Efficacy, Safety, and Acceptability of Pharmacologic Treatments for Pediatric Migraine Prophylaxis: A Systematic Review and Network Meta-analysis

Key Points

Question  What are the most effective, safe, and accepted pharmacologic treatments for migraine prophylaxis in children and adolescents?

Findings  In this network meta-analysis, comparing head-to-head and placebo-controlled trials found no significant long-term effects for migraine prophylaxis relative to placebo. Medium-sized short-term effects were found for propranolol and topiramate, but the prediction interval indicates that significant beneficial effects are to be expected in only 70% of similar studies conducted in the future.

Meaning  Considering the limited effect size, a cautious, individual, and tailored treatment approach to migraine prophylaxis is of great importance.

Importance  Migraine is one of the most common neurologic disorders in children and adolescents. However, a quantitative comparison of multiple preventive pharmacologic treatments in the pediatric population is lacking.

Objective  To examine whether prophylactic pharmacologic treatments are more effective than placebo and whether there are differences between drugs regarding efficacy, safety, and acceptability.

Data Sources  Systematic review and network meta-analysis of studies in MEDLINE, Cochrane, Embase, and PsycINFO published through July 2, 2018.

Study Selection  Randomized clinical trials of prophylactic pharmacologic treatments in children and adolescents diagnosed as having episodic migraine were included. Abstract, title, and full-text screening were conducted independently by 4 reviewers.

Data Extraction and Synthesis  Data extraction was conducted according to Preferred Reporting Items for Systematic Reviews and Meta-Analysis network meta-analysis guidelines. Quality was assessed with the Cochrane Risk of Bias tool. Effect sizes, calculated as standardized mean differences for primary outcomes and risk ratios for discontinuation rates, were assessed in a random-effects model.

Main Outcomes and Measures  Primary outcomes were efficacy (ie, migraine frequency, number of migraine days, number of headache days, headache frequency, or headache index), safety (ie, treatment discontinuation owing to adverse events), and acceptability (ie, treatment discontinuation for any reason).

Results  Twenty-three studies (2217 patients) were eligible for inclusion. Prophylactic pharmacologic treatments included antiepileptics, antidepressants, calcium channel blockers, antihypertensive agents, and food supplements. In the short term (<5 months), propranolol (standard mean difference, 0.60; 95% CI, 0.03-1.17) and topiramate (standard mean difference, 0.59; 95% CI, 0.03-1.15) were significantly more effective than placebo. However, the 95% prediction intervals for these medications contained the null effect. No significant long-term effects for migraine prophylaxis relative to placebo were found for any intervention.

Conclusions and Relevance  Prophylactic pharmacologic treatments have little evidence supporting efficacy in pediatric migraine. Future research could (1) identify factors associated with individual responses to pharmacologic prophylaxis, (2) analyze fluctuations of migraine attack frequency over time and determine the most clinically relevant length of probable prophylactic treatment, and (3) identify nonpharmacologic targets for migraine prophylaxis.

Migraines are common in children and adolescents, with a prevalence of 7.7%¹ that increases with age, from 3% in young children (age 3 to 7 years) to 8% to 23% in adolescents (age 11 to ≥15 years).² The prevalence is slightly higher in boys compared with girls prior to puberty, but incidence and prevalence of migraine increase more rapidly in girls than boys, and migraines are more prevalent in girls after age 11 years.¹ Migraine is a leading cause of disability across all age groups and a leading cause of outpatient and emergency department visits.³ Children and adolescents with migraine report impaired functioning in school, family, and extracurricular and social activities, on par with other pediatric chronic diseases including cancer and juvenile rheumatoid arthritis.⁴ With regard to pharmacotherapy, treatment options for children and adolescents have been largely based on adult studies. While the US Food and Drug Administration has approved individual triptan and combination products for adolescent migraine,⁵ pharmacologic interventions are often used off label in clinical practice.⁶ First-line intervention for acute pediatric migraine involves early treatment with a nonsteroidal anti-inflammatory drug,^7,8 and the use or addition of a triptan is often considered⁷ despite low efficacy vs placebo in pediatric trials.⁹ Furthermore, treatments deemed to be effective in adults and adapted for use in children might not be effective,^10-12 and a trial¹³ comparing topiramate, amitriptyline, and placebo showed no specific effect on migraine frequency. Also, analysis of adverse events indicated that the risk-benefit profile of amitriptyline and topiramate in preventing pediatric migraine is unfavorable.¹³

