Abstract
Objective
The objective of this article is to see whether the effect of candesartan for migraine prevention, shown in one previous study, could be confirmed in a new study, and if so, whether the effect was comparable to that of propranolol (non-inferiority analysis), and whether adverse events were different.
Methods
In a randomised, triple-blind, double cross-over study, 72 adult patients with episodic or chronic migraine went through three 12-week treatment periods on either candesartan 16 mg, propranolol slow-release 160 mg, or placebo. The main outcome measures were days with migraine headache per four weeks (primary outcome), days with headache, hours with headache, proportion of responders (>50% reduction of migraine days from baseline), and adverse events.
Results
In the modified intention-to treat-analysis, candesartan and propranolol were both superior to placebo: 2.95 (95% confidence interval: 2.35–3.55%) and 2.91 (2.36–3.45%), versus 3.53 (2.98–4.08%) for migraine days per month (p = 0.02 for both comparisons, Wilcoxon's paired signed rank test, blinded statistical analysis). Candesartan was non-inferior to propranolol (and vice versa). The proportion of responders was significantly higher on candesartan (43%) and propranolol (40%) than on placebo (23%) (p = 0.025 and <0.050, respectively). There were more adverse events on candesartan (n = 133%) and propranolol (n = 143%) than on placebo (n = 90%), and the adverse event profiles of the active substances differed somewhat.
Conclusion
It is confirmed that candesartan 16 mg is effective for migraine prevention, with an effect size similar to propranolol 160 mg, and with somewhat different adverse events.
Trial registration: EUDRACT (2008-002312-7), ClinicalTrials.gov (NCT00884663).
Introduction
Migraine is the most common neurological disorder, imposing a large burden on many of the sufferers (1,2) and also on society (3). The majority of patients have relatively rare attacks which can be treated with acute medication. However, a sizeable proportion of patients fulfils criteria for prophylactic treatment, but use of such treatment is limited (4), possibly owing to lack of effectiveness, contraindications, or side effects of currently available drugs (5).
Among the most commonly used preventive drugs are beta-blockers, of which propranolol (PRO) has been most studied and is the most used active comparator for new migraine prophylactics. In 2003, our group reported that the angiotensin II receptor1A blocker candesartan (CAN) was effective for migraine in a cross-over study (6). CAN is now used for migraine in many countries (see e.g. Silberstein et al. (7), Evers et al. (8) and Stark and Stark (9)), in spite of the limited evidence for its efficacy.
The present study was undertaken to test whether it was possible to confirm that CAN is superior to placebo (PLA) as a migraine prophylactic, and if so, whether the effect was comparable to that of PRO, evaluated by a non-inferiority analysis. Also, it was an aim to compare the side effect profiles of the two active substances.
Materials and methods
The study was designed as a placebo-controlled double-blind, double cross-over trial, with a four-week open baseline period, and three 12-week treatment periods with a four-week wash-out period between each treatment period. It was conducted at the neurological outpatient clinic of St. Olavs University Hospital between April 2009 and March 2012. Patients were recruited either from patients referred to the clinic, or among those who contacted the study nurse after advertisements in newspapers or on the Internet, or after information on a national TV channel.
Proportion answering yes to a predetermined list of questions about AEs recorded during the consultation in week 10–11 of each treatment period (SD).
AEs: adverse events; SD: standard deviation; PLA: placebo; CAN: candesartan; PRO: propranolol.
Inclusion criteria
Inclusion criteria were: age 18–65 years; signed informed consent; migraine with or without aura (10), or chronic migraine (11); in retrospect, ≥2 migraine attacks per month during the last three months before inclusion, and ≥2 migraine attacks during the four-week baseline period documented in the diary; debut of migraine ≥1 year prior to inclusion, and before the age of 50.
