68.9	45.8	118.7	56.6	44.3	27.3	84.8
A40-74	5.1	3.2	2.4	6.2	121.1	90.9	65.9	164.1	76.7	57.3	40.9	116.6
Notes: A: annual; B: biennial; CISNET-median: median value of the outcomes estimated by the Cancer Intervention and Surveillance Modeling Network (CISNET) models. Numbers indicate age ranges. The lower and upper error bars in the figures represent the minimum and maximum values, respectively, of the outcomes estimated by the CISNET models. Sources: OncoSim-Breast and CISNET breast cancer model outputs.

When no screening was used as the reference strategy, breast cancer deaths averted and LYs gained with screening projected by OncoSim-Breast were slightly greater than those projected by the CISNET models for all screening strategies (Table 2). However, LYs gained with screening estimated by OncoSim-Breast were within the range of the CISNET models for several screening strategies (Table 2), and QALYs gained with screening estimated by OncoSim-Breast were within the range of the CISNET models for all screening strategies (Table 2). In particular, QALYs gained with screening compared with no screening estimated by OncoSim-Breast were very similar to those estimated by the CISNET models, where the deviation between QALYs estimated with OncoSim-Breast and the CISNET models ranged from 18% to 30%, with a mean value of 26%, across 10 screening strategies (Table 2).



Harms and resource needs of screening

The numbers of mammograms per 1,000 40-year-old women estimated by OncoSim-Breast for all screening strategies were comparable to those estimated by the CISNET breast models (Table 1). However, the number of false-positive recalls projected by OncoSim-Breast were lower than those projected by the CISNET breast models for most screening scenarios, and the number of benign biopsies was lower across all scenarios (Table 1). For example, OncoSim-Breast estimated that biennial screening from ages 40 to 74 results in 16,618 mammograms, 1,250 false positives, and 111 benign biopsies per 1,000 women, compared with a median of 16,092 mammograms, 1,540 false positives, and 210 benign biopsies per 1,000 women estimated by the CISNET models.  

Ranking of the screening strategies

The incremental benefit-to-harm graphs, which plot the incremental benefits of screening against the harms and resource needs, demonstrate that, despite the differences in the absolute values, the rankings of the screening strategies are comparable between OncoSim-Breast and the CISNET models in most cases (Charts 2 to 5). 

Description of Chart 2



Data table for Chart 2
Table summary
This table displays the results of Data table for Chart 2. The information is grouped by Screening strategies (appearing as row headers), OncoSim-Breast, Cancer Intervention and Surveillance Modeling Network models, median estimate, Breast cancer deaths averted and Number of mammograms, calculated using per 1,000 women units of measure (appearing as column headers).

Screening strategies	OncoSim-Breast		Cancer Intervention and Surveillance Modeling Network models, median estimate
	Breast cancer deaths averted	Number of mammograms	Breast cancer deaths averted	Number of mammograms
	per 1,000 women
B50-74	10.4	11,698	6.9	11,192
B45-74	11.2	13,747	7.8	13,283
B40-74	12.1	16,618	8.4	16,092
A45-49, B50-74	11.9	16,570	8.6	15,992
A45-54, B55-74	12.1	18,551	8.8	18,006
A40-49, B50-74	12.8	21,508	9.3	20,898
A50-74	13.2	22,426	9.2	21,439
A45-74	14.6	27,296	10.4	26,272
A40-74	15.5	32,234	11.0	31,178
Notes: A: annual; B: biennial. Numbers indicate age ranges. Sources: OncoSim-Breast and Cancer Intervention and Surveillance Modeling Network breast cancer model outputs.

Description of Chart 3



Data table for Chart 3
Table summary
This table displays the results of Data table for Chart 3. The information is grouped by Screening strategies (appearing as row headers), OncoSim-Breast, Median Cancer Intervention and Surveillance Modeling Network model, Breast cancer deaths averted and Number of FPs, calculated using per 1,000 women units of measure (appearing as column headers).

