A complex systems model of breast cancer etiology: The Paradigm II Model

Abstract

Background Complex systems models of breast cancer have previously focused on prediction of prognosis and clinical events for individual women. There is a need for understanding breast cancer at the population level for public health decision-making, for identifying gaps in epidemiologic knowledge and for the education of the public as to the complexity of this most common of cancers. Methods and findings We developed an agent-based model of breast cancer for the women of the state of California using data from the U.S. Census, the California Health Interview Survey, the California Cancer Registry, the National Health and Nutrition Examination Survey and the literature. The model was implemented in the Julia programming language and R computing environment. The Paradigm II model development followed a transdisciplinary process with expertise from multiple relevant disciplinary experts from genetics to epidemiology and sociology with the goal of exploring both upstream determinants at the population level and pathophysiologic etiologic factors at the biologic level. The resulting model reproduces in a reasonable manner the overall age-specific incidence curve for the years 2008–2012 and incidence and relative risks due to specific risk factors such as BRCA1, polygenic risk, alcohol consumption, hormone therapy, breastfeeding, oral contraceptive use and scenarios for environmental toxin exposures. Conclusions The Paradigm II model illustrates the role of multiple etiologic factors in breast cancer from domains of biology, behavior and the environment. The value of the model is in providing a virtual laboratory to evaluate a wide range of potential interventions into the social, environmental and behavioral determinants of breast cancer at the population level.

Making evidence go further: Advancing synergy between agent-based modeling and randomized control trials

“…ABM, in turn, can strongly benefit from partnership with RCTs to overcome important limitations of its own. Although ABM has seen rapid uptake along with other complex systems science tools across the health sciences ( 6 , 8 , 15 – 26 ), data availability often constrains potential applications. This is because an important strategy for testing and improving ABM is to compare output from the computational simulation to real-world empiric data taken from observational or experimental sources.…”

“…Use of ABM to extrapolate across contexts requires a solid representation of the mechanistic structure of a system. Where such structure is not already well understood, ABM can be used to build theory by identifying hypothesized structures and testing them using existing evidence ( 6 , 7 , 14 , 15 , 24 , 25 ). Resultant models can be used with suitable caution to simulate potential real-world intervention effects ( 7 , 16 , 17 ).…”