A Comparative Effectiveness Meta-Analysis of Drugs for the Prophylaxis of Migraine Headache

Objective To compare the effectiveness and side effects of migraine prophylactic medications. Design We performed a network meta-analysis. Data were extracted independently in duplicate and quality was assessed using both the JADAD and Cochrane Risk of Bias instruments. Data were pooled and network meta-analysis performed using random effects models. Data Sources PUBMED, EMBASE, Cochrane Trial Registry, bibliography of retrieved articles through 18 May 2014. Eligibility Criteria for Selecting Studies We included randomized controlled trials of adults with migraine headaches of at least 4 weeks in duration. Results Placebo controlled trials included alpha blockers (n = 9), angiotensin converting enzyme inhibitors (n = 3), angiotensin receptor blockers (n = 3), anticonvulsants (n = 32), beta-blockers (n = 39), calcium channel blockers (n = 12), flunarizine (n = 7), serotonin reuptake inhibitors (n = 6), serotonin norepinephrine reuptake inhibitors (n = 1) serotonin agonists (n = 9) and tricyclic antidepressants (n = 11). In addition there were 53 trials comparing different drugs. Drugs with at least 3 trials that were more effective than placebo for episodic migraines included amitriptyline (SMD: -1.2, 95% CI: -1.7 to -0.82), -flunarizine (-1.1 headaches/month (ha/month), 95% CI: -1.6 to -0.67), fluoxetine (SMD: -0.57, 95% CI: -0.97 to -0.17), metoprolol (-0.94 ha/month, 95% CI: -1.4 to -0.46), pizotifen (-0.43 ha/month, 95% CI: -0.6 to -0.21), propranolol (-1.3 ha/month, 95% CI: -2.0 to -0.62), topiramate (-1.1 ha/month, 95% CI: -1.9 to -0.73) and valproate (-1.5 ha/month, 95% CI: -2.1 to -0.8). Several effective drugs with less than 3 trials included: 3 ace inhibitors (enalapril, lisinopril, captopril), two angiotensin receptor blockers (candesartan, telmisartan), two anticonvulsants (lamotrigine, levetiracetam), and several beta-blockers (atenolol, bisoprolol, timolol). Network meta-analysis found amitriptyline to be better than several other medications including candesartan, fluoxetine, propranolol, topiramate and valproate and no different than atenolol, flunarizine, clomipramine or metoprolol. Conclusion Several drugs good evidence supporting efficacy. There is weak evidence supporting amitriptyline’s superiority over some drugs. Selection of prophylactic medication should be tailored according to patient preferences, characteristics and side effect profiles.


Introduction
Migraine headaches are common, with a worldwide prevalence ranging between 8 and 18% [1][2][3][4][5][6][7]. Migraines cause significant disability [8][9][10][11], even during periods between attacks [12], and are responsible for $1 billion in medical costs and $16 billion in lost productivity per year [13,14] in the US alone. The diagnostic criteria for migraine headaches have evolved over time. Currently, the International Headache Society (IHS) diagnostic criteria for migraine includes having at least 5 attacks that last 4-72 hours, that are unilateral, pulsating, moderate or severe in intensity and aggravated by or cause avoidance of routine physical activity and are also accompanied by nausea and/or vomiting, photophobia or phonophobia [15]. IHS further classifies migraine as with or without an aura and as episodic or chronic. Chronic migraine is defined as more than 15 migraine headaches per month for more than 3 months. Chronic migraines result in significantly greater disability than episodic migraines [16].
