Mehmet Ali Ergun

Collaborative Modeling of the Benefits and Harms Associated With Different U.S. Breast Cancer Screening Strategies.

Context Multiple alternative mammography screening strategies exist. Contribution This modeling study estimated outcomes of 8 strategies that differed by starting age and interval. Biennial screening from age 50 to 74 years avoided a median of 7 breast cancer deaths; in contrast, annual screening from age 40 to 74 years avoided an additional 3 deaths but yielded 1988 more false-positive results and 11 more overdiagnoses per 1000 women screened. Annual screening from age 40 years for high-risk women had similar outcomes as screening average-risk women biennially from 50 to 74 years of age. Caution Imaging technologies other than mammography and nonadherence were not modeled. Implication Biennial mammography screening for breast cancer is efficient for average-risk women. Despite decades of mammography screening for early detection of breast cancer, there is no consensus on optimal strategies, target populations, or the magnitude of harms and benefits (111). The 2009 US Preventive Services Task Force (USPSTF) recommended biennial film mammography from age 50 to 74 years and suggested shared decision making about screening for women in their 40s (12). Since that recommendation was formulated, new data on the benefits of screening have emerged (2, 6, 8, 9, 11, 13, 14), digital mammography has essentially replaced plain film (15), and increasingly effective systemic treatment regimens for breast cancer have become standard (16). There has also been growing interest in consumer preferences and personalized screening approaches (1720). These factors could each affect the outcomes of breast cancer screening programs or alter policy decisions about population screening strategies (17). Modeling can inform screening policy decisions because it uses the best available evidence to evaluate a wide range of strategies while holding selected conditions (such as treatment effects) constant, facilitating strategy comparisons (21, 22). Modeling also provides a quantitative summary of outcomes in different groups and assesses how preferences affect results. Collaboration of several models provides a range of plausible effects and illustrates the effects of differences in model assumptions on results (1, 7, 23). We used 6 well-established simulation models to synthesize current data and examine the outcomes of digital mammography screening at various starting ages and intervals among average-risk women. We also examined how breast density, risk, or comorbidity levels affect results and whether preferences for health states related to screening and its downstream consequences affected conclusions. Methods Strategies We evaluated 8 strategies that varied by starting age (40, 45, or 50 years) and interval (annual, biennial, and hybrid [annual for women in their 40s and biennial thereafter]); all strategies stop screening at age 74 years. We included no screening as a baseline. Model Descriptions The models used to evaluate the screening strategies were developed within the Cancer Intervention and Surveillance Modeling Network (CISNET) (2430), and the research was institutional review boardapproved. They were named model D (Dana-Farber Cancer Institute, Boston, Massachusetts), model E (Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands), model GE (Georgetown University Medical Center, Washington, DC, and Albert Einstein College of Medicine, Bronx, New York), model M (MD Anderson Cancer Center, Houston, Texas), model S (Stanford University, Stanford, California), and model W (University of Wisconsin, Madison, Wisconsin, and Harvard Medical School, Boston, Massachusetts). The Appendix provides information on model validation. Since earlier analyses (1), the models have undergone substantial revision to reflect advances in breast cancer control, including portrayal of molecular subtypes based on estrogen receptor (ER) and human epidermal growth factor-2 receptor (HER2) status (23); current population incidence (31) and competing nonbreast cancer mortality; digital screening; and the most current therapies (32). All models except model S include ductal carcinoma in situ (DCIS). The general modeling approach is summarized in this article; full details, including approach, construction, data sources, assumptions, and implementation, are available at https://resources.cisnet.cancer.gov/registry and reference 33. Additional information is available on request, and the models are available for use via collaboration. The models begin with estimates of breast cancer incidence (31) and ER/HER2-specific survival trends without screening or adjuvant treatment, and then overlay data on screening and molecular subtypespecific adjuvant treatment to generate observed incidence and breast cancerspecific mortality trends in the U.S. population (1, 7, 17, 23, 33, 34). Breast cancer has a distribution of preclinical screen-detectable periods (sojourn time) and clinical detection points. Performance characteristics of digital mammography depend on age, first versus subsequent screen, time since last mammogram, and breast density. ER/HER2 status is assigned at diagnosis on the basis of stage and age. Molecular subtype and stage-specific treatment reduces the hazard of breast cancer death (models D, GE, M, and S) or results in a cure for some cases (models E and W). Women can die of breast cancer or other causes. Screen detection of cancer during the preclinical screen-detectable period can result in the identification and treatment of earlier-stage or smaller tumors than might occur via clinical detection, with a corresponding reduction in breast cancer mortality. We used a cohort of women born in 1970 with average-risk and average breast density and followed them from age 25 years (because breast cancer is rare before this age [0.08% of cases]) until death or age 100 years. Model Input Parameters The models used a common set of age-specific variables for breast cancer incidence, performance of digital mammography, treatment effects, and average and comorbidity-specific nonbreast cancer causes of death (20, 35). The parameter values are available at www.uspreventiveservicestaskforce.org/Page/Document/modeling-report-collaborative-modeling-of-us-breast-cancer-1/breast-cancer-screening1 (33). In addition, each group included model-specific inputs (or intermediate outputs) to represent preclinical detectable times, lead time, and age- and ER/HER2-specific stage distribution in screen- versus nonscreen-detected women on the basis of their specific model structure (1, 7, 2330). These model-specific parameters were based on assumptions about combinations of values that reproduced U.S. trends in incidence and breast cancerspecific mortality, including proportions of DCIS that were nonprogressive and would not be detected without screening. Models M and W also assumed some small nonprogressive invasive cancers. The models adopted an ageperiodcohort modeling approach to project incidence rates of breast cancer in the absence of screening (31, 36); model M used 19751979 rates from the Surveillance, Epidemiology, and End Results program. The models assumed 100% adherence to screening and receipt of the most effective treatment to isolate the effect of varying screening strategies. Four models used age-specific sensitivity values for digital mammography that were observed in the Breast Cancer Surveillance Consortium (BCSC) for detection of invasive and DCIS cancers combined (model S uses data for invasive cancers only). Separate values were used for initial and subsequent mammography by screening interval, using standard BCSC definitions: Annual includes data from screens occurring within 9 to 18 months of the prior screen, and biennial includes data on screens within 19 to 30 months (37, 38). Model D used these data as input variables (28), and models GE, S, and W used the data for calibration (24, 25, 27). Models E and M fit estimates from the BCSC and other data (26, 29). Women with ER-positive tumors received 5 years of hormone therapy and an anthracycline-based regimen accompanied by a taxane. Women with ER-negative invasive tumors received anthracycline-based regimens with a taxane. Those with HER2-positive tumors also received trastuzumab. Women with ER-positive DCIS received hormonal therapy (16). Treatment effectiveness was based on clinical trials and was modeled as a reduction in breast cancerspecific mortality risk or increase in the proportion cured compared with ER/HER2-specific survival in the absence of adjuvant treatment (32). Benefits Screening benefits (vs. no screening or incremental to other strategies) included percentage of reduction in breast cancer mortality, breast cancer deaths averted, and life-years and quality-adjusted life-years (QALYs) gained because of averted or delayed breast cancer death. Benefits (and harms) were accumulated from age 40 to 100 years to capture the lifetime effect of screening. We considered preferences, or utilities, to account for morbidity from screening and treatment. A disutility for age- and sex-specific general population health was first applied to quality-adjust the life-years (39). These were further adjusted to account for additional decrements in life-years related to undergoing screening (0.006 for 1 week, or 1 hour), evaluating a positive screen (0.105 for 5 weeks, or 3.7 days), undergoing initial treatment by stage (for the first 2 years after diagnosis), and having distant disease (for the last year of life for all women who die of breast cancer) (Appendix Table 1) (33, 40, 41). Appendix Table 1. Utility Input Parameter Values Use and Harms Use of services focused on the number of mammograms required for the screening strategy. Harms included false-positive mammograms, benign biopsies, and overdiagnosis. Rates of false-positive mammograms were calculated as mammograms read as abnormal or needing further work-up in women without cancer divided by the total number of screening