Exclusion criteria
Exclusion criteria were: interval headache not distinguishable from migraine; chronic tension-type or other headache occurring on ≥15 days/month; pregnancy, nursing or not using contraceptives in fertile women; heart conduction block or other significant abnormality on electrocardiogram; heart rate <54 (sitting, after three minutes' rest); asthma or diabetes; decreased hepatic or renal function; hypersensitivity to active substances; history of angioneurotic oedema; psychiatric illness; use of daily migraine prophylactics less than four weeks prior to start of study; having tried ≥3 prophylactic drugs against migraine during the last 10 years; previous use of PRO or CAN in adequate doses (≥16 mg or ≥160 mg) and duration (≥6 weeks); previous discontinuation of CAN or any beta-blocker because of AEs; current use of antihypertensive medication; use of rizatriptan 10 mg tablet; regular ergotamines or opioids use; consistent failure to respond to any acute migraine medication; alcohol or illicit drug dependence.
At the screening visit to the neurologist, a thorough history was taken to ascertain compliance with inclusion and exclusion criteria. All subjects had a neurological examination, blood pressure (BP) and pulse measurement, and electrocardiogram. During the whole study, patients were to keep a headache diary recording relevant attack variables, as well as AEs or other health-related condition.
The four-week baseline period without medication before randomisation was used to ascertain whether patients had ≥2 attacks per month. If <2, the baseline period was extended by four more weeks, and if there were ≥4 attacks during the whole eight-week period, the patient could be included. After the baseline period, there was a new visit to evaluate whether the patient could be included, before randomisation and dispensing study medicines for the first period.
Adherence was defined as using the medicines >2 months in accordance with the prescription, and it was measured using self-reports and pill counts.
Seventy-two patients were randomised in blocks of 12 to one of six possible treatment sequences. The study drug consisted of tablets with either CAN 8 mg or PLA, or capsules of either PRO slow-release formulation or PLA. In week 1 and 12, the patient took one tablet and one capsule each day (CAN 8 mg or PRO 80 mg or placebo), and during weeks 2–11, two tablets and two capsules (CAN 16 mg or PRO 160 mg or placebo). After two weeks in each treatment period, all patients were contacted by telephone by the study nurse who checked for AEs and compliance with the protocol. If there were AEs after increasing the dose, the dose could be reduced to one tablet and one capsule per day for the rest of the period. If the drug was not tolerated at all, it could be discontinued for the rest of that period, and the patient could start on the next period after 12 weeks and still be included in the modified intention-to-treat (mITT) analysis. In week 10 or 11, the patients visited the doctor for evaluation, and they were subjected to heart rate variability and baroreceptor sensitivity tests (to be reported in a separate publication). Adherence was checked and study drug for the next period was also handed out. Approximately one month after the end of study period 3, there was a final visit.
AEs were recorded in the headache diary (free text, one line for each day), asked about at telephone calls and visits and also registered after an open question about side effects and other health-related complaints. An AE reported several times during one treatment period was counted only once.
The study was approved by the local ethics review board (where the full trial protocol can be accessed on request:
Randomisation, blinding and data handling
Participants fulfilling criteria for randomisation were consecutively given a randomization number (1–72) assigning them to one of the six treatment sequences (Figure 1) according to a computer-generated list, pre-made by the company producing the drugs, but unknown to participants, clinicians and statistician. Randomisation numbers were pre-printed on study medication labels, and on three sealed envelopes containing information about the medication in each period for each participant. Hence, in case of serious AEs (SAEs), it was possible to unblind a single period. These envelopes were kept in a limited-access area. PLA tablets and capsules were identical to those with active medication and packaged in identical bottles.
Flow of participants through the study (figures refer to number of patients).
Adverse events (AEs) (n of patients reporting the AE ≥1 time during the treatment period).
Number of patients who took at least one dose of the substance.
p< 0.05 compared to PLA.
p≤ 0.001 compared to PLA.
p< 0.05 compared to CAN, difference in proportions (standardised normal deviation). PLA: placebo; CAN: candesartan; PRO: propranolol.
Data entry was performed by the study nurse. After completion of the study, the data file, together with the unopened envelopes containing randomisation codes, were handed over to personnel at the Unit for Applied Clinical Research at the Faculty of Medicine who opened the codes and returned the file with each treatment type having a code (A, B or C). A predetermined statistical protocol had been written for the analysis of the primary and secondary efficacy variables, and the statistician (TS) performed analysis of this file without knowing the actual type of treatment. This was revealed first after tables with efficacy data had been created (triple-blind study). A few data entry errors were detected during the blinded analysis, and these were corrected before the final unblinding.