Treatment of headaches can be either abortive or prophylactic. Abortive treatment provides symptom relief for the acute headache [17,18], while prophylactic treatment aims to reduce the frequency or severity of headaches over time. We focus on prophylactic migraine headache treatment in this manuscript. There are a large number of prophylactic treatment options available; common ones include alpha antagonists, anti-convulsants [19], beta-blockers [20], botulinum-A [21], calcium channel blockers [22], serotonin agonists [23], serotonin reuptake inhibitors (SSRIs) [24] and tricyclic antidepressants (TCAs) [25]. Two emerging prophylactic candidates are angiotensin converting enzymes (ACE) and angiotensin receptor antagonists (ARB). Unfortunately nearly half of males and a third of females who are candidates for prophylactic therapy do not receive it [26]. Selection of prophylactic treatment is tailored on individual patient characteristics, costs and side effects of the available options. However, for patients and their providers, the decision about which prophylactic regimen to use is hampered by the lack of head to head trials comparing the different classes of medications. In addition, previous systematic reviews have focused on single classes of drugs. Two recent systematic reviews that looked more broadly at different drug options have been published. One only included studies since 1999 and did not pool any results, providing qualitative statements about relative treatment effectiveness [27]. Another review analyzed focused only on dichotomous outcomes among patients with episodic migraines and found no difference in likelihood of experiencing at least 50% improvement in headaches between different classes of oral medications [28]. Previous systematic reviews have also had methodological problems. Some combine outcomes from the end of the study, regardless of study duration. This inappropriately combines study results at markedly different time points. This also tends to overstate the strength of the evidence by making it appear that there are more studies contributing data to the results and produces inappropriately narrow confidence intervals. We conducted a metaanalysis asking what is the comparative effectiveness and side effects of the prophylactic treatment of migraine headaches in adults using oral pharmacological medications.

Materials and Methods
This report closely adheres to the PRISMA guidelines for conducting a systematic review [29]. We searched MEDLINE, EMBASE, the bibliographies of all retrieved articles, published systematic reviews and the Cochrane Database of Clinical Trials for each of the classes of medications (Table 1) through 7 November 2014. The search was conducted independently in duplicate. We included published, randomized clinical trials that evaluated efficacy in reducing the frequency or severity of migraine headaches that were at least 4 weeks in duration among adults. These comparisons could be between active treatment with placebo controls or comparative trials comparing two or more active treatments. We did not include unpublished data as there is no systematic means of searching for it. Because the classification of headache has changed over time [30,31], two authors independently reviewed each included article's headache definition and, where possible, classified it according to the 3rd edition of the International Headache Society (IHS) criteria (ICDH-III) and included only those that could reasonably be defined based on these diagnostic criteria [15]. For headache trials before 2004, we classified trials as focusing on episodic or chronic migraine based on the number of headaches experienced by participants at baseline.
Two authors independently abstracted data. Because measures of headache outcomes varied, a priori we followed International Headache Society outcome recommendations by prioritizing abstraction and analysis in this order: 1) headache frequency, 2) a headache index that included frequency, 3) severity or 4) duration [32]. Headache frequency was standardized to number of headaches per month. Whenever possible, we pooled frequency as the number of headaches/month. When not possible, we pooled standardized mean differences between studies, a measure also known as an effect size. By convention, effect sizes greater than 0.8 are considered to be large effect sizes, 0.5-0.8 moderate and 0.2-0.5 small [33]. When missing, variances were calculated from reported mean, sample size and p values [34]; for one non-placebo comparison trial [35] variance was imputed based on sample size and the reported effect size (r 2 = 0.76) When not explicitly reported, to verify we were using the proper variance, we tested the abstracted data for each article to ensure that the p value reported in the article matched our analysis. This helped insure that standard errors weren't abstracted as standard deviations, a common error in systematic reviews [36]. In addition, because of reports on the potential for misleading data [37,38], we only accepted data that was unadjusted and that was either based on a true intention to treat analysis or based on the subjects remaining in the trial. We rejected any "modified intention to treat" analyses or analyses subject to other adjustments. We assessed article quality independently and in duplicate, using both component and scales approaches using the Cochrane Risk of Bias Tool [39] and the Jadad scale [40] with good interrater agreement (Cochrane ICC: 0.83; Jadad kappa: 0.85). Disagreements were resolved by consensus.
