key: cord-0291984-pbnn5oy5 authors: Yong, J. H. E.; Nadeau, C.; Flanagan, W.; Coldman, A.; Asakawa, K.; Garner, R.; Fitzgerald, N.; Yaffe, M.; Miller, A.; Group, OncoSim-Breast Working title: The OncoSim-Breast cancer microsimulation model date: 2020-05-24 journal: nan DOI: 10.1101/2020.05.22.20110569 sha: 35d689040adaed0f1ffde727416dff2daccab813 doc_id: 291984 cord_uid: pbnn5oy5 Background: The increasing demand for health care resources requires measures to evaluate the impact of cancer control approaches. A cancer simulation model can help integrate new knowledge to inform clinical and policy decisions. OncoSim-Breast is a breast cancer simulation model. This paper aims to describe the key assumptions in the OncoSim-Breast model and how well it reproduces more recent breast cancer trends and the observed effects in a randomized screening trial. Methods: The OncoSim-Breast model simulates the onset, growth and spread of invasive and ductal carcinoma in situ tumours. The model combines Canadian cancer incidence, mortality, screening program and cost data to project population-level outcomes. Users can change the model input to answer specific policy questions. Here we report three validation exercises. First, we compared the model's projected breast cancer incidence and stage distributions with the observed data in the Canadian Cancer Registry. Second, we compared OncoSim's projected breast cancer mortality with Vital Statistics. Third, we replicated the UK Age trial to compare the model's projections with the trial's observed screening effects. Results: OncoSim-Breast's projected incidence, mortality and stage distribution of breast cancer were close to the observed data in the Canadian Cancer Registry and the Vital Statistics. OncoSim-Breast also reproduced the breast cancer screening effects observed in the UK Age trial. Interpretation: OncoSim-Breast's ability to reproduce the observed population-level breast cancer trends and the screening effects in a randomized trial increases the confidence of using its results to inform policy decisions related to early detection of breast cancer. Rapidly emerging knowledge in breast cancer control has put pressure on the health system for the adoption of new technologies and policies. Randomized trials are the gold standard of evidence to introduce new interventions in clinical practice and public health; however, such evidence is not always relevant for informing policy decisions because the context of the interventions evolves quickly compared to the time that elapses between the design of a trial and the availability of its results. For example, most breast cancer screening randomized trials were from the era before breast cancer adjuvant treatment was available and used film-screen mammography1,2; breast cancer survival has since vastly improved3 and digital mammography has superseded film-screen mammography. A cancer simulation model can help integrate evidence from multiple sources and make them more relevant to inform contemporary clinical and policy decisions. Several groups have developed sophisticated cancer-specific models based on the natural history of cancer that can be revised for additional analyses and incorporate knowledge from experts in different areas4. OncoSim is an example of a cancer simulation model. The validation and applications of OncoSim colorectal, cervical, and lung cancers have been described previously5-19. Breast cancer is the latest addition to OncoSim's suite of cancer models. The primary objective of OncoSim-Breast is to investigate emerging issues related to breast cancer control in Canada. This work builds on a strong foundation of analyses performed over a decade ago to estimate the impact of diagnostic and therapeutic approaches to non-metastatic breast cancer in Canada, using the Statistics Canada POHEM mathematical model20. The present paper has two goals. First, it aims to describe the key assumptions in OncoSim-Breast. Secondly, it compares OncoSim-Breast's projections with observed data in the Canadian Cancer Registry, projected breast cancer mortality estimates in the Canadian Vital Statistics, and the observed screening effects in a randomized trial. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 24, 2020. The model inputs have several components: • Demography is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05. 22.20110569 doi: medRxiv preprint The model combines these inputs to project outcomes, such as breast cancer incidence and mortality, screening outcomes, stage and age at diagnosis, life-years, quality-adjusted lifeyears, lifetime breast cancer costs, and screening or follow-up procedure costs ( Figure 1 ). OncoSim simulates one individual at a time, replicating the age and sex distributions, and allcause mortality of the population in each province and territory in Canada (Supplemental Appendix 1). Each simulated individual has attributes, such as demography (sex, province/territory), and breast cancer-related risk factors (BRCA1/2 gene mutation, family history, exposure to hormone replacement therapy, Supplemental Table 1 ). OncoSim-Breast simulates the onset, growth and spread of tumours, both invasive cancer and Tumour onset: In OncoSim-Breast, tumours start from 2mm, based on the probable minimum size detectable by mammography screening and similar to the Wisconsin Breast model. Probability of tumour onset varies by age and years (Supplemental Figure 1 ). In addition, the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint risk increases if a woman has any of the breast cancer-related risk factors (BRCA1/2 mutation, family history of breast cancer or exposure to hormone replacement therapy, Supplemental Tables 3-4) ; if a woman has previously had a DCIS tumour, she is also more likely to have an invasive cancer. Larger tumours are more likely to get detected clinically than smaller tumours (Supplementary Table 7 For evaluating screening strategies or related performance, the model allows users to create different screening strategies and scenarios by modifying the following input parameters (Supplemental Appendix 4): . