Impact of Early Detection on Treatment Effects and Cancer Mortality.

Early Detection;Misspecified Model;Bias;Prostate Cancer;Mechanistic Statistical Modeling;Statistics and Numeric Data;Science;Biostatistics

Lee, Shih-YuanTaylor, Jeremy M. ;

University of Michigan

关键词: Early Detection;    Misspecified Model;    Bias;    Prostate Cancer;    Mechanistic Statistical Modeling;    Statistics and Numeric Data;    Science;    Biostatistics;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/77836/shihylee_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

The widespread use of early detection programs has led to profound changes in heterogeneity, prognosis, and the meaning of clinical variables for newly diagnosed cancer patients.All this made an assessment of treatment benefit and the combined effect of early detection (screening) and treatment a challenge.Screening trials that randomize screening vs. no screening and then assess mortality differences are expensive, require huge sample size, suffer from contamination as control arm patients get screened, and generally do not generalize to populations under a different pattern of screening. Clinical trials assessing survival post-diagnosis suffer from heterogeneity of patients induced by screening, including lead-time and length biases, and the fact that they condition on clinical characteristics at diagnosis, the latter being part of the response variable under screening.With this thesis we pursue a mechanistic statistical modeling approach to explain the dynamics of the population and subject-specific effects induced by screening and use the model to provide an unbiased assessment of the combined effect of screening and treatment interventions that would be generalizable to populations under arbitrary screening patterns, provided they are known.In Chapter 2, we assess how the lead time induced by screening affects the treatment effect as estimated by a proportional hazards model. We investigate the behavior of the estimator in the Cox proportional hazards model using partial likelihood and propose a meta-analytic approach to correct the bias based on a joint cancer incidence and survival modeling approach. In Chapters 3 and 4, we propose an analytic joint statistical model based on the natural history of the disease and its interaction with screening to assess the latent chronic disease characteristics. We apply the model to US prostate cancer and analyze it using the Surveillance, Epidemiology and End Results (SEER) data. The natural history of the cancer, the estimated treatment effects, and the predicted mortality are the focus of our study.In our final chapter, we summarize the strengths and the limitations of our current approach and discuss the direction for future research.