Methods of the Indirect Comparison

Evidence Network

The relevant networks available for Ona A in patients with CM are shown in Figure 2, Figure 3, and Figure 4. These networks describe change from baseline in monthly migraine days, change from baseline in monthly headache days, and all-cause discontinuation, respectively. Limited data were available for ≥ 50% reduction in migraine days and quality of life, and networks were therefore not available for these outcomes. Networks for discontinuations due to AEs and serious adverse events (SAEs) were available for the chronic and episodic patient population combined and have not been presented in this appendix.

Figure 2Network of Studies – Monthly Migraine Days (Extracted from ICER [2018], p.197)

Source: Institute for Clinical and Economic Review.¹⁹

Figure 3Network of Studies – Monthly Headache Days (Extracted from ICER [2018], p.199)

Source: Institute for Clinical and Economic Review.¹⁹

Figure 4Network of Studies — All-Cause Discontinuation (Extracted from ICER [2018], p.207)

Source: Institute for Clinical and Economic Review.¹⁹

Indirect Treatment Comparison Methods

An NMA was conducted if data were available from at least three similar studies, with respect to characteristics such as population, intervention, outcome, and time point. Sufficient data were available for the following outcomes in the CM population: change from baseline in monthly migraine days, change from baseline in monthly headache days, change from baseline in days per month using acute medications, and all-cause discontinuations. Aside from monthly acute medication use, the networks for these outcomes (Figure 2, Figure 3, and Figure 4) included Ona A, with comparisons against placebo, topiramate, and CGRP inhibitors. There were insufficient data to conduct an NMA for ≥ 50% reduction in migraine days or quality of life (MIDAS, MSQ, or HIT-6). In addition, NMAs for discontinuations due to adverse events, AEs reported by at least 5% of patients in a trial arm, and SAEs were not available for patients with chronic migraine. A meta-regression with a covariate for time point was also conducted. A treatment was concluded to favour another if the credible interval (CrI) excluded the null.

The NMAs followed a Bayesian framework, with random effects on the treatment parameters and a between-study variance that was assumed to be constant across treatment comparisons. Continuous outcomes were analyzed with a normal likelihood and identity link and binary outcomes with a binomial likelihood and logit link. The treatment effects were presented as mean differences with 95% CrIs for continuous outcomes and odd ratios with 95% CrIs for binary outcomes. Non-informative prior distributions were used for all model parameters. The first 50,000 iterations were discarded as “burn-in” and base inferences were made on an additional 50,000 iterations using three chains, and chain convergence was assessed visually with trace plots. If studies reported multiple time points, the NMAs included the latest time point data. Separate NMAs were conducted at monthly time points (e.g., four, eight, 12, and 26 weeks) where data were available. A subgroup of patients who had failed at least one prior preventive treatment was also analyzed.

Results

Fourteen trials were available in patients with chronic migraine. Of these, four RCTs and one crossover trial compared Ona A with placebo (Table 50), two RCTs compared Ona A with topiramate (Table 50), four RCTs compared topiramate with placebo (Table 51), and three RCTs compared CGRP inhibitors (i.e., erenumab and fremanezumab) with placebo (Table 52). Sample sizes, baseline characteristics, and treatment doses in these trials are provided in Table 50, Table 51, and Table 52.

Six trials (Tepper,⁷¹ Bigal,⁷² Silberstein,⁷³ Aurora,⁵⁹ Diener,⁶⁰ and Silberstein⁶³) were included in the NMA for the mean change from baseline in monthly migraine days. The time point of analysis was the full 16-week period for the topiramate trial, the full 24-week period for the two Ona A trials, and the last four weeks of the randomization period for the three CGRP inhibitor trials, and is a potential source of heterogeneity. An average change from baseline of 3.8 to 6.3 fewer migraine days per month was reported in patients receiving placebo across the individual trials.

Eight trials (Bigal,⁷² Cohen,⁸¹ Aurora,⁵⁹ Diener,⁶⁰ Cady,³⁷ Freitag,⁷⁵ Silberstein,⁸² and Cady⁸⁰) were included in the NMA for the mean change in monthly headache days. The analysis time point was the last four weeks of the randomization period for two of the Ona A trials (Freitag⁷⁵ and Cady³⁷) and the two fremanezumab trials,^72,81 the full 12-week period for the head-to-head Ona A and topiramate trial,⁸⁰ and the full 24-week period for the two PREEMPT trials,^59,60 and is a potential source of heterogeneity. An average change from baseline of 3.3 to 8.0 fewer headache days per month was reported in patients receiving placebo across the individual trials.

In Table 54, Table 55, and Table 56, the results for change from baseline in monthly migraine days, change from baseline in monthly headache days, and all-cause discontinuation, respectively, for Ona A from NMAs are shown. No treatment was favoured for monthly migraine days or monthly headache days. For monthly migraine days, Ona A favoured placebo (mean difference = −1.95; 95% CrI, −3.62 to −0.28) and change from baseline in monthly headache days (mean difference = −2.06; 95% CrI, −3.48 to −0.63). No treatment was favoured for all-cause discontinuation compared with placebo, topiramate, or CGRP inhibitors.

