Effectiveness of botulinum neurotoxin A, topiramate, and acupuncture in preventive treatment of chronic migraine: a network meta-analysis

Background Botulinum neurotoxin A (BoNT-A) was the primary choice for preventive treatment of chronic migraine. Topiramate and acupuncture showed promising effect for the condition, but their effectiveness relative to BoNT-A was rarely studied. We aimed to perform a network meta-analysis to compare the effectiveness and acceptability between BoNT-A, topiramate, and acupuncture. Methods We searched OVID Medline, Embase, the Cochrane register of controlled trials (CENTRAL), the Chinese Clinical Trial Register, and clinicaltrials.gov for randomized controlled trials (RCTs) that compared BoNT-A, topiramate, and acupuncture with any of them or placebo in the preventive treatment of chronic migraine. Two independent reviewers screened RCTs, extracted data, and assessed risk of bias. A network meta-analysis was performed by using a frequentist approach and a random-effects model. The primary outcomes were reduction in monthly headache days and monthly migraine days at week 12. Acceptability was assessed by adverse event rate. The effect size of the primary outcomes was measured by using standard mean difference (SMD). Results We included 15 RCTs (n = 2545). Eleven RCTs were at low risk of bias. The network meta-analyses (n = 2061) showed that acupuncture (2061 participants; SMD − 1.61, 95%CI -2.35 to -0.87) and topiramate(582 participants; SMD − 0.4, 95%CI -0.75 to -0.04) ranked the most effective in the reduction of monthly headache days and migraine days, respectively; but they were not signicantly superior over BoNT-A. Topiramate caused the most treatment-related adverse events and the highest rate of dropouts owing to adverse events. Conclusions Topiramate and acupuncture were not superior over BoNT-A. In terms of acceptability and accessibility of treatments, BoNT-A was still the primary choice for preventive treatment of chronic migraine.


Introduction
Migraine is a recurrent headache usually accompanied with nausea, photophobia, and phonophobia. Migraine affects 12% of the general population and has substantial impact on patient's quality of life and work productivity. Migraine is classi ed as episodic migraine and chronic migraine. Episodic migraine causes less frequent headache attacks (2-8 attacks per month). Chronic migraine affects 1-3% of the general population [1] and causes more severe socioeconomic impact than episodic migraine. In addition, patients with chronic migraine was usually accompanied with medication overuse headaches, which causes treatment failure in migraine prophylaxis. Few treatment options are available for the preventive treatment of chronic migraine. Botulinum neurotoxin A (BoNT-A) was the rst drug that was speci cally approved for the prophylactic treatment of chronic migraine by the US Food and Drug Administration [1,2]. As indicated by several systematic reviews and randomized controlled trials (RCTs), BoNT-A reduced the number of headache days and global pain intensity in patients with chronic migraine [3,4]. One systematic review indicated that BoNT-A had a relatively higher acceptance rate than other prophylactic agents [5], but the hypothesis was not statistically tested because of the lack of head-to-head comparisons. The e cacy of topiramate in the preventive treatment of chronic migraine was tested in several RCTs and meta-analyses [6][7][8][9]. Two multicenter RCTs showed superiority of topiramate over placebo in the reduction of headache days [8,9], and the ndings were further con rmed in a metaanalysis [10]. Acupuncture, a treatment that was e cacious for episodic migraine [11], was also tested for its effectiveness for chronic migraine in two recent RCTs [12,13]. Owing to the cumulating evidence for the e cacy of topiramate and acupuncture in the treatment of chronic migraine, clinical practitioners might wonder: (1) which of the 3 treatments-BoNT-A, topiramate, and acupuncture-was relatively more effective; (2) which treatment was relatively safer and more tolerable? The comparative effectiveness between BoNT-A, topiramate, and acupuncture in the treatment of chronic migraine were examined in three RCTs [7,13,14], but the three treatments were not compared simultaneously in one RCT.
Network meta-analysis, also known as mixed treatment comparison meta-analysis, combined both direct and indirect evidence to provide a more accurate estimate of a treatment effect [15]. It solves the problem of convention meta-analysis that pairwise comparison between treatments without head-to-head comparison in RCTs, and it therefore saves additional research expenditure and helps the clinical practitioners and participants to select treatment options that are suitable for their condition with data support. Our study primarily aimed to investigate the comparative effectiveness acceptability of BoNT-A, topiramate, and acupuncture in patients with chronic migraine through a network meta-analysis.

