Primary analysis of time to first confirmed Expanded Disability Status Scale score progression

Primary analysis using a Cox regression model showed no evidence of an effect of age (p = 0.36), disease type (p = 0.12), sex (p = 0.56), weight (p = 0.11) or treatment (p = 0.57; see Table 8) on time to confirmed EDSS score progression. The HR for first EDSS score progression event in patients randomly assigned to dronabinol compared with those assigned to placebo was 0.92 [95% confidence interval (CI) 0.68 to 1.23; see Table 8]. At trial completion, Kaplan–Meier estimates of the probability of EDSS score progression were 0.55 (95% CI 0.46 to 0.63) in the dronabinol group compared with 0.60 (95% CI 0.44 to 0.71) in the placebo group (Figure 3).

FIGURE 3

Kaplan–Meier estimates of the probability of progression on the EDSS in the two treatment groups, confirmed after 6 months within the study period. The numbers at risk (cumulative number of censored observations) are given. Those patients who (more...)

We noted evidence of some study site effects and of an effect of baseline EDSS score on time to confirmed progression (Figure 4). Most notably, relative to a baseline EDSS score of 4.0, there was an increased hazard of disease progression among those with a baseline EDSS score of 5.5 (HR 3.17, 95% CI 1.45 to 6.93; p = 0.004) and a reduced hazard among those with a baseline EDSS score of 6.5 (HR 0.49, 95% CI 0.24 to 0.98; p = 0.04). However, the numbers of participants in the individual EDSS groups are small (Figure 5), as are the numbers in some study sites (see Table 16 and Appendix 3).

FIGURE 4

Estimated HRs and 95% CIs for EDSS score progression from a Cox regression model. HRs for study sites are compared with site 01; HR for SPMS is compared with PPMS; active compared with placebo; each baseline EDSS is compared with a baseline EDSS score (more...)

FIGURE 5

Kaplan–Meier estimates of the probability of progression on the EDSS, by baseline EDSS score, confirmed after 6 months within the study period. The numbers at risk (cumulative number of censored observations) are given. Those patients who were (more...)

TABLE 16

Premature discontinuations of trial medication and losses to follow-up, by study site

The global PH test gave no evidence that the PH assumption was violated (χ² = 36, 36 degrees of freedom; p = 0.47).

Sensitivity analyses of time to first confirmed Expanded Disability Status Scale score progression

Results of sensitivity analysis showed that when losses to follow-up were treated as progression events rather than censored observations, the estimated HR (active : placebo) for EDSS score progression changed to 1.11 (95% CI 0.86 to 1.44; see Table 8), but the estimated effect of treatment remained non-significant (p = 0.41). This change in HR might be because the dronabinol group had a higher proportion of losses to follow-up for EDSS assessment [56 of 71 (79%)] than the placebo group [15 of 71 (21%)] and represents a worst-case scenario in terms of patient deterioration and hence the potential benefit of dronabinol.

The EPC reviewed data on 95 patients [71 active (74.7%); 24 placebo (25.3%)], for which there were ambiguities regarding EDSS scores. The EPC considered 22 (12 active; 10 placebo) of these patients to have progressed. These patients had no confirmed progression according to the data collected from the trial schedule. A further four patients (three active; one placebo) were considered to have progressed prior to the time of progression determined from the trial schedule. Clinical information on the remaining 69 patients reviewed by the EPC either confirmed non-progression or was insufficient to draw any further conclusions over those made on the primary data. As a result, data derived following EPC review consisted of a total of 240 first progression events compared with 218 in the primary data (with losses to follow-up considered as censored observations in both).

Conclusions from the main analyses of time to first EDSS score progression were robust to sensitivity analyses in terms of whether or not conclusions from the EPC were considered in defining EDSS progressions under both approaches to dealing with losses to follow-up, that is treated as censored observations or as progression events (see Table 8).

