Supplementary webappendix This webappendix formed part of the original submission and has been peer reviewed. We post it as supplied by the authors. Supplement to: Giovannoni G, Gold R, Selmaj K, et al, for the SELECTION Study Investigators. Daclizumab high-yield process in relapsing-remitting multiple sclerosis (SELECTION): a multicentre, randomised, double-blind extension trial. Lancet Neurol 2014; published online March 19. http://dx.doi.org/10.1016/s1474-4422(14)70039-0.
Methods (additional information) Eligibility criteria Eligible patients were those who participated for at least 52 weeks and were compliant in the SELECT study. Patients were required to practice effective contraception during the study and to be willing and able to continue contraception for 4 months after the last dose of study medication. Exclusion criteria included any significant change in medical status from SELECT that would preclude administration of daclizumab high-yield process (HYP); permanent discontinuation of treatment during SELECT (except patients unblinded during evaluation of an adverse event and found to be receiving placebo); planned or ongoing treatment with any approved or experimental treatment for MS (except for protocol-allowed use of interferon beta); current enrolment in any investigational drug study other than SELECTION; and unwillingness or inability to comply with requirements of the study protocol. 1
Statistical analyses Unless otherwise stated, efficacy (ie, clinical and MRI endpoints) was assessed in the intention-to-treat population with the exception of patients who had a delay (>55 days) in starting SELECTION after the last daclizumab HYP dose in SELECT. The intention-to-treat population included all patients who underwent randomisation, except for 18 patients from a single study centre who were prospectively excluded due to detection of systematic misdosing at that centre during trial monitoring. Statistical analyses were done by treatment group and by combined dose groups. 2
Clinical endpoints The effect of daclizumab HYP treatment on the annualised relapse rate (ARR) during years 1 and 2 (SELECT and SELECTION, respectively) was estimated using a negative binomial regression model adjusting for the number of relapses in the previous year. The ARR was calculated for each treatment group and combined dose group to assess treatment differences. Relapse rates were based on relapses that were confirmed by the independent neurology committee. Relapses that occurred after rescue treatment with alternative MS medication were excluded from the analyses, and the patient s time on study was censored at the time of starting the alternative MS medication. To assess within-treatment differences, the ARR during years 1 and 2 were compared by a generalised estimating equation (GEE) using a Poisson model adjusted for the number of relapses in the year before entry into SELECT (main effects only). The Poisson distribution was used for this analysis, as there were problems with convergence when the negative binomial distribution was used. The logarithmic transformation of time during the period was included in the model as an offset parameter and the model was adjusted for over-dispersion using a Pearson scale parameter. An exchangeable correlation structure was used in this analysis. For year 1, confirmed disability progression was defined as a 1 0-point increase on the Expanded Disability Status Score (EDSS) during year 1 from a baseline EDSS 1 0 that was sustained for 3 months, or a 1 5-point increase on the EDSS from a baseline EDSS of 0 that was sustained for 3 months; baseline values were at entry to SELECT. For year 2, confirmed disability progression was defined as a 1 0-point increase on the EDSS during year 2 from a baseline EDSS 1 0 that was sustained for 3 months or a 1 5-point increase on the EDSS during year 2 from a baseline EDSS of 0 that was sustained for 3 months; baseline values were at entry to SELECTION. A cut-off of 74 days was used to determine sustained progression for 3 months since patients were allowed a 5-day window for their visit schedule. The estimated percentage of patients who were relapse-free and the estimated percentage of patients with sustained progression were estimated from the Kaplan-Meier survival distribution. The proportion of patients with sustained progression for 3 months during each of years 1 and 2 were compared by a GEE using a binomial distribution, adjusted for the EDSS score at baseline of SELECT (main effects only). An exchangeable correlation structure was used. The proportion of patients who relapsed during each of years 1 and 2 were compared by a GEE using a binomial distribution, adjusted for number of relapses at baseline of SELECT (main effects only). An exchangeable correlation structure was used. 3
MRI endpoints The numbers of cumulative new gadolinium lesions between week 52 of the SELECT study (baseline of SELECTION) and week 104 (end of treatment period in SELECTION) were assessed by a GEE using a Poisson distribution, adjusted for the number of gadolinium lesions at baseline to SELECT (main effects only). The Poisson distribution was used as there were convergence problems with the negative binomial distribution. Differences in the number of new or newly enlarging T2 hyperintense lesions between week 52 of the SELECT study and week 104 were assessed by a GEE using a negative binomial distribution, adjusted for age and number of T2 lesions at baseline to SELECT (main effects only). The GEE analyses were adjusted for over-dispersion using a Pearson scale parameter. An exchangeable correlation structure was used. Differences in the percentage change in total volume of T2 hyperintense lesions, volume of new T1 hypointense lesions and percentage change in total volume of T1 hypointense lesions between week 52 of the SELECT study and week 104 were assessed by ranking the data and performing Friedman s two-way nonparametric ANOVA because the data were not normally distributed. The rate of percentage change in brain volume from baseline of SELECT to week 52 of the SELECT study was compared with the rate of percentage change from week 52 of the SELECT study to week 104. The analysis was done using a repeated measures mixed model adjusted for normalised brain volume at baseline to SELECT (main effects only). For all MRI endpoints, missing data were imputed using the mean within the treatment group. 4