Most previous pairwise meta-analyses^9,10 were not able to generate hierarchies among available treatments because, to our knowledge, medications have not been directly compared. Therefore, we conducted a network meta-analysis (NMA) to systematically compare and rank medications regarding acceptability, safety, and efficacy in the prophylactic treatment of migraine in children and adolescents. We hypothesized that, compared with previous methodologies, use of network approaches could uniquely allow the integration of multiple direct and indirect treatment comparisons of medications and generate stronger conclusions regarding rankings among medications for safety, acceptability, and efficacy.

This study was a systematic review and NMA and is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement.^14,15 We searched MEDLINE, Cochrane, Embase, and PsycINFO from inception until July 2, 2018. Further trials were identified from an existing systematic review of prophylactic treatments for migraine.¹⁶ We applied a combination of keywords and text words related to migraine and prophylactic treatments, combined with validated filters for controlled clinical trials (eAppendix 1 in the Supplement). In total, the search returned 8799 articles (eFigure 1 in the Supplement). The search included nonpharmacologic studies as well as follow-up projects. The screening and selection processes were conducted independently by 4 coauthors (C.L., J.K., J.B., T.L.L.).

We included randomized clinical trials (RCTs) of prophylactic pharmacologic treatments for children and adolescents younger than 18 years. Participants were required to have a diagnosis of episodic migraine (with or without aura) according to the International Headache Society criteria, or criteria for migraine diagnosis had to be in close agreement with the International Headache Society classification. Quiz Ref IDEligible trial designs included RCTs that make head-to-head comparisons of at least 2 pharmacologic agents (ie, comparator trials) as well as RCTs that compare at least 1 pharmacologic agent with a placebo (ie, placebo-controlled trials).Quiz Ref ID Studies had to report at least 1 clinical outcome related to migraine (eg, migraine frequency or number of migraine days). We excluded crossover studies except when the results of the first period were given separately. We also excluded studies in which migraine was associated with other neurologic disorders as well as studies on menstrual migraine. We only considered studies including patients who experienced other headaches (eg, tension-type headache) if separate results for migraine patients were presented.

All study data were extracted in duplicate (C.L., J.B., T.L.L., and H.K., in pairs of 2) on a standardized form. Disagreements were resolved through consensus and, if necessary, consultation with a third reviewer. We extracted means and standard deviations (SDs) for continuous outcomes. If SDs were not provided, we calculated them from standard errors, confidence intervals, or other measures.^17,18 If we were unable to calculate SDs, we imputed them by the mean of SDs reported for the same outcome measure.¹⁹ If the sample size was missing in the table of analysis, we used the sample size of the descriptive statistics. In the absence of continuous outcome data, we extracted the number of persons who fulfilled the response criterion as defined by the study authors.

Our choice of primary outcomes was a pragmatic one, reflecting the wide variation in choice of primary measure in the studies, and outcomes were chosen to comply with a previous meta-analysis on migraine in adults conducted by 2 authors (K.M. and K.L.).¹⁶ To accommodate different primary outcomes in the included studies, we assembled a hierarchy of measures and chose the highest measure available in an individual study. Quiz Ref IDIn descending order, the primary efficacy outcome measures were (1) frequency of migraine attacks (means and SDs) per month at baseline and follow-up, (2) the number of migraine days per month, (3) number of headache days per month, (4) headache frequency, or (5) headache index/activity (composite score of intensity and duration/frequency of headache). As a secondary efficacy outcome measure, we extracted the proportion of responders. We preferably defined responders as patients with a reduction in attack frequency per month of at least 50%. If these data were not available, we used (in descending order of preference) patients (1) with at least a 50% reduction in the number of migraine days, (2) with at least a 50% reduction in the number of headache days, (3) with at least a 50% reduction in headache frequency, (4) at least 50% reduction in headache index, or (5) global assessment of improvement by patients and physicians. Regarding safety and acceptability outcomes, we defined primary outcomes as study discontinuation owing to adverse effects and for any reason, respectively.

