A Bayesian Simulation Model for Breast Cancer Screening, Incidence, Treatment, and Mortality

Background. The important but complicated research questions regarding the optimization of mammography screening for the detection of breast cancer are unable to be answered through any single trial or a simple meta-analysis of related trials. The Cancer Intervention and Surveillance Network (CISNET) breast groups provide answers using complex statistical models to simulate population dynamics. Among them, the MD Anderson Cancer Center (Model M) takes a unique approach by not making any assumptions on the natural history of breast cancer, such as the distribution of the indolent time before detection, but simulating only the observable part of a woman's disease and life. Methods. The simulations start with 4 million women in the age distribution found in the year 1975, and follow them over several years. Input parameters are used to describe their breast cancer incidence rates, treatment ecacy, and survival. With these parameters, each woman's history of breast cancer diagnosis, treatment, and survival are generated and recorded each year. Research questions can then be answered by comparing the outcomes of interest, such as mortality rates, quality-adjusted life years, number of false positives, differences between hypothetical scenarios, such as different combinations of screening and treatment strategies. We use our model to estimate the relative contributions of screening and treatments on the mortality reduction in the United States, for both overall and different molecular (ER, HER2) subtypes of breast cancer. Results. We estimate and compare the benefits (life-years gained) and harm (false-positives, over-diagnoses) of mammography screening strategies with different frequencies (annual, biennial, triennial, mixed) and different starting (40 and 50 years) and end ages (70 and 80 years). Conclusions. We will extend our model in future studies to account for local, regional, and distant disease recurrences.