Methods
The systematic review was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement and PRISMA for network meta-analysis [16]. Additional details of this review were provided in Appendix A-G.

Study source
We searched OVID Medline, Embase, the Cochrane register of controlled trials (CENTRAL), Chinese Clinical Trial Register (ChiCTR), and clinicaltrial.gov for randomized controlled trials (RCTs) that compared BoNT-A, topiramate, and acupuncture with each other or placebo in the preventive treatment of chronic migraine. The literature search was performed from inception to march 1, 2020 without language restriction. Search strategies were provided in Appendix A. We searched google scholar (https://scholar.google.com) and Bing scholar (https://cn.bing.com/academic) for any missing RCTs.

Study selection
We selected only studies with randomized design to ensure the credibility of the ndings. We included: (1) RCTs with parallel design or crossover design with rst-phase data, since acupuncture and BoNT-A were reported with effect persisting for even 12 weeks after completion of treatment [17,18]; (2) RCTs that recruited participants with chronic migraine and with ≥15 monthly headache days and ≥8 monthly migraine days; (3) RCTs that compared the BoNT-A, topiramate, and acupuncture with each other or placebo interventions; and (4) RCTs that assessed at least one of the following outcomes: monthly headache days, monthly migraine days, headache intensity (assessed with visual analog scale or other scales), responder rate, six-items headache impact test (HIT-6), migraine disability assessment (MIDAS), quality of life scales, or adverse event. We excluded RCTs that examined acupuncture techniques not using liform needles and RCTs that examined the e cacy of BoNT-A or topiramate at doses outside the licensed range.
We screened the retrieved articles according to PICOS principle-Participants, Interventions, Controls, Outcomes, and Study design. Two reviewers screened titles and abstracts of the retrieved articles in duplicate, and further screened the full-text copies of the potentially eligible articles. Disagree in study selection was solved by group discussion and arbitrated by a third reviewer.

Data abstraction and risk of bias assessment
Two reviewers independently abstracted trial characteristics, participant's characteristics, risk of bias, and outcome parameters from the included RCTs. The trial characteristics included name of rst author, publication year, diagnostic criteria, study settings, participating countries, sample size, interventions, comparisons, and main outcomes. The participant's characteristics included mean age, proportion of female, and disease duration.
Dichotomous data for meta-analysis were extracted for the number of participants in each arm, the number of events, the follow-up time points; continuous data were extracted for mean, standard deviation, and the number of participants in each arm. The number of participants was extracted according to the intention-to-treat principle.
Risk of bias was assessed in the following aspects: random sequence generation, allocation concealment, blinding participants, blinding care providers, blinding outcome assessors, unpublished data, selective outcome reporting, and attrition bias. RCTs would be classi ed as low risk of bias if none of these aspects was rated as high risk of bias and three or fewer were rated as unclear risk of bias; moderate if one was rated as high risk of bias, or none was rated as high risk but four or more were rated as unclear risk; high risk of bias if all other cases were met [19].

Outcome assessments
The primary outcomes were changes in monthly headache days and monthly migraine days measured at week 12. Secondary outcomes included monthly headache days and monthly migraine days measured at other timepoints (week 4, week 8, and week 24), moderate and severe headache or migraine days, responder rate, headache intensity, headache impact test (HIT-6), migraine disability scale (MIDAS), quality of life scales, adverse event rate, and tolerability. A responder was de ned a >50% reduction in monthly headache days or monthly migraine days. HIT-6 and MIDAS are scales for testing the impact of migraine on social life, and higher scores indicate larger impact of migraine [20,21]. Tolerability was de ned as the number of dropouts owing to adverse events.