For studies with more than one arm or using a cross-over design, we followed the recommendations of the Cochrane collaboration by pooling the arms into a single arm (if the study reported no differences between arms) or by reducing the sample sizes for cross-over trials by 50% [41]. We abstracted data from each trial at the following time points: baseline, 4, 8, 12, 24, 30 and 36 weeks using the DerSimonian and Laird random effects model [42]. Because of controversy about the accuracy of reporting of off-label use of one of gabapentin [37,38], we relied on data in McCrory's reanalysis of misleading data presented in one of the studies [43] based on drug company trial data. The main focus of our analysis is between active treatment and placebo controls. We also included data from comparative effectiveness trials. In addition to direct comparisons between drugs, we also conducted a network meta-analysis [44][45][46][47]. In brief, network meta-analysis asks if one drug has a pooled efficacy compared to placebo of X and another drug has a pooled efficacy compared to placebo of Y, are X and Y statistically different? We only included drugs with at least 2 clinical trials and at least 8 weeks in duration, adjusting for duration and for correlation between outcomes reported from the same trial. Because these studies did not always report their outcomes in frequency of headaches, the network meta-analysis was done using standardized mean differences (SMD) rather than weighted mean differences.
Heterogeneity was assessed visually using Galbraith plots [48], and I-square [49].We assessed for small study effects (publication bias) using the methods of Peters [50] for dichotomous outcomes and Eggers [51] for continuous ones. We explored the potential source of heterogeneity using stratified analysis and random-effects meta-regression [52]. These analyses included assessment of the impact of quality, study duration, percentage women, losses to follow-up, and drug dose. All analyses were done using STATA (v 13.1, College Station TX). There was no external funding for this study.

Placebo Comparisons
Overall, the studies varied in quality. Quality ratings for placebo controlled trials are given in Table 4. By Jadad criteria, 34% of studies had scores 3.0, suggesting low quality, 39% had scores between 3 and 5 consistent with modest quality and only 37% had scores 5 suggesting high quality. Only 36% used an intention to treat analysis, 27% assessed compliance, 26% had concealed allocation, and 51% had adequate blinding. There was no difference in the overall effect sizes for placebo controlled trials using Jadad criteria as a scale (p = 0.44) or when coded as high, modest or low quality (p = 0.37), or when assessed by most of the specific Jadad or Cochrane Risk of Bias quality characteristics (compliance p = 0.59; blinding p = 0.36; adequacy of blinding p = 0.50, industry sponsorship p = 0.52; incomplete outcome reporting p = 0.96, reporting of withdrawals p = 0.24). However, trials which had inadequate concealed allocation had significantly (p = 0.02) higher reported effects (SMD: -0.52, 95% CI: -0.63 to -0.41) than those who had concealed allocation (SMD: -0.26, 95% CI: -0.34 to 0.17).
Alpha-blockers. There were 9 trials comparing alpha blockers to placebo with a total of 4590 participants who averaged 39.3 (range 12-76) years in age with 84% women ( Table 2). All of the studies measured headache frequency. Eight of these trials focused on episodic migraine headaches; all studied clonidine. One trial focused on chronic migraines using tizanidine. The average rate of withdrawals was 32%. Studies averaged 11 weeks (range 4-82) with a mean of 71.3 participants (range 11-67). At no time point was clonidine more effective than placebo for episodic migraines (Table 5, Fig 2) and tizanidine was no more effective than placebo for chronic migraine headaches ( Table 6). None of these trials reported on the likelihood of a 50% reduction in headaches. Angiotensin Converting Enzyme Inhibitors (ACE)/ Angiotensin Receptor Blockers (ARB). There were three ACE (captopril, enalapril, lisinopril) and three ARB (candesartan x2, telmisartan) placebo-controlled trials, all focusing on episodic migraines ( Table 2). The ACE studies were 8, 12 and 16 weeks in duration with 120 participants who averaged 7.3 headaches per month. All three ARB studies were 12 weeks in duration with a total of 231 participants, averaging 6.5 headaches/month. One of the ACE trials suggested no benefit at 4 or 8 weeks (enalapril), another found benefit at 12 weeks (lisinopril) and a third benefit at 16 weeks (captopril, Table 5, Fig 3); none of the trials reported outcomes at a common time-point. At twelve weeks, ARBs were better than placebo in reducing the frequency of headaches (Table 5, Fig 3). The likelihood of experiencing at least 50% improvement was not reported in all clinical trials. One of the ACE trials (captopril) was more likely than placebo to achieve at least a 50% reduction in headache frequency (Table 7). This was not found in the trial studying lisinopril or for two of the ARB trials.