Furthermore, estimated HRs (active : placebo) for EDSS score progression remained similar after sequential removal of study sites with high loss to follow-up rates, under each of the two ways of treating losses to follow-up and each of the two data sets, that is according to trial schedule or following EPC review (Figure 6).

FIGURE 6

Estimated HRs (active : placebo) for EDSS score progression, with 95% CIs and p-values from Cox regression models, sequentially removing study sites with high loss to follow-up for each of the two data sets and each of the two ways of (more...)

Pre-specified subgroup analyses of time to first confirmed Expanded Disability Status Scale score progression

Pre-specified subgroup analyses of time to first EDSS score progression suggested a differential effect of treatment between participants with lower (4.0–5.5) and higher (6.0–6.5) baseline EDSS scores (Figure 7). There was little evidence of differential effects of treatment among subgroups defined in terms of sex, disease type, or age or weight at registration.

FIGURE 7

Estimated HRs (active : placebo) and 95% CIs for EDSS score progression by subgroup.

Primary analysis of change in Multiple Sclerosis Impact Scale-29 20-point physical subscale

A multilevel model fitted to repeated measures of MSIS-29phys score showed no evidence of an effect of treatment [estimated between-group difference (active–placebo) −0.91 points, 95% CI −2.01 to 0.19 points; p = 0.11; see Table 8], or of disease type, sex, weight or study site (data not shown; p > 0.05 for all).

It was estimated that MSIS-29phys score reduced by a mean of 1.4 points (95% CI 0.3 to 2.5 points; p = 0.02) for every 10-year increase in age. In both treatment groups, mean MSIS-29phys score decreased from baseline to month 3, after which it tended to increase (Figure 8).

FIGURE 8

Estimated mean MSIS-29phys score, with 95% CIs, at each visit, separated by treatment group. The numbers of patients with total scores calculated at each visit, in each treatment group, are given directly above the CI for placebo and directly below the (more...)

With the exception of a small reduction in MSIS-29phys score in patients with a baseline EDSS score of 5.0 compared with those with a score of 4.0, MSIS-29phys score tended to increase with increasing baseline EDSS score (data not shown).

Results from the primary analysis of repeated measures of MSIS-29phys remained unchanged after removal of non-significant terms from the fitted model and under an alternative analysis based on comparison of treatment groups in terms of change from baseline to last valid observation [estimated between-group difference (active–placebo) –1.4 points, 95% CI –3.3 to 0.4 points; p = 0.13].

Multiple Sclerosis Walking Scale-12

A multilevel model fitted to repeated measures of MSWS-12v2 score showed no evidence of an effect of treatment [estimated effect –0.19 (95% CI –0.97 to 0.60); p = 0.74; see Table 8], or of disease type, sex or weight (data not shown; p > 0.05 for all). There was some evidence of study site effects (data not shown) and of effects of baseline EDSS score. Compared with those with a baseline EDSS score of 4.0, MSWS-12v2 was estimated to be, on average, 5.7 (95% CI 2.3 to 9.0), 6.1 (95% CI 3.7 to 8.5) and 9.3 (95% CI 6.8 to 11.8) points higher in those with a baseline EDSS score 5.5, 6.0 and 6.5, respectively. In both treatment groups, mean MSWS-12v2 score decreased from baseline to month 3, after which it tended to increase (Figure 9).

FIGURE 9

Estimated mean MSWS-12v2, with 95% CIs, at each visit, separated by treatment group. The numbers of patients with total scores calculated at each visit, in each treatment group, are given directly above the CI for placebo and directly below the CI for (more...)

Multiple Sclerosis Functional Composite

A multilevel model fitted to repeated measures of MSFC composite z-score showed no evidence of an effect of treatment; estimated between-group difference (active–placebo) –0.03 (95% CI –0.19 to 0.09; p = 0.72; see Table 8). Multilevel models fitted to the MSFC component-wise z-scores each showed no evidence of an effect of treatment. Estimated between-group differences (active–placebo) were: T25-FW –0.08 (95% CI –0.25 to 0.09; p = 0.37); 9-HPT 0.05 (95% CI –0.04 to 0.13; p = 0.28); and PASAT –0.01 (95% CI –0.10 to 0.09; p = 0.92). Across both treatment groups, mean T25-FW, 9-HPT and composite z-scores increased from baseline to week 1, after which they tended to decrease. After an initial increase at week 1, PASAT z-scores remained relatively constant over the 3-year study period (Figure 10).

