Abstract
Traditionally, triptan trials require patients to treat migraine attacks only when the headache is moderate or severe on a 4-point severity scale; the primary outcome measure is improvement to mild or no pain (1). There are good methodological reasons for this rather artificial design. It better ensures that migraine attacks are being treated rather than nonmigraine (tension-type headache) attacks. This should minimize placebo response. It also simplifies the assessment of the improvement as all patients start from a same baseline pain rather than from different levels. For, shifts from severe to moderate pain are not similar to shifts from mild to no pain.
In clinical practice, however, patients treat their migraine attacks differently. Not surprisingly, clinicians have been searching for trial designs that better reflect the real world. In this issue of Cephalalgia, Scholpp et al. (2) and Klapper et al. (3) and colleagues report on two randomized prospective trials that study the efficacy of early treatment with triptans, while the pain is still mild. Both studies find high responder rates for improvement to pain-free at two hours post dose. The authors conclude: (i) that their findings extend the results of earlier trials that had been using historical comparisons; and (ii) that ‘early treatment’, when the pain is still mild, affords improved efficacy over ‘late treatment’ when the pain is moderate or severe. The elegant studies from Burstein's group, showing differential responses in patients with and without allodynia, seem to provide a pharmacological basis for the superiority of early treatment (4, 5). But are all these conclusions justified?
A number of methodological and clinical pitfalls need to be considered first, before accepting the hypothesis that treating early affords clinically relevant improved efficacy without increased risk of medication-overuse headaches. These pitfalls will be discussed below.
Historical evidence and confusing terminologies
The historical evidence in favour of treating early; see Scholpp et al. (2) can be regarded as hypothesis generating only. It is based on retrospective analyses of protocol violators who accidentally treated mild attacks and comparisons of responder rates for treating mild headaches in current trials with those for treating severe headaches in previous trials. These comparisons did not account for the different study designs and the higher placebo responses in the ‘mild group’. In fact, the therapeutic gains vs. placebo in this group were smaller than those in the ‘severe group’. Just comparing absolute response rates in different study designs, without accounting for different placebo responses, may be misleading (see below). Furthermore, in the past, ‘treating early’ and ‘treating mild’ have been incorrectly used as interchangeable terms. This ofcourse may only be correct in patients with slowly progressing attacks but not in patients with rapidly progressing attacks.
Comparing different study populations
When comparing different treatments, the baseline characteristics of the study populations (especially disease severity) and other prognostic factors must be balanced. This is the very reason for randomization of the treatments. Thus, studies comparing patients who treat mild and patients who treat severe headaches, run the risk of actually comparing two different study populations: those with slowly progressing vs. those with rapidly progressing attacks. In order to minimize this imbalance only one type of patients should be included in a study and asked to treat attacks either early or late, preferentially in a randomized, within-patient, cross-over design. Only in the slowly progressing type, this will also equate with ‘treating mild’ vs. ‘treating severe’.
The illustrious placebo response and placebo effect
Patients may improve after placebo treatment. This placebo response may be due to different mechanisms. Some patients will improve due to regression to the mean or natural course of the disease, others seemingly improve due a wrong diagnosis (a different disease with a better outcome was treated). Finally, some patients will improve due a true placebo effect, due to endorphine-mediated and psychological mechanisms.
When asked to treat early, migraine patients may acidentally treat short migraine or tension-type headache attacks. This could result in accidental improvement due to natural course rather than pharmacological mechanisms. Expectation of relief, e.g. triggered by information in the patient informed consent that early treatment may be superior, could bias patients and therefore may promote a placebo effect. Thus the effect must be placebo-controlled and the analysis should be based on comparing the therapeutic gains of active over placebo treatment for early vs. later treatment. This could be done in a four-arm parallel-group design, in which patients treat one attack randomly allocated to one of four treatments: (i) early with placebo; (ii) early with verum; (iii) late with placebo; or (iv) late with verum. Even better, but more complicated, one could use a partial parallel-group/cross-over design, in which patients are randomized to treat two attacks early or two attacks late, one with placebo and one with verum. A full cross-over design in which patients treat four attacks would probably be too complicated.
Is pain-free at two hours a misleading endpoint?
The 2-h pain-free endpoint is useful to assess improvement from pain (1). For early treatment, when the pain is mild, we should primarily assess prevention of progression. Just measuring pain severity at two hours is misleading as (i) improvement from mild to no pain hardly seems clinically relevant; and (ii) patients want the attack to be fully blocked and not just delayed beyond a few hours. Obtaining pain-free by two hours post dose and maintaining pain-free for the subsequent 22 h without recurrence of the headache and without using additional medication seems the clinically relevant outcome measure for early treatment trials. This variant of the sustained pain-free endpoint (1) can be used to assess treatments of any baseline severity and reflects a true blockade of the attack. The success rates will ofcourse be lower, but certainly more realistic and clinically relevant than the currently marketed and misleadingly high success rates.
