Abstract:
|
We have previously described a suite of statistical data models to identify data entry errors and inform risk-based monitoring. Including the treatment assignment in the models should account for variability thereby providing greater power to detect outliers. However, it carries the risk of unblinding the treatment assignment, being applied differentially across the groups, telegraphing results to clinical investigators and disturbing equipoise. We conducted a simulation study to investigate the effects of either including or excluding treatment assignment. Different combinations of equal/unequal means and variances as well as outliers differentially affecting one group more than another were considered. Linear models that excluded treatment, included treatment as a covariate, or conducted separate analyses by treatment group were compared. Evaluation criteria included sensitivity, specificity, bias in treatment estimates and difference, and power to detect a treatment difference. With some exceptions, including treatment generally resulted in better statistical properties. However, safeguards are crucial to ensure all results are kept confidential at the statistical center.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.