The study was monitored by an employee at the Norwegian University of Science and Technology (NTNU) Department of Neuroscience who had experience and formal training in monitoring clinical trials.
Statistical methods
The predetermined hypotheses were: H1: CAN is better than PLA (superiority analysis). H2: CAN is not inferior to PRO (non-inferiority analysis). H3: CAN has fewer AEs than PRO (superiority analysis).
A power analysis before study start indicated that 60 evaluable patients would give >80% power to detect a difference between CAN and PLA of 0.5 standard deviation (SD) (two-sided test) for the main efficacy variable. Expecting a drop-out rate of 15%–20%, we decided that 72 patients would have to be included.
All statistical tests were between treatment periods, and did not include baseline data (except for analyses of responder rate). In accordance with the predetermined statistical protocol, H1 (CAN-PLA-difference) was tested with Wilcoxon's paired signed rank test, and likewise the secondary comparisons (CAN-PRO and PLA–PRO). The differences in responder rate and AE proportions were tested with a standardised normal deviate. SYSTATv11 (Systat Software Inc, Chicago, IL, USA) was used with two-sided level of significance 0.05, and subjects fulfilling mITT-requirements were included in the main analysis.
H2 was tested in accordance with CONSORT recommendations for non-inferiority analysis (12%), including only PP completers (n = 54). A mean difference of 1.2 days per four weeks was chosen as the cut-off delta value. This was deemed to be a clinically relevant improvement corresponding to an expected standardised difference of 0.35 SD, but clearly less than the effect (0.5 SD) considered in the power calculation for the main superiority hypothesis. In order to select an a priori reasonable delta, we calculated the probability of reaching an inconclusive result, i.e. we calculated power for detecting true inferiority as a function of both delta and the ‘true active comparator difference’. For a reasonably small true active comparator difference = 0.1 SD (or even 0 SD) and delta = 0.35 SD, the power for detecting true non-inferiority was deemed acceptable = 58% (or 83% for 0 SD, i.e. true equality).The expected treatment effect difference between the active drugs was estimated from a previous migraine trial (13).
Results
Of the 72 patients randomised to treatment, 17 were recruited from the outpatient clinic and 55 were among those who contacted the study nurse after media coverage. Flow of participants is shown in the Figure.
Baseline variables of the whole (n = 72), modified intention-to-treat (mITT) (n = 61) and per-protocol (PP) population (N = 54).
Patients who dropped out of the study (n = 10), were partial completers (n = 7), or who reduced dose in one period (n = 3).
CAN: candesartan; PLA: placebo; PRO: propranolol; F: female; M: male.
SAE: randomisation code not broken.
SAE: randomisation code broken.
Efficacy and adverse event variables in different treatment periods, modified ITT-analysis. Wilcoxon's signed rank test, or *test for differences in proportions.
ITT: intention to treat; AEs: adverse events; CAN: candesartan; PLA: placebo; PRO: propranolol; CI: confidence interval; n.a.: not applicable.
None, mild, moderate, severe.
Normal, reduced, markedly reduced, in bed.
From 1: extremely dissatisfied, to 7: extremely satisfied.
From 1: very unpleasant, to 5: not unpleasant at all.
The proportion of responders was significantly higher on CAN (24/56, 43%) and PRO (24/60, 40%) than on PLA (14/60, 23%) (p = 0.025 and <0.050, respectively). Among the 55 who completed both CAN and PRO periods, 13 responded to both treatments, 10 only to CAN, eight only to PRO, and 24 to neither of the two.
There were five SAEs (Table 4, footnote), three pregnancies, and two requiring hospitalisation. Patients reported more discomfort due to AEs on CAN and PRO than on PLA (Table 5). There were more AEs since last visit on PRO than on PLA, but no significant difference between CAN and PRO in this respect. Tiredness and the sum of free-text AEs were also significantly higher on either CAN or PRO compared to PLA (Table 4). Some AEs seemed to be specific for either PRO (bodily pain and low pulse at exercise) or CAN (dizziness and paraesthesia) (Table 2). As to the predefined AEs (Table 1), the sum was higher on both CAN and PRO than on PLA, but also significantly higher on PRO than CAN.
