Measures of Expected Influence Provide Useful Constraints to Enrollment in Randomized Multi-Center Clinical Trials for Binomial, Continuous and Time-to-Event Endpoints.  

作　　者：Shankar Srinivasan, Ph.D. Arlene Swern, Ph.D. 

机构地区：[1]Biometrics and Data Operations, Celgene Corporation

出　　处：《Journal of Statistical Science and Application》2015年第2期39-49,共11页统计科学与应用（英文版）

摘　　要：Avoiding excessive enrollment of a single cohort in a clinical trial is prudent in order to avoid imbalances and to prevent one cohort from having a disproportionate influence on the results of a trial and perhaps even negating positive findings of the clinical trial. Numerical criteria are provided here to evaluate the expected influence of a large cohort as a function of both its size and the relative effect of interventions, in comparison to those of other groups. Measures of expected influence are obtained as a function of the parameters of the distribution of statistics measuring influence. Calculated numerical criteria for the binomial, continuous and time-to-event contexts are presented. Details of the application of this method and sensitivity analyses conducted during the planning stages of a multiple myeloma clinical trial are provided. Numerical criteria are derived under asymptotic conditions and thus results hold for large cohorts. The numerical criteria are easy to compute and are useful tools to assess possible detrimental effects of large cohorts during the design of a study or during enrollment prior to any un-blinding. The numerical criteria allow for a-priori sensitivity analyses of the likely influence of large cohorts under varying conditions.

关 键 词：Influence of large sites large strata large cohorts scaled inflation in influence BINOMIAL TIME-TO-EVENT CONTINUOUS expected influence. 

分 类 号：C8[社会学—统计学]