Treatment Effect Estimation for Randomized Clinical Trials Subject to Noncompliance and Missing Outcomes.

Missing Data;Clinical Trials;Treatment Switching;Noncompliance;Masked Missing Not at Random;Missing Not at Random;Public Health;Statistics and Numeric Data;Health Sciences;Science;Biostatistics

Kang, ShanTaylor, Jeremy M. ;

University of Michigan

关键词: Missing Data;    Clinical Trials;    Treatment Switching;    Noncompliance;    Masked Missing Not at Random;    Missing Not at Random;    Public Health;    Statistics and Numeric Data;    Health Sciences;    Science;    Biostatistics;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/110464/shankang_1.pdf?sequence=1&isAllowed=y

瑞士|英语

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

【 摘 要 】

Noncompliance and missing outcomes are common in randomized clinical trials. In this dissertation, we explore treatment arm switching issues for survival data and nonrandom dropout issues for masked clinical trials.In Chapter 2, we consider noncompliance in phase III clinical trials in oncology. Although patients are randomized to their treatment assignments, the option of treatment switching may be offered to patients who experience disease recurrence for ethical considerations. Standard methods that ignore this nonrandom noncompliance can lead to bias. Although methods do exist to account for the effect of treatment switching, several of these methods focus on quantifying an overall switching effect, which can still lead to biased results if the benefit derived from switching varies among patients. We propose a new parametric method to address this limitation that factorizes the likelihood into two parts in order to evaluate the individual benefit of switching. A more robust latent event time approach is also proposed. In simulation studies, our proposed methods outperform the existing methods.In Chapter 3, we consider missing outcome problems in masked clinical trials. Most standard models for analyzing the data make the missing at random (MAR) assumption, but in practice, there are often situations where MAR is not valid. For masked trials, we propose a specific missing not at random assumption, which we call masked MNAR (MMNAR): since the specific treatment received is unknown, missingness does not depend on treatment assignment after conditioning on outcomes and side effects. We suggest that methods based on MMNAR are useful for masked clinical trials, either in their own right or to provide a sensitivity analysis for deviations from MAR. We formulate two models under this assumption. Simulations show that our proposed methods outperform other methods when MAR is violated and the efficiency of treatment effect estimates is similar to that of MAR methods when MAR is true. We apply our methods to the TROPHY study.In Chapter 4, we develop regression-based multiple imputation models that exploit the MMNAR assumption proposed in Chapter 3 for longitudinal data. Simulation studies are conducted and the idea is also illustrated with the TROPHY study.