To Screen or Not To Screen: Prevention Decisions and Competing Risks

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Routine cancer screening can lead to early detection of disease, thereby prevent morbidity associated with more advanced disease and reduce mortality. Many older Americans, however, have comorbid medical conditions that present competing risks; because many benefits of screening, such as reduced mortality, accrue in the future, persons with shorter life expectancies are less likely to realize the benefits. Current guidelines, however, provide recommendations based only on age, and only for some ages in large part because of the paucity of relevant trial data. Moreover, it may be infeasible to conduct randomized controlled trials that are able to address when individuals should consider stopping routine screening because of advanced age or increasing numbers of comorbid conditions. The absence of clear recommendations or rigorous evidence arguably contributes to the difficulty that patients and their physicians have when making these decisions. Using large, observational datasets from the Medicare program, and causal inference approaches, we propose to emulate clinical trials of mammography screening for breast cancer, and of colonoscopy screening for CRC. Our 2002-12 data from a 20% random sample of Medicare fee-for-service beneficiaries (Parts A, B, and D) contain detailed, longitudinal information on screening tests, cancer diagnoses, deaths, medical spending, and important covariates. Our three aims examine the impact of screening on outcomes for subjects age 65+ who have varying numbers of comorbid medical conditions: 1) cancer detection; 2) survival; and 3) medical spending. We will use a series of approaches including traditional survival and repeated measures analyses, and more recently developed techniques for causal analysis with observational data, e.g., Marginal Structural Models with inverse probability weighting. We also will explore additional approaches using instrumental variables and g-formula models. While the methodology supporting causal inference has advanced considerably in recent years, its use in clinical, policy, and comparative effectiveness research is only starting. With over 22 million person-years of data, we will have adequate power to detect clinically relevant differences in our outcomes. With detailed longitudinal data, we will adjust for a rich set of demographic, clinical, physician, geographic, and insurance characteristics. This study provides arguably the best opportunity to evaluate alternative cancer screening strategies among older Americans with varying numbers of comorbid medical conditions and thus inform policy makers, clinicians, and patients.