An NMA was conducted at multiple time points (i.e., four weeks, eight weeks, and 12 weeks) and, additionally, a network meta-regression was performed with study duration as a covariate. The results for monthly migraine days and monthly headache days by time point were available for Ona A 155 U versus placebo and are provided in Table 57. For monthly migraine days, Ona A favoured placebo at week 4 and week 8, but at week 12 there was no difference. No treatment was favoured for monthly headache days at any time point.

Critical Appraisal

The NMAs were based on a systematic review of the literature to identify all relevant published trials from multiple databases, although the focus of the review was on CGRP inhibitors as the intervention, rather than Ona A. Of note, the FORWARD study, which was an open-label RCT that compared Ona A (155 U) with topiramate (up to 200 mg/d) in 282 patients with CM, was not included in the NMAs.²² The FORWARD study was available to CADTH as a Clinical Study Report and has not been published to date. While the patient population (i.e., adults with CM and eligible for preventive migraine therapy) was in alignment with the reimbursement request, there were limited data available for patients who failed previous therapies. The CGRP inhibitor trials excluded patients who experienced failures from two or three previous treatments and therefore the applicability of the evidence to the patient population of interest is limited. This is also a potential source of heterogeneity in the NMAs. The Health Canada–approved dosing for Ona A is 155 U up to 195 U. While the main trials in the NMA (i.e., PREEMPT-1 and PREEMPT-2) followed the Health Canada–approved dosing, several trials used either a smaller dose (i.e., 100 U) or a higher dose (200 U). This is another factor that limits the applicability of the NMA results to the patient population of interest and is a source of heterogeneity. A comprehensive set of safety and efficacy outcomes was evaluated, and included quality-of-life scales such as MIDAS, MSQ, and HIT-6. However, the data available for quality of life were insufficient for NMA and follow-up on all outcomes was limited from 12 to 26 weeks.

The evidence base for monthly migraine days, monthly headache days, and all-cause discontinuation formed connected networks of trials. Direct and indirect evidence was available only for Ona A versus topiramate 100 mg/d for all-cause discontinuation (Figure 4). The ICER report did not present the direct and indirect estimates separately for this treatment comparison, and the consistency of the direct and indirect estimates is therefore unclear. However, the report did indicate that for networks that were loops, the assumption of consistency among indirect and direct estimates was examined empirically using a node-splitting approach, and that no evidence of inconsistency was observed.

The report did not provide a discussion about whether the transitivity assumption was met in the networks of trials. Table 50 to Table 52 show that there were differences among the trials in the mean number of years since onset (i.e., shorter in the topiramate trials). There were also differences among the trials in the exclusion of previous treatment failures, whether ongoing preventive therapy was allowed, and the percentage of patients with medication overuse headache (trials either excluded these patients or prevalence ranged from 41% to 68%). These factors may be important treatment-effect modifiers, but they were not examined in analyses.

The NMA considered time points in meta-regression, and attempted a subgroup analysis for patients who had failed previous therapies. However, no other sources of potential heterogeneity, such as number of previous treatment failures, use of concomitant migraine preventive therapy, compliance with headache diary, Ona A dose, or study quality, were considered.

The clinical expert consulted on this review indicated that placebo response would vary based on the method placebo was received (i.e., injection versus oral tablets) and that placebo response is higher when it is received as an injection. Across the trials included in the NMA, the average change from baseline for placebo ranged from 3.8 to 6.3 fewer migraine days per month and 3.3 to 8.0 fewer headache days per month, which suggests that the placebo response was different between trials. The ICER report did not perform an NMA meta-regression on placebo response where meta-regression models impose a common interaction effect between baseline risk and relative effectiveness that account for variation in reference arm response across trials. While adjusting for placebo response might be the preferred approach, there are limitations to the approach, because there is an assumption that study and patient characteristics (the effect modifiers of the relative treatment effect) are also prognostic factors of the outcome with placebo.^83,84 And given that the extent to which placebo response is an adequate proxy for specific characteristics or effect modifiers is unclear, uncertainty remains in such analysis.

The strength of the network was low, with only six studies for seven treatment options (for change from baseline in monthly migraine days) and only eight studies for seven treatment options (change in monthly headache days). The networks were centered on placebo and most comparisons were indirect. All of the studies included in the analysis for change from baseline in monthly migraine days were of good quality; however, three of the eight studies included in the analysis for the mean change in monthly headache days were of poor quality. A sensitivity analysis based on study quality was not conducted.

The ITC did not include any health-related quality-of-life data, patient-reported symptoms, or key safety outcome SAEs, and withdrawals due to an adverse event.

As with all NMA, inclusion of the null value in the 95% CrIs of the difference between treatments does not necessarily imply that the treatments are equivalent or noninferior.