This phase 3 study (ClinicalTrials.gov identifier: NCT02579759, December 2015–March 2018) was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. A total of thirty centers in Belgium, Canada, France, Germany, the Netherlands, Spain, the United Kingdom and the United States were involved. Ethical approval was obtained from institutional review boards or ethics committees. All subjects provided written informed consent.
Overview
Eligible subjects were men or women aged 16–65 years, with a genetically proven diagnosis of CMT1A of mild to moderate severity (i.e., a CMT Neuropathy Score version 2 [CMTNS-v2] score of 2–18), clinically confirmed muscle weakness in at least the foot dorsiflexors, and motor nerve conduction of at least 15 m/s in the ulnar nerve. Subjects diagnosed with any other type of concomitant peripheral neuropathy, such as diabetic neuropathy, another significant neurological disease or major systemic disease were excluded. Full eligibility criteria are provided in Additional file 3: Table S3.
After a screening period of up to four weeks, eligible subjects were randomly allocated in a 1:1:1 ratio to receive high-dose PXT3003, low-dose PXT3003 or placebo. Randomization was done using a block scheme stratified by study center, and study treatments were numbered according to a material randomization list, which was separate from the subject randomization list. The randomization code, which was not available to the study centers, the Sponsor or the subjects, was created by an independent statistician who was not involved in any other activity related to the study. All treatments were liquid formulations indistinguishable by look or taste and packaged in identical kits.
After randomization, subjects entered the treatment phase. To improve tolerability, subjects began by taking a half dose (2.5 mL) orally twice daily for two weeks and then escalating to 5 mL (full dose) orally twice daily for the remainder of the study. The treatment period lasted for up to 15 months. To avoid excluding subjects who completed a significant part of the treatment period, a functional measurement at the 12-month visit was sufficient for a subject to have an evaluable endpoint. The mean of measurements at the 12- and 15-month visits was used as the endpoint for subjects who completed the full 15 months of treatment.
During the study, subjects were asked to attend up to six visits (V1 to V6) (Fig. 4) after the initial screening visit (V0), during which informed consent was obtained and demographic data, medical history as well as baseline disease characteristics were recorded. Eligibility criteria were evaluated at both V0 and at the randomization visit (V1). Physical and neurological examinations, nerve conduction studies, vital signs (blood pressure and heart rate) measurements and CMTNS-v2 evaluations were performed at V0, V1, V3, V5 and V6. In V1, V3, V5 and V6, subjects were assessed using the ONLS, nine-hole peg test, 10MWT, quantified muscle testing (dorsiflexion and grip strength), and the EQ-5D-5L quality of life questionnaire, which included a visual analog scale ranging from “the best health you can imagine” to “the worst health you can imagine.” Safety was assessed at each treatment visit (V1–V6), including physical and neurological examinations, electrocardiography, clinical laboratory tests, vital signs measurements and recording of adverse events.
Objectives and outcomes
The primary objective of this study was to evaluate the efficacy of the two PXT3003 doses compared to placebo. The primary efficacy endpoint was the treatment effect on disability improvement as measured by the change in the ONLS total score from baseline the end of treatment. The ONLS was selected as the primary endpoint given that regulatory authorities require the use of clinical outcome measures for the primary endpoint. Furthermore, the 136th European Neuromuscular Committee meeting in 2006 [22] recommended the ONLS as the core disability scale and as a secondary endpoint, even though it discouraged the use of disability scales as primary endpoint measures. The ONLS, which ranges from 0 (no disability) to 12 (maximum disability), was originally proposed as an improved form of the ODSS [19] and is considered a reliable disability scale for patients with CMT with acceptable inter- and intra-rater reliability [22,23,24]. Secondary efficacy endpoints included functional tests (10MWT and nine-hole peg test) as well as the CMTNS-v2 Sensory score (sum of items 1, 4 and 5: sensory symptoms, pinprick sensibility and vibration) and Examination score (sum of items 1–7: sensory symptoms, motor symptoms [legs], motor symptoms [arms], pinprick sensibility, vibration, strength [legs] and strength [arms]). Exploratory efficacy endpoints included ONLS arm and leg subitems, the odds of a subject’s ONLS score improving or remaining stable in response to treatment, quantified muscular testing, electrophysiological parameters and quality of life measures. These physiological parameters were chosen as exploratory endpoints as they have been shown to change as CMT1A progresses [25].
The secondary objective was to evaluate the safety and tolerability of the two PXT3003 doses compared to placebo. Safety and tolerability were assessed by the incidence and severity of TEAEs, their relationship to the study drug, the incidence of TEAEs that led to withdrawal of study drug, the incidence of serious adverse events and changes in physical and neurological examinations, vital signs, 12-lead electrocardiograms and laboratory data. Treatment compliance was assessed by counting the number of bottles dispensed and the number of used bottles returned.