FIGURE 10

Estimated mean MSFC composite (a), T25-FW (b), 9-HPT (c) and PASAT (d) z-scores, with 95% CIs, at each visit, separated by treatment group. The numbers of patients with scores calculated at each visit, in each treatment group, are given directly above (more...)

Rivermead Mobility Index

A multilevel model fitted to repeated measures of RMI showed no evidence of an effect of treatment [the estimated between-group difference (active–placebo) was 0.04 (95% CI –0.24 to 0.32; p = 0.76; see Table 8)] or of disease type, sex or weight (data not shown; p > 0.05 for all). There was some evidence of study site effects (data not shown) and of effects of baseline EDSS score. Compared with those with a baseline EDSS score of 4.0, RMI was estimated to be, on average, 1.57 points (95% CI 0.38 to 2.76 points), 2.44 points (95% CI 1.59 to 3.30 points) and 4.99 points (95% CI 4.10 to 5.88 points) lower in those with a baseline EDSS score of 5.5, 6.0 and 6.5, respectively. In both treatment groups, RMI decreased from baseline to 30 months, after which it remained fairly constant (Figure 11).

FIGURE 11

Estimated mean RMI, with 95% CIs, at each visit, separated by treatment group. The numbers of patients with total scores calculated at each visit, in each treatment group, are given directly above the CI for placebo and directly below the CI for active. (more...)

Short Form questionnaire-36 items (physical health subscale)

A multilevel model fitted to repeated measures of SF-36(PH) showed no evidence of an effect of treatment [the estimated between-group difference (active–placebo) was –0.15 (95% CI –0.83 to 0.53; p = 0.67; see Table 8)], or of disease type, sex or weight (data not shown; p > 0.05 for all). In both treatment groups, mean SF-36(PH) score increased from baseline to month 3, after which it tended to decrease (Figure 12).

FIGURE 12

Estimated mean SF-36(PH), with 95% CIs, at each visit, separated by treatment group. The numbers of patients with total scores calculated at each visit, in each treatment group, are given directly above the CI for placebo and directly below the CI for (more...)

With the exception of a small increase in SF-36(PH) score in patients with a baseline EDSS score of 5.0 compared with those with a score of 4.0, SF-36(PH) score tended to decrease with increasing baseline EDSS score (data not shown).

Multiple Sclerosis Spasticity Scale-88

Figure 13 shows estimated mean MSSS-88 scores, with 95% CIs, by visit and treatment group, for each of the eight subscales.

FIGURE 13

Estimated mean MSSS-88 subscales 1–8 scores, with 95% CIs, at each visit, separated by treatment group. The numbers of patients with total scores calculated at each visit, in each treatment group, are given directly above the CI for placebo and (more...)

For each of the physical components of the MSSS-88, subscales 1 to 6 inclusive, after an initial decrease from baseline to month 3, mean scores tended to increase. Estimated means were consistent across treatment groups, as seen by the overlapping CIs. For the two psychological components, subscales 7 and 8, after an initial decrease from baseline to month 3, mean scores remained relatively constant over the study period. Estimated mean scores for these components tended to be higher in the active group than in placebo, but any differences failed to reach statistical significance.