The following time windows were applied: 8 weeks or 2 months after randomization, 3 to 4 months after randomization, 5 to 6 months after randomization, and more than 6 months after randomization. To increase the comparability between studies, our main analysis focused on outcomes reported at 3 to 4 months after randomization. If no data were reported for that time window, outcomes at 8 weeks or 2 months after randomization were used. Two studies did not report results for these time windows and were only included in the long-term analyses.^13,20 For the analysis of long-term effects, we considered all reported outcomes and included the last reported measurements in the analysis. All outcomes relied on patient reports, mainly collected in headache diaries. If a study contained multiple treatment groups that differed only in the dosage, we pooled the values.²¹ This was the case for 3 of the included studies.^22-24

Critical appraisals of included studies were conducted using the Cochrane risk of bias tool for RCTs.²⁵ Two reviewers assessed each study, with conflicts resolved through consensus or, if required, consultation with a third reviewer.

Our primary efficacy outcomes are continuous data, and we calculated the effect size (ES) of the interventions using the standardized mean difference (SMD). The magnitude of ESs was interpreted as small, moderate, or large, with 0.20, 0.50, and 0.80 SD units, respectively.²⁶ If no continuous data were available, we calculated odds ratios (ORs) as ES between groups¹⁷ and transformed them into SMDs according to the recommendations in the Cochrane Handbook of Systematic Reviews.²⁷ We decided to rely on continuous outcome data because dichotomizing continuous scores into categorical outcome data leads to a loss of information, reduces power, and creates an artificial boundary.^28,29 Regarding safety outcomes, we calculated the risk ratio (RR), ie, the probability of study discontinuation owing to adverse effects. Additionally, we calculated RRs for overall acceptability, ie, study discontinuation measured by the proportion of patients who withdrew for any reason.³⁰ For efficacy, safety, and acceptability outcomes, we chose to apply random-effects models rather than fixed-effects models because the studies we included were heterogeneous, and the number of studies was relatively small.³¹

To combine direct and indirect evidence for migraine prophylaxis, an NMA was conducted using the R package “netmeta”³² (the R Foundation), which implements a frequentist method based on a graph-theoretical approach according to the electrical network theory.³³ We ranked the various treatments for the efficacy outcomes using P scores, the most frequent analogue of the surface under the cumulative ranking curve.^34,35 The P scores are values between 0 and 1 and have an interpretation analogous to the surface under the cumulative ranking curve values and measure the extent of certainty that a treatment is better than another treatment, averaged over all competing treatments. P scores induce a ranking of all treatments that mostly follows that of the point estimates but takes precision into account.³⁵ Statistical significance was defined at a 2-sided α level of less than .05. We assumed that the between-study heterogeneity was the same for all treatment comparisons in the NMAs. Heterogeneity was quantified using the (within-design) Q statistic,³⁶ the between-study variance τ², and the heterogeneity statistic I².³⁷ An I² value of 0% to 40% might not be important; a value of 30% to 60% may represent moderate heterogeneity; a value of 50% to 90% may represent substantial heterogeneity; and a value of 75% to 100% may represent considerable heterogeneity.²⁷ Network meta-analyses rely on the assumption of transitivity to estimate indirect treatment effects.³⁸ We discussed the transitivity assumption by epidemiologic judgment considering 3 of its equivalent expressions by Salanti^39,40: (1) whether studies are comparable in terms of the distribution of effect modifiers; (2) whether the direct and indirect treatment effects are in statistical agreement (via an assessment for inconsistency); and (3) whether participants included in the network could in principle be randomized to any of the treatments. We conducted a statistical evaluation of consistency, ie, the agreement between direct and indirect evidence, using local (ie, separating direct from indirect evidence)⁴¹ as well as global (ie, design-by-treatment interaction test)⁴² approaches. The certainty of evidence in network estimates of the main outcomes (ie, efficacy, acceptability, and safety) was assessed using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) ratings.⁴³ The GRADE ratings were conducted in Confidence in Network Meta-analysis (CINeMA; Institute of Social and Preventive Medicine).⁴⁴ In GRADE, the quality of a body of evidence is characterized on the basis of the study limitations, imprecision, inconsistency, indirectness, and publication bias (eAppendix 2 in the Supplement).⁴³ To understand the clinical interpretation of the heterogeneity, prediction intervals were calculated to estimating what true treatment effects can be expected in future settings.⁴⁵ So far, there is a lack of a concrete methodology of assessing across-studies bias (publication bias) in NMA. Therefore, a comparison-adjusted funnel plot with accompanying Egger test for asymmetry was conducted.⁴⁶