Screening strategies	OncoSim-Breast		Median Cancer Intervention and Surveillance Modeling Network model
	Breast cancer deaths averted	Number of FPs	Breast cancer deaths averted	Number of FPs
	per 1,000 women
B50-74	10.4	905	6.9	1,021
B45-74	11.2	1,049	7.8	1,230
B40-74	12.1	1,250	8.4	1,540
A45-49, B50-74	11.9	1,246	8.6	1,416
A45-54, B55-74	12.1	1,384	8.8	1,514
A40-49, B50-74	12.8	1,592	9.3	1,896
A50-74	13.2	1,653	9.2	1,543
A45-74	14.6	1,994	10.4	1,943
A40-74	15.5	2,340	11.0	2,423
Notes: A: annual; B: biennial. Numbers indicate age ranges. Sources: OncoSim-Breast and Cancer Intervention and Surveillance Modeling Network breast cancer model outputs.

Description of Chart 4



Data table for Chart 4
Table summary
This table displays the results of Data table for Chart 4. The information is grouped by Screening strategies (appearing as row headers), OncoSim-Breast, Cancer Intervention and Surveillance Modeling Network models, median estimate, LYs gained and Number of mammograms, calculated using per 1,000 women units of measure (appearing as column headers).

Screening strategies	OncoSim-Breast		Cancer Intervention and Surveillance Modeling Network models, median estimate
	LYs gained	Number of mammograms	LYs gained	Number of mammograms
	per 1,000 women
B50-74	158.7	11,698	114.6	11,192
B45-74	187.4	13,747	140.0	13,283
B40-74	211.6	16,618	170.1	16,092
A45-49, B50-74	202.0	16,570	151.3	15,992
A45-54, B55-74	210.8	18,551	159.3	18,006
A40-49, B50-74	232.5	21,508	178.9	20,898
A50-74	206.5	22,426	153.2	21,439
A45-74	249.3	27,296	187.3	26,272
A40-74	279.7	32,234	208.7	31,178
Notes: A: annual; B: biennial. Numbers indicate age ranges. Sources: OncoSim-Breast and Cancer Intervention and Surveillance Modeling Network breast cancer model outputs.

Description of Chart 5



Data table for Chart 5
Table summary
This table displays the results of Data table for Chart 5. The information is grouped by Screening strategies (appearing as row headers), OncoSim-Breast, Cancer Intervention and Surveillance Modeling Network models, median estimate, QALYs gained and Number of mammograms, calculated using per 1,000 women units of measure (appearing as column headers).

Screening strategies	OncoSim-Breast		Cancer Intervention and Surveillance Modeling Network models, median estimate
	QALYs gained	Number of mammograms	QALYs gained	Number of mammograms
	per 1,000 women
B50-74	102.8	11,698	80.7	11,192
B45-74	123.9	13,747	97.5	13,283
B40-74	140.2	16,618	118.8	16,092
A45-49, B50-74	132.7	16,570	104.7	15,992
A45-54, B55-74	138.3	18,551	109.8	18,006
A40-49, B50-74	152.8	21,508	122.2	20,898
A50-74	130.0	22,426	104.2	21,439
A45-74	159.4	27,296	125.0	26,272
A40-74	179.5	32,234	138.0	31,178
Notes: A: annual; B: biennial. Numbers indicate age ranges. Sources: OncoSim-Breast and Cancer Intervention and Surveillance Modeling Network breast cancer model outputs.

Discussion

When evaluating the impact of starting breast cancer screening at age 40 or 45, compared with age 50, OncoSim-Breast estimated breast cancer outcomes that were comparable to those estimated by the CISNET breast cancer models. This provides additional confidence that OncoSim-Breast’s projections of the impact of starting screening at ages younger than 50 are reasonable because they are consistent with more established models.