Effects of Screening and Systemic Adjuvant Therapy on ER-Specific US Breast Cancer Mortality

BACKGROUND Molecular characterization of breast cancer allows subtype-directed interventions. Estrogen receptor (ER) is the longest-established molecular marker. METHODS We used six established population models with ER-specific input parameters on age-specific incidence, disease natural history, mammography characteristics, and treatment effects to quantify the impact of screening and adjuvant therapy on age-adjusted US breast cancer mortality by ER status from 1975 to 2000. Outcomes included stage-shifts and absolute and relative reductions in mortality; sensitivity analyses evaluated the impact of varying screening frequency or accuracy. RESULTS In the year 2000, actual screening and adjuvant treatment reduced breast cancer mortality by a median of 17 per 100000 women (model range = 13-21) and 5 per 100000 women (model range = 3-6) for ER-positive and ER-negative cases, respectively, relative to no screening and no adjuvant treatment. For ER-positive cases, adjuvant treatment made a higher relative contribution to breast cancer mortality reduction than screening, whereas for ER-negative cases the relative contributions were similar for screening and adjuvant treatment. ER-negative cases were less likely to be screen-detected than ER-positive cases (35.1% vs 51.2%), but when screen-detected yielded a greater survival gain (five-year breast cancer survival = 35.6% vs 30.7%). Screening biennially would have captured a lower proportion of mortality reduction than annual screening for ER-negative vs ER-positive cases (model range = 80.2%-87.8% vs 85.7%-96.5%). CONCLUSION As advances in risk assessment facilitate identification of women with increased risk of ER-negative breast cancer, additional mortality reductions could be realized through more frequent targeted screening, provided these benefits are balanced against screening harms.