In total, 26 double-blind RCTs met our inclusion criteria, but 3 studies had to be excluded because they did not connect with the other studies in the networks^47-49: 2 studies compared a pharmacologic treatment with a psychological treatment (ie, metoprolol vs relaxation vs biofeedback; triptan vs Behavioral Migraine Management Program), and 1 study examined an unconnected pharmacologic combination (ie, sodium valproate and ω-3 VS sodium valproate and placebo). Overall, 23 double-blind, parallel RCTs (comprising 2217 patients) conducted between 1974 and 2018 and comparing 13 pharmacologic treatments with each other or with placebo were included in our analyses^{13,20,22-24,50-67} (eFigure 1 in the Supplement). The individual characteristics of the 23 studies included in the NMAs are given in eTable 1A and B in the Supplement.

In total, 1698 participants were randomly assigned to active drugs and 519 were randomly assigned to placebo. The mean (SD) age was 10.9 years (2.41), and 1042 of the sample population were girls (46.8%). Most reported trials (14 trials; 61%) included patients who had either migraine with or without aura; 2 studies focused on migraine without aura (9%); and the remaining 7 trials (30%) did not specify. The median duration of the short-term treatment was 12 weeks (range, 4-24 weeks). Furthermore, 6 studies (26%) were multicenter studies. Nine of 23 trials (39%) recruited patients from Asia, 6 from Europe (26%), 5 from North America (22%), and 1 from Australia (4%) (1 trial [4%] was cross-continental and the remaining 1 did not specify [4%]).

Nineteen studies provided sufficient data in the prespecified timeframe to be included in the main efficacy analysis; 2 additional studies could only be included in the efficacy long-term analysis^13,20 (eTable 1A and B in the Supplement). Three studies^50,61,62 reported only categorical data, and the ORs were transformed into SMDs. Figure 1A shows the network of eligible comparisons for efficacy. All prophylactic pharmacologic treatments, except pregabalin, had at least 1 placebo-controlled trial. Quiz Ref IDOnly 2 treatments were significantly more effective than placebo when data were combined in the NMA: propranolol, with an SMD of 0.60 (95% CI, 0.03-1.17), and topiramate, with an SMD of 0.59 (95% CI, 0.03-1.15). All other pharmacologic interventions revealed nonsignificant SMDs, ranging from −0.21 (95% CI, −1.40 to 0.99 for l-5-hydroxytrypto) to 0.93 (95% CI, −0.12 to 1.98 for flunarizine) (Figure 2A). There were no significant differences between the different prophylactic treatments (eTable 2A in the Supplement). While flunarizine appeared to have the largest effect size (SMD = 0.93, 95% CI = −0.12 to 1.98) and P score (P = .81) with respect to efficacy outcomes (Figure 2A), it was only evaluated by 1 trial, with a sample size of 21 patients per arm.⁶⁵

Certainty of evidence was only low to moderate (see eAppendices 2-4 and eFigures 2-4 in the Supplement for details). Evaluating imprecision, we found that the 2 statistically significant comparisons (ie, propranolol vs placebo and topiramate vs placebo) revealed a clinically significant effect size (ie, SMDs >0.20). However, in the comparison between propranolol and placebo, the prediction interval (95% prediction interval, −0.62 to 1.82) was considerably wider than the significant confidence interval (95% CI, 0.03-1.17; Figure 2A) and included both clinically beneficial and detrimental effects. Similar results were observed in the comparison between topiramate and placebo (95% CI, 0.03-1.15; Figure 2A); the prediction interval (95% prediction interval, −0.62 to 1.80) extended into nonsignificant effects. Quiz Ref IDFor both propranolol vs placebo and topiramate vs placebo, only a prediction interval of 70% revealed significant effects (eAppendix 5 in the Supplement). In the long-term analysis, no treatment was significantly more effective than placebo (eAppendix 6, eFigures 5-7, and eTable 3A-C in the Supplement).

