Simulation modeling to enhance comparative effectiveness research in cancer

Abstract

Policy-makers need compelling evidence of the relative benefits and harms or "comparative effectiveness" (CE) of new interventions in order to guide clinical practice and prioritize future research. In cancer control, the primary indicator of intervention efficacy is reduction in cancer deaths. However, CE studies often do not follow subjects to the time of death due to time and resource constraints. Even when CE studies do reach the time of death, the results may quickly become obsolete due to advances in technology that occur during the long study period. Models can integrate data from various studies to project a coherent picture of disease progression from intervention to death. The goal of this research is to construct a novel modeling framework to translate existing CE results into their implications for deaths prevented in three cancer control settings: 1) using diagnostic tests to target treatment 2) using new biomarkers to enhance screening 3) screening under evolving treatment technology. In each setting, we build a transparent model that can integrate data from multiple studies to project deaths prevented. We then apply the model to answer an important current question in that setting. In the first setting, we evaluate whether a new diagnostic test for breast cancer used to identify patients most likely to benefit from treatment actually changes the risk of cancer death by changing the rate at which patients get treated. In the second setting, we consider whether a new biomarker for prostate cancer that changes the sensitivity and specificity of prostate cancer screening actually changes the risk of prostate cancer death and overdiagnosis. In the third setting, we address whether advances in breast cancer treatments over the last two decades have implications for interpreting screening mammography trials conducted before these treatments were available. In each case, we use published data from a CE study as a starting point and build on this to extrapolate the observed CE results into a projection of deaths prevented.