In single trials, several anticonvulsants were no better than placebo for episodic migraines including acetazolamide, carbamazepine, carisbamate, clonazepam, oxcarbazepine and vigabatrin (Table 5). In single trials, lamotrigine was found effective at 4 weeks though ineffective at 12 weeks (Table 5). In several trials, gabapentin was not superior to placebo ( Table 5). Several of these anticonvulsants were assessed for ability to reduce headaches by 50% (Table 7). Carisbamate was less effective than placebo and anticonvulsants no more likely than placebo to reduce headaches by at least 50% included acetazolamide, gabapentin, lamotrigine, levetiracetam and oxcarbazepine.
Anticonvulsants that were found to be more effective than placebo for episodic migraine included levetiracetam (Table 6), topiramate (Fig 4) and valproate (Fig 5). Both topiramate and valproate had numerous trials demonstrating benefit at multiple time points (Table 5).
Topiramate. Topiramate has been evaluated in twelve placebo-controlled trials that reported outcomes at numerous time points and different doses (50, 100 and 200mg). Pooled results suggest that topiramate was more effective than placebo at all time points (4-24 weeks, Table 5) and at all doses assessed. There was evidence that higher doses of topiramate was more effective than lower ones, with a stepwise increase as the dose increased from 50 to 100 to 200mg (Fig 6). For chronic migraine, 2 studies of topiramate suggested effectiveness for up to 16 weeks (Table 6). In several studies (n = 8) topiramate was also demonstrated to be more effective than placebo at reducing migraine by more than 50% (Table 7).
Valproate. Valproate also had been compared to placebo in six trials with multiple time points and varying doses (500-1500mg). Valproate was found to be more effective than placebo for episodic migraine at all time points assessed including 4, 8 and 12 weeks (Table 5, Fig 5). However, unlike topiramate there was no evidence of a difference in response to increased doses (dose-response p = 0.83). Valproate was also found in numerous trials (n = 5) to reduce headaches by more than 50% (Table 7).
Calcium Channel Blockers. Calcium blockers headache trials tested cyclendelate (n = 1), nicardipine (n = 1), nifedipine (n = 2), nimodipine (n = 5) and verapamil (n = 2). All studies focused on episodic migraines ( Table 2). Overall there were a total of 878 participants who  . No calcium channel blocker was more effective than placebo, including cyclendelate, nicardipine, nifedipine, nimodipine and verapamil (Table 5). When the dihydropyridines (nicardipine, nifedipine, nimodipine) were pooled, they were no better than placebo at reducing headaches. Flunarizine. While classified as a calcium channel blocker, flunarizine has no influence on blood pressure and its side effect profile suggests that its site of action is on cellular receptors other than the calcium channel [231,232]. Flunarizine is not available in the United States. There were 7 studies of episodic migraines, totaling 332 participants ( Table 2). Studies averaged 47 participants, 36.4 years in age, 77% women, 12.5 weeks in duration and 9% dropouts. Four studies reported headache frequency and three reported headache outcomes based on a headache index. Flunarizine was superior to placebo at 8 and 12 weeks ( Table 5, Fig 8), though not at 4 weeks. Only a single trial reported the likelihood of a 50% reduction in headache with flunarizine with insignificant results (Table 7).
