A CCEPTED MANUSCRIPT
degree polynomial and the total AUC was calculated by computing the integral. This approach was
taken for all VCE-obtained data.
Statistical tests were pre-defined and agreed in the statistical analyses plan finalised and signed
prior to unblinding of the randomization key. The randomization list was made, and the labelling of
trial product was performed by third parties not otherwise involved in the trial. No imputation of
data was carried out in cases of missing data, but all available data were used.
Subject characteristics and all efficacy data presented are based on the Full Analysis Set (FAS)
population. Criteria for inclusion in FAS was defined as maximum one missing visit in between the
randomization visit (visit 2) and end of trial (Visit 7). The safety reporting by listing of adverse
events included all subjects that were randomized (n=75).”
A sample size calculation was performed prior to trial initiation based on the primary endpoint of
the trial. The curve shapes were assumed to fit with a third-degree polynomial. We considered a
MANUSCRIPT
30% lower AUC following treatment of Bif195 compared to placebo to be clinically relevant and
aimed at a trial design that would have 80% power in detecting an intervention effects of this size as
statistical significant. No previously knowledge exists on AUC values and SD. Sample size
calculation was therefore performed on percent difference of AUC between two normalised curves
(Active vs. placebo) as an approximation. We assumed similar standard deviation in each arm and
planned for two-sided testing with a significance level of 5%. Given the above assumptions the
number of subjects needed in each arm was 30. To account for potential drop-out subjects, we
aimed to randomize a total of 75 subjects. Subjects who withdrew within one week of
randomization were replaced by standby-subjects.
In general, datasets were modelled as the dependent variable in a linear mixed model. The model
included the baseline value as covariate and gender and Bif195/placebo intervention as factors.
Model check was always assessed for all datasets using QQ residual plots together with
12
Powered by FlippingBook