The University of Wisconsin Breast Cancer Epidemiology Simulation Model: An Update:

The University of Wisconsin Breast Cancer Epidemiology Simulation Model (UWBCS), also referred to as Model W, is a discrete-event microsimulation model that uses a systems engineering approach to replicate breast cancer epidemiology in the US over time. This population-based model simulates the lifetimes of individual women through 4 main model components: breast cancer natural history, detection, treatment, and mortality. A key feature of the UWBCS is that, in addition to specifying a population distribution in tumor growth rates, the model allows for heterogeneity in tumor behavior, with some tumors having limited malignant potential (i.e., would never become fatal in a woman’s lifetime if left untreated) and some tumors being very aggressive based on metastatic spread early in their onset. The model is calibrated to Surveillance, Epidemiology, and End Results (SEER) breast cancer incidence and mortality data from 1975 to 2010, and cross-validated against data from the Wisconsin cancer reporting system. The UWBCS model generates detailed outputs including underlying disease states and observed clinical outcomes by age and calendar year, as well as costs, resource usage, and quality of life associated with screening and treatment. The UWBCS has been recently updated to account for differences in breast cancer detection, treatment, and survival by molecular subtypes (defined by ER/HER2 status), to reflect the recent advances in screening and treatment, and to consider a range of breast cancer risk factors, including breast density, race, body-mass-index, and the use of postmenopausal hormone therapy. Therefore, the model can evaluate novel screening strategies, such as risk-based screening, and can assess breast cancer outcomes by breast cancer molecular subtype. In this article, we describe the most up-to-date version of the UWBCS.

Association of screening and treatment with breast cancer mortality by molecular subtype in US women, 2000-2012

Importance Given recent advances in screening mammography and adjuvant therapy (treatment), quantifying their separate and combined effects on US breast cancer mortality reductions by molecular subtype could guide future decisions to reduce disease burden. Objective To evaluate the contributions associated with screening and treatment to breast cancer mortality reductions by molecular subtype based on estrogen-receptor (ER) and human epidermal growth factor receptor 2 (ERBB2, formerly HER2 or HER2/neu). Design, Setting, and Participants Six Cancer Intervention and Surveillance Network (CISNET) models simulated US breast cancer mortality from 2000 to 2012 using national data on plain-film and digital mammography patterns and performance, dissemination and efficacy of ER/ERBB2-specific treatment, and competing mortality. Multiple US birth cohorts were simulated. Exposures Screening mammography and treatment. Main Outcomes and Measures The models compared age-adjusted, overall, and ER/ERBB2-specific breast cancer mortality rates from 2000 to 2012 for women aged 30 to 79 years relative to the estimated mortality rate in the absence of screening and treatment (baseline rate); mortality reductions were apportioned to screening and treatment. Results In 2000, the estimated reduction in overall breast cancer mortality rate was 37% (model range, 27%-42%) relative to the estimated baseline rate in 2000 of 64 deaths (model range, 56-73) per 100 000 women: 44% (model range, 35%-60%) of this reduction was associated with screening and 56% (model range, 40%-65%) with treatment. In 2012, the estimated reduction in overall breast cancer mortality rate was 49% (model range, 39%-58%) relative to the estimated baseline rate in 2012 of 63 deaths (model range, 54-73) per 100 000 women: 37% (model range, 26%-51%) of this reduction was associated with screening and 63% (model range, 49%-74%) with treatment. Of the 63% associated with treatment, 31% (model range, 22%-37%) was associated with chemotherapy, 27% (model range, 18%-36%) with hormone therapy, and 4% (model range, 1%-6%) with trastuzumab. The estimated relative contributions associated with screening vs treatment varied by molecular subtype: for ER+/ERBB2−, 36% (model range, 24%-50%) vs 64% (model range, 50%-76%); for ER+/ERBB2+, 31% (model range, 23%-41%) vs 69% (model range, 59%-77%); for ER−/ERBB2+, 40% (model range, 34%-47%) vs 60% (model range, 53%-66%); and for ER−/ERBB2−, 48% (model range, 38%-57%) vs 52% (model range, 44%-62%). Conclusions and Relevance In this simulation modeling study that projected trends in breast cancer mortality rates among US women, decreases in overall breast cancer mortality from 2000 to 2012 were associated with advances in screening and in adjuvant therapy, although the associations varied by breast cancer molecular subtype.