Nineteen studies assessing acceptability (ie, treatment discontinuation for any reason), were included in the main analysis; 2 additional studies^13,20 could be included in the long-term analysis (eTable 1A and B in the Supplement). We found no significant differences between prophylactic pharmacologic treatments and placebo in terms of the RRs, ranging from 0.49 (95% CI, 0.12-1.97 for riboflavin) to 1.50 (95% CI, 0.70-3.21 for sodium valproate) (Figure 2B). There were no significant differences between the various prophylactic pharmacologic treatments (eTable 2B in the Supplement for details). In the long-term analysis, none of the treatments was significantly more acceptable than placebo (eAppendix 6 in the Supplement).

With respect to safety (ie, treatment discontinuation owing to adverse effects), only 11 trials reported sufficient data; 2 additional studies could be included in the long-term analysis^13,20 (eTable 1A and B in the Supplement). There were no significant differences between prophylactic pharmacologic treatments and placebo in terms of RRs, ranging from 0.78 (95% CI, 0.02-37.83 for riboflavin) to 7.00 (95% CI, 0.38-128.47 for flunarizine) (eTable 2C in the Supplement). Again, the various pharmacologic treatments were ranked using P scores (Figure 2C). In the long-term analysis, no treatment was significantly safer than placebo (eAppendix 6 in the Supplement). The certainty of evidence for the acceptability and safety network estimates (ie, in line with GRADE) is reported in eAppendixes 2-4 and eFigures 2-4 in the Supplement.

This systematic review and NMA of double-blind RCTs with 2217 patients with pediatric migraine assessed the efficacy, safety, and acceptability of β-blockers, anticonvulsants, antidepressants, and antihistaminic and calcium channel blockers to natural supplements and placebo. The NMA revealed a significant effect of propranolol and topiramate compared with placebo. However, the 95% prediction interval for both these studies, which reflects the variation in true treatment effects over different settings, including what effect is to be expected in future patients,⁴⁵ was nonsignificant. Pregabalin and flunarizine both showed high mean SMDs compared with placebo, yet were each based on 1 study and nonsignificant given the variance within the studies. Further, none of the investigated drugs demonstrated convincing evidence that it reduces the migraine frequency in the long run more than a placebo. While our results indicate the potential specifically for these 4 medications, they also emphasize the need for studies aimed at identifying children who are likely to benefit from pharmacologic prophylaxis and determining the most clinically relevant length of probable prophylactic treatment. With regard to safety, the lower risk ratio and higher P score indicate a more favorable benefit-risk profile of propranolol compared with topiramate. However, the difference between the 2 drugs in terms of safety was nonsignificant (RR, 0.56; 95% CI, 0.10-3.08).

Our results confirm the results of the Childhood and Adolescent Migraine Prevention (CHAMP) study,¹³ which found no significant difference between topiramate, amitriptyline, and placebo in reducing migraine headaches in children. Because specific effects of drugs are associated with the size of the placebo effect,¹⁶ the lack of drug efficacy in our NMA could be owing to a comparatively high placebo effect in children. In fact, there is indirect evidence that the placebo effect is more pronounced in children and adolescents than in adults.^68-70 Noteworthy, in the CHAMP study, 61% of the children in the placebo group met the criterion of response. (ie, ≥50% reduction in headache frequency),¹³ whereas the mean placebo response rate in pharmacologic studies of migraine prophylaxis in adults was only 22%.¹⁶ However, placebo response rates can include improvements owing to regression to mean, unidentified cointerventions, and spontaneous improvement.⁷¹ The quantification of the placebo effect would therefore require comparison with a nontreated group, which is rarely included in clinical trials. Therefore, a large placebo effect size is a possible explanation for the lack of convincing efficacy in our NMA.

The direct comparison of different drugs is of great importance for clinicians and patients to choose the safest and most effective among existing treatments. A network meta-analytic approach uniquely allows us to combine direct and indirect evidence to get the most precise estimate of the treatment differences and associated standard errors.

