Statistical methods

Three hundred and fifty participants will be required to detect at least a 20 % difference in the percentage of successes at 18 months post-randomisation (where success is defined as a device in use and ≥50 % CCIS improvement from baseline) between FENIX^TM MSA and SNS at 5 % level of significance, 90 % power, assuming approximately 40 % success on the SNS arm and allowing for 20 % loss to follow-up.

Analyses will be performed on an intention-to-treat (ITT) basis (primary analysis), where participants will be included according to the surgical procedure they were randomised to, and by actual treatment group, where participants will be included according to the surgery actually received (SNS device or FENIX^TM MSA device implantation). All hypothesis tests will be two-sided and use a 5 % significance level.

Analyses will exclude training cases, although data collected on training cases will be summarised. Analysis and reporting will be in line with Consolidated Standards of Reporting Trials (CONSORT) guidelines [26]. For the primary analysis, multi-level logistic regression will be used, including adjustment for the factors included in the minimisation algorithm.

Secondary endpoints including SF-12®, EQ-5D-5L®, CCIS and OD-score recorded at baseline and at 6, 12 and 18 months post-randomisation will be analysed using random effects (multi-level) models to account for the hierarchical nature of repeated measures data. The models will include adjustments for minimisation factors, and a categorical covariate will be used to assess the effect of length of time of device in use on these endpoints.

Pattern-mixture multi-level models, which will treat all participant data observed after the removal of their device (explant) as missing data, but also account for the informative nature of the missing data, will be fitted to the secondary endpoints outlined above. Note that this is in contrast to the random effects models outlined above, which incorporate data from participants ‘post-explant’. Therefore, the results yielded by the pattern-mixture multi-level models will act as sensitivity analyses, which can be used to explore the potential issue of disparity in treatment of participants post-explant in each treatment arm.

A subgroup analysis will be performed on participants in the FENIX^TM arm in order to explore which potential patients could benefit most from FENIX^TM. A multi-level logistic regression model will be fitted using the primary endpoint, and the effects of various patient-level covariates (e.g. age, gender, baseline quality of life) on the odds of ‘success’ will be assessed.

Data collected on the safety of FENIX^TM MSA and SNS will be analysed using multi-level logistic regression. No formal interim analyses are planned; hence, no statistical testing will take place until final analysis.