Effect of Time to Diagnostic Testing for Breast, Cervical, and Colorectal Cancer Screening Abnormalities on Screening Efficacy: A Modeling Study

Background: Patients who receive an abnormal cancer screening result require follow-up for diagnostic testing, but the time to follow-up varies across patients and practices. Methods: We used a simulation study to estimate the change in lifetime screening benefits when time to follow-up for breast, cervical, and colorectal cancers was increased. Estimates were based on four independently developed microsimulation models that each simulated the life course of adults eligible for breast (women ages 50–74 years), cervical (women ages 21–65 years), or colorectal (adults ages 50–75 years) cancer screening. We assumed screening based on biennial mammography for breast cancer, triennial Papanicolaou testing for cervical cancer, and annual fecal immunochemical testing for colorectal cancer. For each cancer type, we simulated diagnostic testing immediately and at 3, 6, and 12 months after an abnormal screening exam. Results: We found declines in screening benefit with longer times to diagnostic testing, particularly for breast cancer screening. Compared to immediate diagnostic testing, testing at 3 months resulted in reduced screening benefit, with fewer undiscounted life years gained per 1,000 screened (breast: 17.3%, cervical: 0.8%, colorectal: 2.0% and 2.7%, from two colorectal cancer models), fewer cancers prevented (cervical: 1.4% fewer, colorectal: 0.5% and 1.7% fewer, respectively), and, for breast and colorectal cancer, a less favorable stage distribution. Conclusions: Longer times to diagnostic testing after an abnormal screening test can decrease screening effectiveness, but the impact varies substantially by cancer type. Impact: Understanding the impact of time to diagnostic testing on screening effectiveness can help inform quality improvement efforts. Cancer Epidemiol Biomarkers Prev; 27(2); 158–64. ©2017 AACR.

Using Active Learning for Speeding up Calibration in Simulation Models

Background. Most cancer simulation models include unobservable parameters that determine disease onset and tumor growth. These parameters play an important role in matching key outcomes such as cancer incidence and mortality, and their values are typically estimated via a lengthy calibration procedure, which involves evaluating a large number of combinations of parameter values via simulation. The objective of this study is to demonstrate how machine learning approaches can be used to accelerate the calibration process by reducing the number of parameter combinations that are actually evaluated. Methods. Active learning is a popular machine learning method that enables a learning algorithm such as artificial neural networks to interactively choose which parameter combinations to evaluate. We developed an active learning algorithm to expedite the calibration process. Our algorithm determines the parameter combinations that are more likely to produce desired outputs and therefore reduces the number of simulation runs performed during calibration. We demonstrate our method using the previously developed University of Wisconsin breast cancer simulation model (UWBCS). Results. In a recent study, calibration of the UWBCS required the evaluation of 378 000 input parameter combinations to build a race-specific model, and only 69 of these combinations produced results that closely matched observed data. By using the active learning algorithm in conjunction with standard calibration methods, we identify all 69 parameter combinations by evaluating only 5620 of the 378 000 combinations. Conclusion. Machine learning methods hold potential in guiding model developers in the selection of more promising parameter combinations and hence speeding up the calibration process. Applying our machine learning algorithm to one model shows that evaluating only 1.49% of all parameter combinations would be sufficient for the calibration.

Using Collaborative Simulation Modeling to Develop a Web-Based Tool to Support Policy-Level Decision Making About Breast Cancer Screening Initiation Age

Background: There are no publicly available tools designed specifically to assist policy makers to make informed decisions about the optimal ages of breast cancer screening initiation for different populations of US women. Objective: To use three established simulation models to develop a web-based tool called Mammo OUTPuT. Methods: The simulation models use the 1970 US birth cohort and common parameters for incidence, digital screening performance, and treatment effects. Outcomes include breast cancers diagnosed, breast cancer deaths averted, breast cancer mortality reduction, false-positive mammograms, benign biopsies, and overdiagnosis. The Mammo OUTPuT tool displays these outcomes for combinations of age at screening initiation (every year from 40 to 49), annual versus biennial interval, lifetime versus 10-year horizon, and breast density, compared to waiting to start biennial screening at age 50 and continuing to 74. The tool was piloted by decision makers (n = 16) who completed surveys. Results: The tool demonstrates that benefits in the 40s increase linearly with earlier initiation age, without a specific threshold age. Likewise, the harms of screening increase monotonically with earlier ages of initiation in the 40s. The tool also shows users how the balance of benefits and harms varies with breast density. Surveys revealed that 100% of users (16/16) liked the appearance of the site; 94% (15/16) found the tool helpful; and 94% (15/16) would recommend the tool to a colleague. Conclusions: This tool synthesizes a representative subset of the most current CISNET (Cancer Intervention and Surveillance Modeling Network) simulation model outcomes to provide policy makers with quantitative data on the benefits and harms of screening women in the 40s. Ultimate decisions will depend on program goals, the population served, and informed judgments about the weight of benefits and harms.