Rank-based tests for non-inferiority and equivalence hypotheses in multi-centre clinical trials using mixed models

Rank-based procedures for testing non-inferiority and equivalence hypotheses for data (continuous) arising from multi-centre clinical trials are developed using mixed models. The centres are presumed to be a random sample from the population of centres which might conceivably use the treatments under investigation, and the treatment effects are assumed to be fixed. It is assumed that there are no interactions between treatments and centres. Non-inferiority and equivalence hypotheses are tested using rank (R) estimators. The R-estimators are obtained by minimizing a combined dispersion function. The combined dispersion function is a sum of separate dispersion functions from each centre. Both the combined dispersion function and the centre-specific dispersion functions are piecewise linear functions. As a result, the R-estimates are expected to be less influenced by outliers. Large sample properties of the R-estimators including efficiencies are developed. A small scale simulation study is conducted to compare the performances (empirical levels and powers) of the non-parametric procedures and the normal theory based procedures.