Breast cancer models overview
Six breast cancer modeling groups have been collaborating as part of CISNET since 2000, referred to as CISNET-Dana-Farber Cancer Institute (DFCI (Dana-Farber)), MIcrosimulation SCreening Analysis (MISCAN) Fatal Diameter model (MISCAN-Fadia (Erasmus)), Simulating Population Effects of Cancer Control Interventions - Race and Understanding Mortality Georgetown-Einstein (Spectrum/G-E), Bayesian Simulation Model (MD Anderson Cancer Center (MDACC)), Breast Cancer Outcomes Simulator (BCOS (Stanford)), and University of Wisconsin Breast Cancer Simulation Model (UWBCS (Wisconsin).
The models are designed to match breast cancer incidence and mortality rates observed in the Surveillance, Epidemiology and End Results (SEER) Program for U.S. women aged ≥ 25 beginning in 1975. Four models are microsimulations (MISCAN-Fadia, Bayesian Simulation Model, Spectrum/G-E, and UWBCS), one model uses an analytic approach (CISNET-DFCI), and the remaining model (BCOS) is a hybrid analytic/microsimulation. The microsimulation models include natural history components that approximate tumor progression in size and stage. While the six models employ different approaches, model structure, and underlying assumptions, the models share common inputs. These inputs include incidence rates in the absence of screening, mammography and treatment dissemination rates, mammography performance, and non-breast cancer mortality. Common inputs are based on observational data from a variety of sources, including: the National Health Interview Survey (NHIS), the SEER Program including Patterns of Care studies, the Breast Cancer Surveillance Consortium (BCSC), the Connecticut Tumor Registry, the Berkeley Mortality Database, and the National Center for Health Statistics (NCHS).
Each model has faced a central challenge to match SEER rates showing the dramatic increase in incidence in 1980s due to the widespread adoption of screening mammography. This rapid increase in incidence requires the availability of undiagnosed breast cancers in the models as well as accommodations for the phenomenon that breast cancer mortality largely remained stable throughout this period. CISNET-DFCI mathematically models a distribution of lead times in the presence of screening, and assumes that any survival gain from screening is a result of a change in the stage distribution because of early diagnosis. The MISCAN-Fadia model incorporates the concept of a fatal tumor diameter, where a woman can only be cured if her tumor is diagnosed (clinically or by screening) and treated before it reaches its unique fatal diameter. Screen-detected breast cancers tend to have slower growth rates. The Spectrum/G-E and BCOS models also incorporate the feature that screening preferentially detects slower growing tumors, but survival is determined through the assigned stage of cancer as a function of tumor size, so that there is no survival benefit beyond the stage shift attributable to screening. Conversely, the Bayesian Simulation Model assigns a benefit beyond stage shift so that screening not only can detect a tumor in an earlier stage than in the absence of screening, but an additional survival benefit of screening is implemented. The UWBCS model assigns a fraction of breast cancers, identified through calibration, with the traits of a tumor with “limited malignant potential,” such that these tumors never lead to death if left untreated. The MISCAN-Fadia and Spectrum/G-E models similarly assume that a portion of in situ tumors are non-progressive and do not result in death. Models also differ by whether treatment affects the hazard for death from breast cancer (CISNET-DFCI, Spectrum/G-E, Bayesian Simulation Model, and BCOS) or results in a cure for some fraction of breast cancer cases (MISCAN-Fadia and UWBCS).
Breast CISNET models have been extended to predict US breast cancer incidence and mortality as more years of data become available from the SEER Program. All models incorporate recent screening trends including the transition from film to digital mammography and adjuvant treatment trends including newer multi-agent chemotherapy regimens. Four of the six models historically included in situ as well as invasive breast cancer, and the remaining two models are now also modeling ductal carcinoma in situ (DCIS). Updates for all models include the addition of breast cancer subtypes defined by Estrogen Receptor (ER) and Human Epidermal Growth Factor Receptor 2 (HER2) with specific treatment regimens for each subtype. Race-specific models of breast cancer have also been developed to investigate questions surrounding disparities in incidence, mortality and survival.
Key areas of current investigation include: 1) evaluate the use of polygenetic breast cancer risk to guide breast cancer screening strategies; 2) assess breast cancer screening strategies involving tomosynthesis; 3) evaluate management strategies for screen-detected DCIS; 4) evaluate the impact of new molecular pathway and genomic tumor profile-targeted treatments. Current collaborative partners include the NCI-funded Breast Cancer Surveillance Consortium (BCSC), Genetic Associations and Mechanisms in Oncology (GAME-ON), the Canadian Partnership Against Cancer (CPAC), and the NCI-funded Population-based Research Optimizing Screening through Personalized Regimens (PROSPR) program. Additional projects with other partners are also in progress.