Selective Serotonin Reuptake Inhibitors (SSRI)/ Selective Norepinephrine Reuptake inhibitors (SNRI). There were six SSRI and one SNRI placebo controlled trials, five focusing on migraines and 1 on chronic daily headaches. There were a total of 335 participants who  Table 2). The average rate of withdrawals was 25% (range 0-41%). Study duration averaged 12 weeks (range 8-20) with a mean of 56 participants (range . Specific drugs tested include three SSRIs (femoxitine, n = 1, fluoxetine, n = 4 and sertraline, n = 1), and one SNRI (venlaxafine, n = 1). Four of the SSRI trials reported a headache index. One SSRI trial and the SNRI trial reported frequency of headaches per month. For migraine headaches, two SSRI's, femoxitine and sertraline, were no more effective than placebo while fluoxetine was effective at 12 weeks (Fig 9). A single trial of venlafaxine found benefit at 8 weeks (Table 5). For chronic daily headache a single trial of fluoxetine found no benefit (Table 6). Only a single trial (fluoxetine) investigated the likelihood of reducing headaches by at least 50% and found no benefit over placebo (Table 7).
Serotonin Antagonists. Pizotifen is a serotonin antagonist, commonly used for migraine treatment in the 1970's and 80's. There were 9 placebo controlled trials with a total of 600 participants and all focused on episodic migraine headaches ( Table 2). The average rate of withdrawals was 20% (range 0-48). Study duration averaged 8 weeks (range 4-12) with a mean of 67 participants (range . Two studies reported a headache index, the other 7 headache frequency. Pizotifen was superior to placebo at all time points (Fig 10, Table 5). No trials reported on the likelihood of achieving at least 50% improvement in headaches.

Tricyclic Antidepressants (TCA)
There were 8 trials comparing a TCA to placebo, one focusing on chronic daily headaches, the remainder on episodic migraine headaches. There were a total of 1570 participants. The average rate of withdrawals was 37% (range 20-52%). Study duration averaged 10 weeks (range 4-24) with a mean of 143 participants (range 10-554). Tricyclic's studied included          amitriptyline (n = 5), clomipramine (n = 2) doxepin (n = 1) and opipramol (n = 1). Four trials reported headache frequency and 4 used a headache index as their outcome measure. For episodic migraines, amitriptyline, clomipramine and doxepin were better than placebo ( Table 5, Fig 11), while opipramol (Table 5) was ineffective. Amitriptyline was the best studied TCA (Fig 12), though two of the studies were only 4 weeks in duration. Amitriptyline was more likely than placebo to produce a 50% reduction in episodic migraine headaches (Table 7). A single trial found amitriptyline ineffective for chronic daily headaches (Table 6).

Comparative Effective Trials
There were a total of 60 trials with comparisons between different prophylactic drugs for headaches, 55 including subjects with episodic headaches, five with chronic migraine headaches. Not all prophylactic drugs were directly compared with each other (Table 8). Quality ratings for these trials are given in Table 9. Drugs that were frequently compared to other active drugs include amitriptyline, metoprolol, pizotifen, propranolol, topiramate and valproate. There were few differences in effectiveness between the different drugs. Amitriptyline was no more effective than SSRIs, venlafaxine, topiramate or propranolol. Among beta-blockers, metoprolol was superior to clonidine, flunarizine and nifedipine and propranolol was better than femoxitine. Propranolol was equivalent to metoprolol, atenolol, nadolol as well as to flunarizine and topiramate (Table 10). Among the anticonvulsants, topiramate was equivalent to flunarizine, lamotrigine and to valproate and valproate was equivalent to flunarizine. For chronic migraines, popranolol was better than nortriptyline.