Multilevel models fitted to three groups of the MSSS-88, where MSSS-88 (1) combines subscales 1–3; MSSS-88 (2) combines subscales 4–6 and MSSS-88 (3) combines subscales 7 and 8 (as described in Chapter 2), each showed no evidence of an effect of treatment. Estimated between-group difference (active–placebo) for MSSS-88 (1) was 0.26 (95% CI –1.99 to 2.52; p = 0.82; see Table 8); for MSSS-88 (2) was –0.02 (95% CI –2.35 to 2.32; p = 0.99; see Table 8); and for MSSS-88 (3) was 1.00 (95% CI –0.70 to 2.70; p = 0.25; see Table 8). In both treatment groups, mean MSSS-88 (1) and mean MSSS-88 (2) decreased from baseline to month 3, after which they tended to increase (Figure 14). After an initial decrease from baseline to month 3, mean MSSS-88 (3) remained relatively constant over the study period (see Figure 14).

FIGURE 14

Estimated mean MSSS-88 (1) total score from subscales 1–3, concerning muscle stiffness/spasms, pain and discomfort; MSSS-88 (2) total score from subscales 4–6, concerning activity, walking and body movements; and MSSS-88 (3) total score (more...)

Investigation of adverse events and serious adverse events

The number of participants experiencing at least one SAE was 114 (35%) in the Δ⁹-THC group and 46 (28%) in the placebo group, the most common SAE being admission to hospital for MS-related events and infections. The number and nature of SAEs experienced was similar across treatment groups (Table 9).

TABLE 9

Occurrences of SAEs and the most common AEs

There were numerous non-serious AEs in both groups, consistent with the effects of MS and the known safety profile of cannabinoids. The median number of events per participant in the active group was 11 (25th–75th percentiles 7–17) compared with 10 (25th–75th percentiles 6–14) in the placebo group. Of those events judged to be either moderate or severe, the most frequent are documented in Table 9. Among these AEs, there was some suggestion that those participants on active treatment were more likely to experience dizziness or light-headedness and dissociative and thinking or perception disorders. On the other hand, a higher proportion of patients in the placebo group experienced musculoskeletal pain and aches than in the active group.

Six SAEs were classified as potential SUSARs in accordance with European clinical trials legislation. Three events occurred in each of the active and placebo groups. Trial treatment was discontinued in three participants as a result of the SAE. Three SAEs were classified as nervous system disorders: two were psychiatric events and one related to the gastrointestinal tract.

Category rating scales

Responses to questions 9–16 of the category rating scales, relating to how the patient felt at the time of completing the questionnaire, compared with just before the start of the study, have been grouped (as described in Chapter 2) and summarised, in terms of frequencies and relative frequencies in the two treatment groups, at each follow-up (Tables 10–13). Unadjusted p-values from chi-squared tests for trend are given.

TABLE 10

Frequencies and relative frequencies (%) of responses to category rating scale questions 9–16, by treatment group, at 3 months after baseline. Percentages are taken with respect to total number of patients in the corresponding treatment group (more...)

TABLE 13

Frequencies and relative frequencies (%) of responses to category rating scale questions 9–16, by treatment group, at 3 years after baseline. Percentages are taken with respect to total number of patients in the corresponding treatment group

Generally, a higher proportion of patients on active treatment than on placebo reported being more forgetful at the time of follow-up compared with before the study. At 3 months from baseline, there was an approximately twofold increase in proportion of responses classified as ‘worse’ in the active group compared with placebo (32% active, 15% placebo; p = 0.0067; see Table 10). These proportions were similar across treatment groups at 1-year follow-up (49% active, 42% placebo; p = 0.25; see Table 11), at 2 years there was an approximately 30% increase in ‘worse’ responses in the active group compared with placebo (57% active, 44% placebo; p = 0.037; see Table 12) and similarly at 3 years, with a 33% increase (60% active, 45% placebo; p = 0.11; see Table 13). Responses to the remaining questions were similar across treatment groups.

TABLE 11

Frequencies and relative frequencies (%) of responses to category rating scale questions 9–16, by treatment group, at 1 year after baseline. Percentages are taken with respect to total number of patients in the corresponding treatment group

TABLE 12

Frequencies and relative frequencies (%) of responses to category rating scale questions 9–16, by treatment group, at 2 years after baseline. Percentages are taken with respect to total number of patients in the corresponding treatment group