2.6. Sample size determination and statistical analysis
Following O’Brian and Fleming [25] the study was planned as a superiority trial and conducted according to a 3-stage group sequential design with pre-planned analyses after n = 32, 46 and 60 patients. The trial specific one-sided type I error rate was set at a = 2.5% (corresponding to a two-sided level of 5%), Consequently, we fixed the adjusted two-sided significance levels for the three analyses at a1 = 0.21%, a2 = 0.97% and a3 = 2.16% and defined to stop the trial for futility whenever the observed p-value exceeded 60%. We assumed that the overall pain would decrease by 24 ± 8 mm from baseline to day 7 in the leech therapy group and by 10 ± 8 mm in the diclofenac group. For this, a maximum sample size of 60 patients was calculated to achieve a power of 80% [40].
All outcome criteria were analysed by intention-to-treat analysis with repeated measurement analyses of covariance (ANCOVA) which took time as the within-subject factor, group as a between-subject factor, and the respective baseline value as a covariate. Missing data were multiply imputed following the suggestions of Little and Rubin [14]. In detail, we used the MCMC method of the MI procedure of the SAS/STAT software, imputed missing values for each treatment group separately, and created 20 different sets of data, analysed them separately with the above described ANCOVA models and combined the results with the SAS MIANALYZE procedure.
Only one interim analysis of the primary outcome criterion was conducted. This analysis resulted in a p-value of p = 0.0003 which was substantially smaller than a1 = 0.21% so that the study was stopped.
Ancillary analyses of the overall pain score were done to adjust for the effects of possibly confounding variables, namely outcome expectation. Here, we added these variables as covariates to the ANCOVA models and estimated the group differences in the presence of these covariates.