Network Meta-analysis
Candidate drugs for the network meta-analysis were those drugs found effective for treatment of episodic migraine headaches with at least 3 randomized clinical trials. These included eleven different drugs used in prophylaxis of episodic migraine headaches (Fig 13). Indirect comparisons of these eleven individual drugs using meta-regression suggested that amitriptyline was more effective than several of the other drugs including candesartan (p = 0.04), fluoxetine   Fig 12), and no different than atenolol (p = 0.20), flunarizine (p = 0.06), clomipramine (p = 0.15) or metoprolol (p = 0.15). The network meta-analysis found no differences between the other drugs in the relative effectiveness in the prophylaxis against migraine headaches. (p = 0.21).

Placebo effect
There were 78 studies that provided baseline headache frequency that included 4579 episodic migraine sufferers who were randomized to placebo. On average, patients randomized to the placebo group experienced 5.3 (95% CI: 4.9-5.8) headaches/month at baseline. Patients receiving placebos experienced a significant decline in headache frequency by 4 weeks, an effect that persisted through 12 weeks. By weeks 16, 20 and 24, the number of headaches experienced by patients given placebo increased back to values that were not different than baseline (Fig 14).

Side Effects
Patients receiving prophylactic treatment were more likely than those receiving placebo to experience side effects (RR: 1.27, 95% CI: 1.19 to 1.37) and to withdraw from treatment (RR: 1.18, 95% CI: 1.08-1.29). The specific side effects varied by study medication (Table 11). Drowsiness was the most common side effect, increased among patients taking gabapentin, pizotifen, topiramate, TCA and valproate. Tricyclic antidepressants also caused dry mouth and weight gain. Beta-blockers were associated with feeling depressed, dizzy and insomnia. Topiramate increased rates of nausea and paresthesia. Pizotifen had marked increased rates of weight gain with participants averaging 4.3 kg (95% CI: 3.0-5.6). Network meta-analysis and direct comparisons found no difference in likelihood of experiencing "any" side effect or in the rate of withdrawing from studies.
The amount of heterogeneity varied considerably among the various drugs and drug classes. Longer duration of treatment was associated with greater effects for tricyclic antidepressants (β = -0.06, 95% CI: -0.09 to -0.03) as well as for valproate (β = -0.02, 95% CI -0.04 to -0.01) and flunarizine (β = -0.03, 95% CI -0.07 to -0.001). The other treatment options did not appear to be time-sensitive. There was no relationship between type of measurement (frequency vs. headache index) and outcomes (p = 0.72). Age, percent women, sample size, dropout rate, percent of maximum dose attained, study design and whether or not depressed patients were allowed to participate had no relationship with outcomes.

Discussion
There has long been consensus that some drugs are useful in prophylaxis against migraine headaches. Our review confirms that there is good evidence for amitriptyline, atenolol, flunarizine, fluoxetine, metoprolol, pizotifen, propranolol, timolol, topiramate and valproate in reducing episodic migraine headache. At baseline, episodic migtaine sufferers averaged slightly over six headaches per month and most drugs reduced the number of headaches by 1 or 2 per month. Amitriptyline had the greatest benefit and while the network meta-analysis suggested that it was the most effective drug for preventing migraine headaches, this was not confirmed in clinical trials in which amitriptyline was directly compared with other drugs (including SSRIs, topiramate and propranolol), though all candidate drugs have not been included. Betablockers (atenolol, propranolol, timolol), anticonvulsants (topiramate, valproate), flunarizine and pizotifen had moderate benefit in reducing headache burden while the serotonin reuptake inhibitors had a small effect. On average, across the effective prophylactic medications, migraine sufferers had about twice the chance of experiencing at least a 50% reduction in headaches as those receiving placebo. Our pooled risk reduction (ARR: 0.15, 95% CI: 0.09-0.21) suggests that 7 people would need to be treated to produce 50% reduction in headache burden in one subject. Side effects were common, but were predictable based on the drug mechanisms of action and are well-known.