Discussion
This study confirms the previous finding that CAN 16 mg is superior to PLA for migraine prophylaxis, and it shows that the effect is not inferior, in fact almost identical for the primary efficacy variable, to that of PRO. With regard to AEs, sleep problems and/or nightmares, well-known AEs of PRO, were present in two of three patients who dropped out on PRO (Table 4).
As to the predefined AEs (Table 1), CAN was better than PRO. However, in retrospect it must be admitted that these AEs were biased towards typical AEs of beta-blockers. Overall, CAN was different from PLA and similar to PRO with regard to general questions about AEs or other illnesses, level of discomfort due to AEs (Table 5), and number of free-text AEs (Table 2). Hence, we conclude that the previous contention that CAN has an AE profile similar to PLA in migraine patients (6) was not confirmed in the present study. It does seem, however, that AE profiles differ between CAN and PRO (Table 2). The practical consequence of this observation, and the fact that some respond only to CAN and some only to PRO, is that the other drug could be tried even if one was not well tolerated or effective in a given patient.
Strengths of this study are the rather strict accordance with International Headache Society guidelines for migraine prophylaxis studies (10,11), pre-determined hypotheses and statistical methods, and blinded statistical evaluation. In addition, our study allowed direct comparison of the two active substances, both with regard to effect sizes and AE profiles.
Comparison of effects across different studies is difficult because of variations in methodology, effect variables and patient samples. The effect of CAN in this study (14% compared to PLA period, 38% compared to baseline period) is somewhat lower than that seen in the previous CAN study with a similar design (29% and 47% respectively) (6). Similarly, the effect of PRO measured as responder rate in this study (40%) seems to be lower than the 68% in a Cochrane review (14). The lower effect size may be due to the very marked PLA effect, known to be higher with cross-over design (15), in which the patients knew they would receive active medication in two of three periods. Compared to the PLA period in the present study, the reduction on active medication was only approximately 0.6 days per four weeks (17%), whereas it was around 1.9 days (40%) compared to baseline. The clinical usefulness of CAN is perhaps better shown by the fact that there were 20% more responders (i.e. ≥50% reduction in migraine days) in the CAN than in the PLA period.
The cross-over design is accepted in the IHS Guidelines (10), although it has some inherent limitations. Carry-over and period effects could not be estimated with sufficient power in the present design (16), and these effects have accordingly been minimised and probably largely eliminated by using adequate wash-out periods and balanced cross-overs. Both drugs are quickly eliminated from the body (T1/2 ≤ 11 hours) but the possibility of long-term effects cannot be excluded (17,18). Any carry-over effect would tend to make differences between substances less than real and could also make AE profiles more similar.
Another problem with cross-over studies may be high drop-out rates and low adherence to protocol, particularly in studies like the present one which lasted almost a year for each patient. In this study, the drop-out rate was relatively low (15%) and the proportion of PP completers high (75%), probably owing to close follow-up by the study nurse.
The chosen non-inferiority margin (delta) may seem large. Delta is often defined as the ‘smallest value that would be a clinically significant effect’ (12), but there is no definite guideline for choosing its value. Our choice was based partly on our clinical impression that most migraineurs would appreciate at least one additional headache-free day per four weeks, and partly on power analysis to ensure a reasonable probability for avoiding ‘inconclusive’ results for definitely clinically insignificant differences (see figure in Piaggio et al. (12)). What would be the advantage of choosing a smaller delta, of e.g. 0.25 SD? In the case of a larger, possibly clinically significant ‘true active comparator difference’, e.g. 0.2 SD, the probability to falsely conclude with ‘non-inferiority’ will then decrease from 30% to 10%. The downside is that even small true differences like 0 or 0.1 SD may be deemed ‘inconclusive’ with 100–58 = 42% and 100–30 = 70% probability, respectively. Hence, delta has to be chosen as a reasonable compromise between the up- and downsides. It should be emphasised that our actual results for non-inferiority still holds, even with a (post-hoc) delta = 0.2 SD (0.7 migraine days per four weeks) because the estimated 95% CI for the active comparator difference, i.e. 0.58 migraine days per four weeks, is still lower than 0.7 days. We admit, however, that we may have overestimated the value of a non-inferiority analysis in a study of the present size in general, because CIs will tend to be large and interpretation could have been difficult if results had been less clear.