Data synthesis
We performed a network meta-analysis with a frequentist approach through R project (www.r-project.org, version 3.6.0). We calculated the standardized mean difference (SMD) for continuous data and the relative ratio (RR) for dichotomous outcome data, and we estimated their relative 95% con dence intervals (95%CIs). The data were primarily analyzed at week 12. We calculated the pooled estimates by combining both direct and indirect evidence with reference to placebo. We used P-score to rank the treatment by estimating the mean probability of a treatment to be the best treatment for a speci c outcome [22]. Higher P-scores in effectiveness outcomes correlate to greater probability in achieving the best treatment effect; higher P-scores in acceptability outcomes (i.e., adverse event rate and tolerability) indicate greater probability in causing more harm. We estimated the global heterogeneity in each network meta-analysis by using I 2 statistics; an I 2 from 0% to 49% indicates small heterogeneity, an I 2 from 50% to 74% indicates moderate heterogeneity, and an I 2 >=75% indicates large heterogeneity. We checked the consistency of the network by a comparison of the direct estimates and the indirect estimates [23]. We also used a design-by-treatment approach to detect the source of inconsistency in a network meta-analysis by using a method of decomposing Cochran's Q [24]. We performed sensitivity analyses by excluding RCTs with moderate to high risk of bias.

Trial characteristics
We identi ed 1133 articles, and we nally included 15 RCTs (n=2545) described by 17 articles [14,[25][26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41]. The process of study screening and selection and the lists of the excluded studies were showed in Appendix B and C. The mean age of the included study ranged from 38.8 to 48.8 years. Female participants constituted 44-91.5% of the total population, and the median proportion was 76.3%. The duration of migraine ranged from 3 to 28.1 years, and the median duration was 16 years. Table 1 shows the details of the included treatments and comparisons. Risk of bias assessment showed that 11 RCTs were at low risk of bias, 3 RCTs at moderate risk of bias, and 1 RCT at high risk of bias. Appendix D shows risk of bias in individual RCTs.

Migraine days
Topiramate was the most effective at week 12 (4 RCTs and 582 participants; SMD -0.4, 95%CI -0.75 to -0.04; P-score=0.96; global I 2 =39.3%; Figure 2B), but no signi cant difference between treatments was found. Network meta-analysis was not performed at the other time points because of few available data.

Discussion
We performed a network meta-analysis to perform pairwise comparison between BoNT-A, topiramate, and acupuncture in the preventive treatment of chronic migraine. We found that: (1) BoNT-A, topiramate, and acupuncture were effective in reducing headache days and migraine days at week 12. Although acupuncture ranked the most effective, no signi cant difference was found between the 3 treatments; (2) Other e cacy outcomes showed similar results; (3) topiramate caused higher total adverse event rate, treatment-related adverse event rate, and number of dropouts due to adverse events than acupuncture and BoNT-A, which indicated a lower acceptability of topiramate than BoNT-A and acupuncture.
Several SRs examined the e cacy of acupuncture [11,42], BoNT-A [5,43], or topiramate [10,44,45] in migraine prophylaxis. Most of the SRs focused on the e cacy of the 3 treatments in episodic migraine; and none of them has compared the 3 treatments by combining both direct and indirect evidence. BoNT-A was the rst treatment that was approved by the Food and Drug Administration for the preventive treatment of chronic migraine, and the American neurology society guideline supported the use of topiramate [1]. These two treatments were rarely compared in head-to-head RCTs. One RCT [30] showed that topiramate was similar to BoNT-A in improving outcomes of patients with chronic migraine, which was further con rmed in our study. Our study added that BoNT-A had better effect in reducing headache days and migraine days at week 16. Regarding the similar treatment effect between BoNT-A and topiramate and the lower acceptability of topiramate, BoNT-A should be considered as the rst choice for chronic migraine prophylaxis.
Most of the RCTs testing the e cacy of acupuncture focused on its speci c effect by comparing it to sham acupuncture. As found in a recent SR[46], we found only 2 trials comparing acupuncture with standard pharmacological treatments in the treatment of chronic migraine [14,25]. One RCT found acupuncture was superior over BoNT-A and valproate in reducing headache days at week 12 [25]; and the other trial found that acupuncture caused more reduction in headache days than topiramate [14]. This explained acupuncture ranking top in causing fewer headache days at week 12. We should interpret this nding with caution, since the two trials were small-scale open-label trials that maybe under the risk of small-study effect, and the inclusion of these two trials brought in between-study heterogeneity. However, the nding might, on the other hand, re ect the true effect of acupuncture in practice. Several trials testing the effect of acupuncture on episodic migraine also showed superiority of acupuncture over standard pharmacological treatments[47-49], and one of our SRs comparing acupuncture with propranolol using indirect evidence also showed similar results [50]. Based on these grounds, a new trial on chronic migraine with larger sample size might also showed similar results. Acupuncture treatment should be given for at least 10 sessions and at least one session per week as suggested by guidelines [51]. Considering the accessibility and cost of acupuncture treatment, BoNT-A should also be considered as the rst choice.
Our study had several limitations. First, we did not perform a network meta-analysis of the 3 treatment in longterm assessment (> 24 weeks), because few studies assessed the long-term effect of acupuncture on chronic migraine. Second, we did not perform subgroup analysis for medication overuse migraine because of insu cient data. Third, with-design heterogeneity in BoNT-A versus placebo precludes a rm conclusion based on current evidence. Fourth, cost-effectiveness evaluation was not performed.