There was a significant placebo effect that was seen within 4 weeks of placebo initiation with a gradual increase in the benefit of placebo on headaches through 12 weeks. By week 16, patients randomized to placebo had a gradual increase in the number of headaches experienced with no difference from baseline through 24 weeks of treatment. This is similar to the placebo effect we saw in our meta-analysis of pediatric migraine trials [233]. Uncontrolled trials of drugs for treatment of migraine headaches are still published, our data reinforces the importance of placebo controls.
Our study is the first to pool all the data from the numerous randomized controlled clinical trials to explore potential differences for both continuous and dichotomous outcomes and for both episodic and chronic migraine headaches. We also avoid a common error found in previous meta-analyses in which researchers pooled the outcome at the end of the study, regardless of the time point. This inappropriately pooled studies of different treatment durations.
There have been no previous systematic reviews of ACE/ARB, flunarizine or beta-blockers other than propranolol for migraine headaches. A recent Neurology Academy review was limited by several factors: 1) it included only studies since 2009, 2) it provided only qualitative statements about the level of evidence with no formal pooling of data and 3) it had no comparative effectiveness data [27]. While our findings are similar to previous reviews of anticonvulsants [234], the beta-blocker propranolol [235], anticonvulsants [236] and tricyclic antidepressants [237], we found some important differences. Anticonvulsants were less effective than a 2004 Cochrane review [234], though our review includes nearly twice as many studies. A 2004 Cochrane beta-blocker review included exclusively propranolol, while we include all beta-blockers. Our 2010 TCA review [237] inappropriately pooled both migraine and tension headaches together. Our 1996 review [238] also combined migraine and tension headaches, likely inappropriate given potentially important pathophysiologic differences. A 2005 Cochrane review of SSRIs found no benefit [239], but that trial was largely based on tension headaches and it also combined both migraine and tension headaches in their pooled analysis. In contrast, our larger review focuses on migraine headaches and suggests a modest effect from fluoxetine. To date, there have been no quantitative systematic reviews comparing the different classes of treatment, though one recent qualitative systematic review concluded that the choice should be tailored to patients based on side effects and comorbidities [240].
A recent systematic review examined the efficacy of prophylactic treatment for episodic migraine headaches [28] in reducing headaches by 50%, a dichotomous outcome. Our study includes both continuous and dichotomous outcomes and examines the effects for both episodic and migraine headaches. That study was limited to English language only and includes a smaller number of studies than this analysis. Our results are similar and in agreement with their conclusion that there is no difference in efficacy between the different drugs; however we found that the benefit for most drugs was less than they reported.
Our study has a number of important limitations. First the pooled differences between the various drugs and classes suggested important clinical differences. Some drugs had a large effect in headache reduction, others only small or modest ones. Our network meta-analysis suggested superiority for amitriptyline, a finding not confirmed in head-head trials. While there have been 51 trials directly comparing different drugs, these comparisons have been somewhat haphazard and many important potential comparisons have not been made.

Conclusions
Our data suggests that the current practice of tailoring prophylactic medication according to patient characteristics and expected side effects is a good approach. Patients with migraine headaches and hypertension should consider trials with a beta blocker. Patients with depression may benefit from either SSRI or TCA. Patients with restless leg syndrome or another indication for an anticonvulsant may benefit from topiramate or valproate. Our analysis suggests that amitryptyline is more effective than the other medications, this has not been confirmed in the limited number of direct comparative effectiveness trials that have been conducted. The placebo effect, that lasts through at least 12 weeks in our study, suggests that non-placebo controlled trials should not be performed. Nearly all studies of headache treatment were 24 weeks or less in duration, this is an important limitation since migraine is a chronic condition. Whether treatment benefit persists, increases or wanes is unknown and deserving of further studies. The paucity of head-to-head comparative effectiveness trials between some classes of medication also indicates a direction for future headache research.