It is interesting, although not surprising, to see the large improvement in the placebo period compared to baseline (Table 5). Also, it is of interest that although the frequency was reduced the intensity of the headaches that did occur in all three periods was not. In addition, the two active drugs, though better than PLA, were not rated all that highly by patients (Table 5, satisfaction with treatment effect). This shows the shortcomings of even the best prophylactic drugs available today.
One may speculate that a higher dose could result in more responders. The choice of dose of CAN (16 mg per day) in this study was based on the previous migraine study that showed effect and excellent tolerability (6), but a dose of 32 mg has been tried with good tolerability in cluster headache (19). A dose-finding study in adults with migraine seems warranted.
For acceptance of the present study as valid corroboration of the previous CAN study, it may be a problem that both studies come from the same research group. It is not known why no other groups have tested CAN for migraine, but problems with funding trials with drugs where the patent is about to expire may be a reason. However, another angiotensin receptor blocker (telmisartan) has been tested for migraine in a parallel group study. The results were in favour of telmisartan over PLA, but not significantly with the pre-determined statistical analysis, possibly because it did not take differences in baseline headache frequency into account, and was underpowered (20). These problems are avoided with cross-over studies in which all patients are exposed to all study drugs, and a much higher statistical power is attained with fewer patients (10). Two uncontrolled and open studies on patients with migraine and hypertension or pre-hypertension have also shown a marked decrease in headache frequency with both CAN (21) and olmesartan (22).
An advantage with CAN is that is has fewer contraindications than beta-blockers and can be prescribed to migraineurs with, for example, asthma, diabetes, or heart conduction block. Hypotheses about the mechanism for the anti-migraine effect of CAN have been proposed previously (23), but on the whole it is poorly understood.
In conclusion, this study confirms that CAN is a good alternative to PRO for migraine prevention. With a somewhat different AE profile and fewer contraindications, this easy-to-use drug may become useful in the treatment of one of the most prevalent disorders worldwide.
Clinical implications
Candesartan is effective for migraine prevention. The effect is similar to that of propranolol. The adverse event profile is different from that of propranolol. Candesartan may become a drug of first choice for migraine prevention.
Footnotes
Author contributions
Lars Jacob Stovner and Trond Sand had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Stovner, Hagen, Sand.
Acquisition of data: Stovner, Hagen, Linde, Tronvik, Aamodt, Gravdahl.
Statistical analysis: Sand, Stovner.
Drafting of the manuscript: Stovner, Sand, Hagen, Linde, Tronvik, Aamodt, Gravdahl
Funding
AstraZeneca provided the drugs and gave an unconditional grant of NOK 500,000 to support the study. The company has had no role in the design and conduct of the study, or in collection, analysis and interpretation of the data. Representatives of the company have reviewed and approved the manuscript.
Conflicts of interest
Lars Jacob Stovner has received speaker fees from GlaxoSmithKline (GSK), honorarium to participate in meetings with Pfizer, and research grants and support from Allergan and AstraZeneca.
Mattias Linde has received speaker fees from Allergan and St. Jude Medical, honoraria from Allergan and GSK to participate in scientific meetings, and research grants and support from AstraZeneca.
Erling Tronvik has received fees for or is currently involved as site clinical investigator in studies initiated by GSK and Abbot.
Anne Hege Aamodt has received an honorarium to participate in meetings with Boehringer Ingelheim and speaker fees from Pfizer and Boehringer Ingelheim.
Knut Hagen is a member of an Allergan international advisory board on chronic migraine and has received a research grant and financial support from Allergan.
Gøril Bruvik Gravdahl and Trond Sand have nothing to declare.
Acknowledgement
We thank Professor Harald Schrader, the lead author of the first controlled study on an angiotensin-converting enzyme (ACE) inhibitor and the first to discover the potential of the angiotensin-II-receptor blocker candesartan for migraine-prophylaxis, for his initiative and help to start the present study.