Conclusions
Our network meta-analysis showed that all 3 treatments were effective in prophylactic treatment of chronic migraine. Considering the acceptability and accessibility of the treatments, BoNT-A should still be the primary choice for preventive treatment of chronic migraine. Availability of data and materials All data generated or analyzed data in study are included in this article.

List Of Abbreviations
Competing interests The authors declare no competing interests.  Footnote: Tolerability was de ned as the rate of participants who dropped out because of adverse event. Figure 1 Primary e cacy outcomes Abbreviations: 95%CI, 95% con dence interval. SMD, standard mean difference.

Figures
Footnote: The network meta-analyses assessed the two primary outcomes at week 12: (A) monthly headache days, (B) monthly migraine days. The left of both (A) and (B) shows the geometry of the networks, and the right shows the forest plots using placebo as a reference comparator. The size of the red nodes corresponds to the number of participants who were allocated to the treatments. Direct comparison was linked by a line between two treatments; the thickness of the lines corresponds to the number of trials that studied the treatment. The blue triangle among treatments indicates a three-arm design of an RCT. The treatments were ranked by P-scores. A P-score is an estimation of the mean probability of a treatment to be the best treatment. A treatment with the highest P-score ranked the most effective. A SMD>0 indicates superiority of a treatment over placebo.
Footnote: The network meta-analyses assessed the two primary outcomes at week 12: (A) monthly headache days, (B) monthly migraine days. The left of both (A) and (B) shows the geometry of the networks, and the right shows the forest plots using placebo as a reference comparator. The size of the red nodes corresponds to the number of participants who were allocated to the treatments. Direct comparison was linked by a line between two treatments; the thickness of the lines corresponds to the number of trials that studied the treatment. The blue triangle among treatments indicates a three-arm design of an RCT. The treatments were ranked by P-scores. A P-score is an estimation of the mean probability of a treatment to be the best treatment. A treatment with the highest P-score ranked the most effective. A SMD>0 indicates superiority of a treatment over placebo. shows the geometry of the networks, and the right shows the forest plots using placebo as a reference comparator. The size of the red nodes corresponds to the number of participants who were allocated to the treatments. Direct comparison was linked by a line between two treatments; the thickness of the lines corresponds to the number of trials that studied the treatment. The blue triangle among treatments indicates a three-arm design of an RCT. The treatments were ranked by P-scores. A P-score is an estimation of the mean probability of a treatment to be the best treatment. A treatment with the highest P-score ranked the most effective. A SMD>0 or an RR>1 indicates superiority of a treatment over placebo.

Figure 3
Safety outcomes Abbreviations: 95%CI, 95% con dence interval. RR, relative ratio. Footnotes: The network metaanalyses assessed adverse events at week 12 and 24. Tolerability was de ned as the number of dropouts owing to adverse events. If a treatment had higher RR of dropouts owing to adverse event, it indicates lower tolerability of treatment. The treatments were ranked by P-scores. A P-score is an estimation of the mean probability of a treatment to be the most harmful treatment. An RR>1 indicates superiority of a treatment over placebo.