Statistical methods
All statistical analyses, performed by DICE NV using Statistical Analysis System (SAS) version 9.4 or later, were conducted in accordance with the statistical analysis plan, which was written before database lock and study unblinding. Treatment group allocation data were stored in a dedicated file sent to the analysts after database lock.
Considering the slow progression of CMT1A and the limited duration of a clinical trial, a 0.3-point mean change in ONLS score over the course of the study was pre-specified as clinically meaningful based on the results from the phase 2 study [18] and a Cohen’s d-value calculation [26, 27]. A sample size of 89 subjects per group should make such a difference detectable versus placebo with a power of 90%. Because a monotonic dose effect was not assumed, each dose was compared at a 2.5% significance level to preserve an overall false–positive of 5%. The dropout rate was anticipated to be 10% [18, 28] meaning that 100 subjects needed to be recruited to each treatment arm and at least 300 subjects needed to be recruited in total.
The primary efficacy analysis was performed using a linear mixed model analysis of covariance on the summary mean of ONLS scores at 12 and 15 months adjusted for baseline ONLS scores. Baseline ONLS scores and treatment were the fixed factors and center was admitted as a random effect. Missing data was to be imputed using a multiple imputation method based on the placebo group distribution. Secondary and exploratory endpoints were analyzed analogously to the primary endpoint.
Two sensitivity analyses were also planned. In the first sensitivity analysis, the primary analysis was performed on the full analysis set, the completers and per protocol populations (Fig. 1) to check for any significant variation in treatment effect by population. The second sensitivity analysis used a longitudinal model without imputation instead of an analysis of covariance in determining efficacy. The longitudinal model hypothesized a linear time–effect relationship and used the actual time of visits to estimate differences in the slope of ONLS evolution in the randomized treatment arms. A sensitivity analysis (described in Additional file 4: Table S4) investigating the impact of subjects’ age and sex on the treatment effect was also conducted.
During the study, the high-dose PXT3003 formulation was found to have a stability issue as crystals formed inside some bottles. As a result of this intercurrent event, all subjects in Germany who had not already completed the study were discontinued (n = 62), as required by the German regulatory authorities, as were all ongoing subjects in the high-dose PXT3003 group (n = 59) in all other countries, following the sponsor’s decision. Although the crystallization did not cause additional safety concerns, it had a significant impact on the subject discontinuation rate. Because a large proportion of subjects discontinued the study due to the intercurrent event rather than for reasons related to outcome, an mFAS was defined before database lock. Defining an mFAS was in line with the intent-to-treat approach as it included all subjects in the full analysis set (FAS, i.e., all randomized subjects) except for those who discontinued the study for reasons not related to clinical outcome (e.g., adverse event). An adjudication committee, consisting of two independent clinical neurologists blinded to both the treatment and the cause leading to discontinuation, assessed each discontinuation and classified them as outcome-related or outcome-unrelated. If consensus was not reached within the committee, the discontinuation was assumed outcome-related and included in the mFAS. The mFAS included 25/62 subjects discontinued due to the German regulatory authorities’ decision and 18/59 subjects discontinued following the sponsor’s decision. Outcome-related missing data were imputed as anticipated using a linear model with a multiple imputation method based on the placebo group distribution, while outcome-unrelated missing data in the FAS was imputed based on the distribution in the subject’s treatment arm.
The sensitivity analyses defined in the study protocol were modified before database lock and unblinding to account for the use of the mFAS as the primary analysis population. The FAS was added as a population to the first sensitivity analysis to investigate the bias arising from the use of the mFAS as the primary analysis population. The second sensitivity analysis, which was expected to correct for the variation by group in the treatment duration that the unexpectedly high discontinuation rate caused, was performed on both the mFAS and FAS. Because the longitudinal model used more data points than the analysis of covariance, and time–effect relationship data from the prior phase 2 study appeared to be linear, it was also expected to have greater statistical power.
The original analysis plan foresaw the use of the Dunnett procedure to correct for multiple testing. However, as a result of the intercurrent event and difficulties in using this procedure with multiple imputation in the longitudinal model, a Bonferroni correction based on unadjusted p-values and 97.5% confidence intervals was used instead. As Bonferroni corrections are more conservative than the Dunnett procedure, this change did not increase the type I error rate [29, 30].
Safety was analyzed using descriptive statistics and Fisher’s exact test without correction for multiple testing (and therefore no control of type I error rate). All randomized subjects who had received at least one dose of the study drug were included in the analysis population.