|
Abstract:
|
Non-inferiority (NI) clinical trials test whether a new treatment is unacceptably worse than an existing treatment. Missing data in clinical trials are known to lead to reduction in statistical power and potentially biased estimates: tests of NI hypotheses in the presence of missing data have additional issues. For one, missing data in NI studies tend to decrease sensitivity to differences between treatment groups and can bias toward concluding the alternative hypothesis of non-inferiority. We conduct a systematic review of NI trials published May 2015 to April 2016 and report the amount of missing data and how they are handled (model-based methods, imputation, complete case). We find the majority of NI trials report less than 10% missing data in the primary outcome. Twenty five percent of trials reviewed contain 15% or more missing data. Most researchers use complete case analysis, and only 7% conducted a sensitivity analysis to test assumptions with respect to missing data. We discuss the value of sensitivity analyses, reporting results for both Intention-To-Treat and Per-Protocol analysis sets, and the implications of missing data for the constancy assumption.
|