The optimal design for early treatment trials
For an optimal and clinically relevant comparison of early vs. later treatment, the following guidelines could be considered:
outcome measure: pain-free by two hours post dose and for the subsequent 22 h without use of additional pain medication within that time frame;
inclusion of only one type of patients, either those with slowly progressing attacks or those with rapidly progressing attacks;
design:
for a four-arm parallel-group design, random allocation of the patients to one of four treatments: (i) early with placebo; (ii) early with verum; (iii) late with placebo; (iv) late with verum;
for a two-arm parallel-group design, random allocation of the patients to treating two attacks either early or late: one attack with placebo and one with verum in randomized order;
statistical analysis should be based on comparing the therapeutic gains (verum minus placebo) for early vs. late treatment;
conclusions only apply to the type of patients included in the trial and cannot be extrapolated to the other type;
only when patients with slowly progressing attacks have been included, one can assume that the findings may also apply to ‘treating mild’ vs. ‘treating severe’;
for scientific reasons, one may include testing for the presence of allodynia, but by an investigator who is blinded to the clinical status of the patient and the treatment protocol; this will complicate the study considerably.
The new studies
So, how far away from the hypothetical ideal are the two new studies? I am afraid, still quite far. The study by Scholpp et al. (2) fulfilled the criteria for including a single type of patients and for randomization between treating early and late, but used the wrong endpoint and an open design without placebo control. This study seem to allow only limited conclusions. The study by Klapper et al. (3) did randomize between placebo and verum treatment, but only for early treatment; none of the other criteria were met. From this study, it can be concluded that when treating early, zolmitriptan seems better than placebo; no conclusion can be drawn whether this is also better than later treatment.
Migraine treatment with triptans: a race against central sensitization?
In a nice series of studies, Burstein et al. (4, 5) have shown that central sensitization may occur during the course of migraine attacks, as evidenced by the development of cutaneous allodynia in over 70% of their patients. They also showed that response rates dramatically differed between patients who were treated late, once allodynia had developed, and those who were treated early, before the sensitization process had fully developed. But, can we extrapolate these important findings in a limited number of selected patients to the general migraine population? In Burstein's studies, both allodynia and triptan efficacy were assessed unblinded, without placebo control. This complicates the interpretation of the results. Many placebo-controlled clinical trials with triptans have failed to show a difference in efficacy between patients treated very early, so presumably before allodynia had developed, and patients treated late, so presumably with a high proportion of patients with allodynia (6, 7). Admittedly, in these studies the presence of allodynia was not assessed, but judging from the very high proportions of ‘late treated’ patients with allodynia in Burstein's studies, many of the ‘late treated’ patients in the old trials must have had allodynia. The observation that patients with allodynia may still respond well to triptan treatment was recently confirmed by a study by Diamond and Freitag (8). In conclusion, the juror is still out whether late treatment and allodynia predict poor response to triptans. An interesting alternative hypothesis could be that the central sensitization process may increase the risk of recurrence rather than the lack of initial relief.
Do we really need early treatment trials?
Why are we seeing so many of these studies being presented at congresses and submitted to medical journals? I can see a few possible motivations:
to attempt to better estimate the real value of triptans in clinical practice;
to improve the treatment of migraine by providing better treatment instructions;
to ‘market’ the idea that success rates with triptans are actually higher than previously believed.
It is obvious that success rates are highly dependent on the choosen endpoint and design. We should not try to come up with success rates as high as possible, but with a realistic attempt to truly assess and improve migraine treatment. For this, we should focus on clinically relevant questions and outcome measures, and apply sound scientific methods. We must also assess potential disadvantages of the new instructions, e.g. the risk of medication-overuse headaches when patients start treating attacks before they really can recognize the aetiology. So, my advice to those who are concerned with improving the treatment of migraine attacks: ‘ask the right question, do the right trial, apply the right methodology, and use the right clinically relevant endpoint’. Remember the adagium
‘satisfaction is reality divided by expectation’ (RB Lipton; Florence, Italy. April 1999).
If we raise the expectation for patients and prescribers to unrealistically high levels and, even worse, start doing more harm, it may work as a boomerang.
Should we advise patients to treat early?
Intuitively, treating early is an attractive approach for achieving higher efficacy, and the results from trials are promising. Unfortunately, thus far there is no convincing scientific evidence that early treatment affords improved efficacy. Furthermore, increasing the risk of medication-overuse headache is a real risk of treating too early. So, my advice to the patient? ‘Treat as soon as you are sure of developing a migraine headache, but not already during the aura’.