Page 22 -
P. 22
Chapter 4
Sample Size
The sample size calculation is based on the hypothesis that the short-term results obtained after IPD are superior to the results obtained after surgical decompression. The ZCQ at eight weeks will be used as a primary result measure both to answer the first research question and to calculate the sample size. The sample size of the trial is based on a superiority design and calculated under the alternative null- hypothesis to reach sufficient power to enable a distinction between the two arms in terms of success according to ZCQ if (according to the literature) results obtained after surgical decompression will be 64% and the results obtained after IPD will be at least 84% (20% difference in favour of IPD). A sample size of 98 patients per group ensures 90% power to confirm the null hypothesis when IPD is more than 20% superior to decompression, using a likelihood ratio test in a logistic regression framework (see figure 1: sample size). Accounting for about 10% loss to follow-up, this trial will enrol 216 patients with INC (108 patients in both groups). A sample size of 80 patients per group (including 10% loss to follow-up) will ensure a power of 80%. The feasibility of reaching 216 patients available for analysis will be checked after reaching 160 evaluable patients without deblinding or even analysing the data as a group comparison. This constitutes a methodological valid approach since no multiple testing is involved and stopping further accrual is not based on an intermediate effect estimate. Since the power is based on a dichotomization of the underlying ZCQ scale, an alternative primary analysis of the ZCQ itself will also have sufficient power. The latter analysis will also take the repeated measurements structure into account.
Statistical analysis
Baseline comparability will be assessed by descriptive statistics to determine whether randomization was successful. Differences in outcome between both groups, together with 95% confidence intervals, will be calculated.
Besides a difference in recovery between the two groups at two specified time points (eight weeks and one year), analysis of a difference in time to recovery will be carried out as well, using a survival analysis framework (COX hazards). All data are analysed according to the “intention-to-treat-principle”. Furthermore a repeated measurements analysis of variance will be performed on the underlying continuous scales. In all analyses the first assessment of treatment effect will be the estimate of the main effect within the appropriate model, adjusted for the
62