When assessing the effects of screening compared with no screening, OncoSim-Breast estimated slightly higher benefits in terms of breast cancer deaths averted and LYs gained than the median of the CISNET models. These discrepancies are likely influenced by the differences in inputs used to develop OncoSim-Breast and the CISNET breast cancer models. First, age-adjusted incidence of female breast cancer in the United States is higher than that in Canada, while age-adjusted breast cancer mortality rates are slightly higher in Canada than in the United States.^Note 18, ^Note 19 Moreover, life expectancy for women in Canada differs substantially from that for women in the United States. According to the World Health Organization, life expectancy for women in Canada was 83.9 years at birth and 26.2 years at age 60 in 2019,^Note 20 while life expectancy for women in the United States was 81.0 years at birth and 24.4 years at age 60 in 2019.^Note 20 Age is a primary risk factor for breast cancer, so the longer life expectancy in Canada implies a greater opportunity to avert breast cancer deaths with screening in Canada, compared with the United States. Also, averting a breast cancer death at age 60 results in approximately two additional years of life saved for Canadian women, compared with their United States counterparts. Another potential factor contributing to a higher level of screening benefit estimated by OncoSim-Breast is the use of country-specific survival inputs for OncoSim-Breast and the CISNET models. In addition, the sensitivity of digital mammography and age-specific general population health-related quality of life scores differ between the two countries.^Note 10, ^Note 12, ^Note 21 However, QALYs saved because of screening, as estimated by OncoSim-Breast, were comparable to those estimated by the CISNET breast models for all screening scenarios, providing additional confidence in using OncoSim-Breast to inform guideline development.

Resource needs associated with screening, as measured by the number of mammograms projected by OncoSim-Breast, were comparable to those projected by the CISNET breast models for all screening scenarios. This is expected and provides further verification for OncoSim-Breast. However, the estimates of OncoSim-Breast and the CISNET models for the harms of screening, including false-positive recalls and benign biopsies, differed significantly, likely because of the major differences in the breast cancer screening practices between the United States and Canada. Previous studies reported that the recall rates after screening mammography are substantially higher in the United States, compared with those in Canada, and this is likely the reason for major discrepancies in the number of false-positive recalls and benign biopsies.^Note 22

Despite the differences in benefit and harm outcomes estimated by OncoSim-Breast and the CISNET breast models, the rankings of screening strategies for various harm and benefit outcomes are mostly consistent (Charts 2 to 5). This observation provides additional confidence that OncoSim-Breast’s projections to compare alternative screening policies are consistent with the CISNET breast cancer models. In particular, when the number of mammograms is used as the measure of resource needs, most policies that appear on the efficiency frontiers estimated by OncoSim-Breast also appear on those estimated by CISNET, regardless of the benefit metric (Charts 2, 4 and 5).

To the authors’ knowledge, no previous study has compared the projections of breast cancer models on the performance of various breast cancer screening strategies for Canada and the United States. Multiple CISNET publications have provided an example for cross-model validation for breast cancer in the United States.^Note 14, ^Note 23, ^Note 24, ^Note 25, ^Note 26, ^Note 27, ^Note 28 The authors are unaware of a cross-model validation study that compares breast cancer models developed for different countries. A recent study used OncoSim-Breast to estimate the performance of five screening strategies in Canada.^Note 29 While the harms of screening estimated by the previous study matched those estimated by the present study, substantial differences exist in the estimated benefits. In particular, the previous study significantly underestimated the breast cancer deaths averted and LYs gained because of screening. This is attributable to the differences in the simulation settings used in the two studies. The previous study simulated breast cancer events only until 2051, ignoring outcomes after age 75. By contrast, the present study simulated women until death, providing an estimation of the lifetime benefits of screening.

A major limitation of the present study is that OncoSim-Breast was not modified using the inputs of the CISNET breast cancer models to identify the sources of the differences between the projections made by OncoSim-Breast and the CISNET breast cancer models—this is left for future research. To alleviate this limitation, information about key potential sources of these disparities was provided. Furthermore, such an experiment would still not affect the conclusions of the present study regarding the level of agreement in the findings based on the separate models. Moreover, this study focused on comparing models from only two countries. The use of established models developed for additional countries would provide more robust and comprehensive information, and this is also left for future research. Another limitation of this study is that while the models include many stochastic components, the runs comparing OncoSim-Breast and the CISNET breast cancer models were conducted in a deterministic way instead of a probabilistic manner. This is primarily because of the computational difficulty of conducting a probabilistic analysis for all major model inputs, including natural history, survival, and screening test performance. 

In conclusion, this comparative study demonstrates that OncoSim-Breast’s projections of the impact of extending mammography screening to women aged 40 to 49 on long-term breast cancer outcomes are consistent with those from more established models in the United States.