This study has several limitations that should be taken into account when interpreting the results. First, a major potential limitation of our efficacy NMA is associated with the fact that most substances (7 of 12) have been tested in less than 100 patients (Figure 2A). It is therefore possible that the effect of some of these substances is owing to a so-called small-study effect: smaller trials show different, often larger, treatment effects than bigger ones.^72,73 Second, substantial heterogeneity was found in our efficacy NMA. The variety of the dosages, the format of the treatment (eg, duration), and the reporting methods differed widely, which may have contributed to the statistical heterogeneity and certainly to the clinical heterogeneity. However, we tried to reduce heterogeneity by defining an a priori list of preference in terms of the reporting methods of the primary outcomes that was in line with a meta-analysis of migraine prophylaxis for adults.¹⁶ Furthermore, we found no evidence of inconsistency. Third, a potential limitation of our safety NMA is that only 50% of the included trials reported dropouts owing to adverse effects, and the way of reporting these effects varied widely. Therefore, the ranking for safety and acceptability should be interpreted with caution. Fourth, although NMAs have the advantage of making use of all available data, the indirect evidence does not directly stem from randomized comparisons.⁷⁴ Further, according to the GRADE framework, the within-study bias of many comparisons was assessed as moderate. However, many trials, especially the older ones, did not report adequate information about allocation concealment and sequence generation, which limits the interpretation of these results. To increase the methodologic value of the contributing evidence, we only included double-blind trials, which are usually similar in terms of the study design. Finally, young children could have difficulties differentiating between a migraine attack and a period of tension type headache. This could severely influence the primary and secondary end points of various studies.

Further studies are needed to determine the best treatment for the prophylaxis of pediatric migraine, especially those that directly compare medicines and psychologic treatments. In addition, trials are needed that allow to quantify the placebo effect in pediatric migraine. If the placebo effect is confirmed to be large in children and adolescents, innovative treatment strategies should be considered that harness the placebo effect in the treatment of pediatric migraine.⁷⁵

According to our results, prophylactic pharmacologic treatments have little evidence supporting efficacy for pediatric migraine. We advise to carefully weigh the benefits of prophylactic medications against their potential harms. Future research could (1) identify factors associated with individual responses to pharmacological prophylaxis, (2) analyze fluctuations of migraine attack frequency over time and determine the most clinically relevant length of probable prophylactic treatment, and (3) identify nonpharmacologic targets for migraine prophylaxis.

Corresponding Author: Joe Kossowsky, PhD, MMSc, Department of Anesthesiology, Critical Care, and Pain Medicine, Boston Children’s Hospital, Harvard Medical School, 333 Longwood Ave, Boston, MA 02115 (joe.kossowsky@childrens.harvard.edu).

Accepted for Publication: September 25, 2019.

Published Online: February 10, 2020. doi:10.1001/jamapediatrics.2019.5856

Author Contributions: Drs Kossowsky and Meissner had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Locher and Kossowsky contributed equally to this study.

Concept and design: Locher, Kossowsky, Linde, Meissner.

Acquisition, analysis, or interpretation of data: Locher, Kossowsky, Koechlin, Lam, Barthel, Berde, Gaab, Schwarzer, Meissner.

Drafting of the manuscript: Locher, Kossowsky, Koechlin, Meissner.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Locher, Kossowsky, Koechlin, Berde, Schwarzer.

Obtained funding: Locher, Meissner.

Administrative, technical, or material support: Locher, Kossowsky, Barthel, Gaab.

Supervision: Locher, Kossowsky, Berde, Gaab, Linde, Meissner.

Conflict of Interest Disclosures: Dr Berde reports grants from Amgen and other support from Grunenthal and Akelos outside the submitted work. Dr Locher reported grants from Swiss National Science Foundation during the conduct of the study. Dr Meissner reported grants from Schweizer-Arau Foundation, Germany, during the conduct of the study. No other disclosures were reported.

Funding/Support: This work was supported in part by the Sara Page Mayo Endowment for Pediatric Pain Research, Education, and Treatment. Dr Locher received funding for this project from the Swiss National Science Foundation (P400PS_180730). Dr Meissner received funding from the Schweizer-Arau-Foundation and the Theophrastus Foundation, Germany.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.