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint • Screening program recruitment strategy (e.g. start/end age and years) • Screening participation and retention • Screening frequency • Screening modality (e.g. digital mammography) • Sensitivity and specificity of screening Table 9 ). Oncosim-Breast captures the benefits of screening on breast cancer survival using lead time calibrated from observed survival data (Supplemental Figure 3 ). Upon cancer detection, the model estimates the survival outcomes based on stage, tumour biology, age at diagnosis, and detection method; in the case of disease progression, it draws time to recurrence and possible breast cancer death (Supplemental Appendix 5). Stage-specific recurrence risks and breast cancer survival outcomes were estimated using data from the British Columbia Cancer Agency because comprehensive staging data only became available recently in the Canadian Cancer Registry. To capture provincial differences in stage-specific . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint survival, the model applies province-specific relative risks, estimated from more recent data in the Canadian Cancer Registry, to the British Columbia survival curves. The model included healthcare costs associated with breast cancer from the perspective of a public healthcare payer, such as the Ministry of Health (Supplemental Appendix 6). The costs included breast cancer surgery, radiation treatment, chemotherapy, imaging tests and oncology physician fees, acute hospitalizations, emergency department visits, home care, long-term care, complex continuing care, and others. The model captures lifetime costs of breast cancer across three phases of care (first 18 months after diagnosis, continuing care and terminal care), a similar approach used in other established breast cancer simulation models.3 To calculate quality-adjusted life-years, the model multiplies the duration of each health state with age and sex-specific preference scores for the Canadian population27 and breast cancerspecific health state utilities28 (upon cancer diagnosis) (Supplemental Appendix 7). When an individual is in several health states at the same time, we assumed the utility score is multiplicative.29 We validated our model in three ways. First, we compared the projected incidence and stage distribution of breast cancer in Canada with the observed data in the Canadian Cancer Registry is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05. 22.20110569 doi: medRxiv preprint The UK Age trial is a randomized trial that compared annual screening in women aged 40-49 years with usual care in the UK in the 1990s31. Other established breast cancer simulation models have validated their predictions against the UK trial results34. We simulated a cohort of women born in 1950-1957 to match the birth cohort in the UK Age trial in two scenarios: (i) no screening; and (ii) annual screening for women age 40-49. In the screening scenario, we calibrated the rescreening rate to the average number of mammograms per woman in the Age trial (4.8)33. For each scenario, we estimated the incidence of breast cancer and breast cancer deaths in women aged 40-49 years. We then compared OncoSim Breast's projected incidence of breast cancer (DCIS and invasive cancers) with the trial's mean estimate and its 95% confidence interval. For breast cancer mortality, we compared the mortality reduction ratio from OncoSim Breast with the trial's mean estimate and 95% confidence interval over a 17 years follow-up period. We chose to compare rate ratios rather than rates because the populations were different: volunteers in the UK Age trial vs. Canadian population. OncoSim's projected incidence, stage distribution and mortality of breast cancer were close to the observed data in the Canadian Cancer Registry and the Vital Statistics32. Figures 2A and 2B compare OncoSim's projected incidence rates of breast cancer (invasive cancer and DCIS) by age group in 1992-2013 with the observed data in the Canadian Cancer Registry. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05. 22.20110569 doi: medRxiv preprint In our external validation exercise simulating the UK Age trial, OncoSim projected that annual breast cancer screening in women age 40-49 years would detect 16% more invasive breast cancers vs. 10% (95% CI: 0.95-1.21) observed in the UK Age trial34. When estimating the impact of screening on breast cancer deaths, OncoSim's projection was also within the confidence interval of the observed trial estimate. Over a 17 years follow-up period, OncoSim estimated that annual breast cancer screening in women age 40-49 years had an 85% relative risk of breast cancer death (i.e. 15% reduction), as compared with 88% (95% CI: 74%-104%) This paper has several limitations. First, OncoSim is a simulation model built using the best available data; the accuracy of projections depends on the quality of data input and the validity of assumptions. To address the issue of rapidly emerging evidence, OncoSim-Breast allows users to modify the inputs and assumptions. Second, our validation exercise comparing OncoSim-Breast's projections with more recent Canadian Cancer Registry data was limited by the availability and quality of data in the Registry. Third, our simulation of the UK Age trial was an exploratory external validation exercise; we did not calibrate the model to match historically poorer breast cancer outcomes at that time. Fourth, OncoSim-Breast was built to be a multipurpose breast cancer simulation tool and could simulate many scenarios; therefore, it would not be feasible to validate all its possible projections against observed data. To ensure OncoSim-Breast's relevance for supporting policy decisions, the team compares OncoSim-Breast's projections with emerging real-world data and refines the model based on new evidence, on an ongoing basis. In the upcoming releases, examples of further enhancements will include adding emerging data on new screening modalities, costs of new treatments, and . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint other factors that might affect screening performance, such as breast density and polygenic risk scores. OncoSim-Breast is a natural history-based simulation model developed using Canadian cancer incidence, mortality, screening program and cost data. It reproduces breast cancer trends in the Canadian Cancer Registry, breast cancer mortality in the Vital Statistics, and the breast cancer screening effects observed in a randomized screening trial. OncoSim is led and supported by the Canadian Partnership Against Cancer, with model development by Statistics Canada, and is made possible through funding provided by Health Canada. We thank the many individuals who have contributed to the conceptualization, development, review, and application of OncoSim (see https://www.partnershipagainstcancer.ca/tools/oncosim/acknowledgements/). In particular, we would like to acknowledge the substantial contribution from other members of the OncoSim-Breast technical working group (full list is in Supplemental Appendix 8). Also, we would like to thank Dr. James Mainprize for his contribution in the Age trial validation exercise. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint The Canadian National Breast Screening Study-1: breast cancer mortality after 11 to 16 years of follow-up: a randomized screening trial of mammography in women age 40 to 49 years Reduced breast cancer mortality after 20+ years of follow-up in the Swedish randomized controlled mammography trials in Malmö Influence of tumour stage at breast cancer detection on survival in modern times: population based study in 173 797 patients Introduction to the Cancer Intervention and Surveillance Modeling Network (CISNET) breast cancer models Projected effect of fecal immunochemical test threshold for colorectal cancer screening on outcomes and costs for Canada using the OncoSim microsimulation model Simulating results from trials of sigmoidoscopy screening using the OncoSim microsimulation model Using the Cancer Risk Management Model to evaluate colorectal cancer screening options for Canada Costs and Cost-Effectiveness of Smoking Cessation Within an Organized CT Lung Cancer (LC) Screening Program Potential Health and Economic Consequences of Organized vs Opportunistic Lung Cancer Screening in Canada Implementing low-dose computed tomography screening for lung cancer in Canada: implications of alternative at-risk populations, screening frequency, and duration Cancer Cancer Risk Management Lung Cancer W. Eligibility for low-dose computerized tomography screening among asbestos-exposed individuals Performance of the cancer risk management model lung cancer screening module The OncoSim model: development and use for better decision-making in Canadian cancer control Biennial lung cancer screening in Canada with smoking cessation -outcomes and costeffectiveness International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted Costeffectiveness of Lung Cancer Screening in Canada Exploring the health outcomes of various pan-Canadian cervical cancer screening programs using microsimulation modeling Evaluation of the natural history of cancer of the cervix, implications for prevention. The Cancer Risk Management Model (CRMM) -Human papillomavirus and cervical components Comparing the health and economic impacts of cervical cancer screening strategies using the Cancer Risk Management Model (CRMM) Using the Cancer Risk Management Model to evaluate the health and economic impacts of cytology compared with human papillomavirus DNA testing for primary cervical cancer screening in Canada The Population Health Model (POHEM): an overview of rationale, methods and applications Preventive Health Care. Recommendations on screening for colorectal cancer in primary care Preventive Health Care. Recommendations on screening for lung cancer Canadian Cancer Society's Advisory Committee on Cancer Statistics Special topic: predictions of the future burden of cancer in Canada The University of Wisconsin Breast Cancer Epidemiology Simulation Model: An Update Age-and sex-specific Canadian utility norms, based on the 2013-2014 Canadian Community Health Survey Population health impact of cancer in Canada Deriving utility scores for comorbid conditions: a test of the multiplicative model for combining individual condition scores International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted Malignant neoplasm of breast (C50) Effect of mammographic screening from age 40 years on breast cancer mortality at 10 years' follow-up: a randomised controlled trial Effect of mammographic screening from age 40 years on breast cancer mortality in the UK Age trial at 17 years' followup: a randomised controlled trial. The lancet oncology Comparing CISNET Breast Cancer Incidence and Mortality Predictions to Observed Clinical Trial Results of Mammography Screening from Ages 40 to 49 Randomised controlled trial of mammographic screening in women from age 40: predicted mortality based on surrogate outcome measures International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity Toronto, ON; 2017.. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint Figure 2A . Incidence of invasive breast cancer (per 100,000 women) by age group in 1992- . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint Figure 2B . Incidence of ductal carcinoma in situ (per 100,000 women) by age group in 1992- . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 24, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110569 doi: medRxiv preprint