Abstract
Methodological aspects of study design affect therapeutic outcome. We tested the hypothesis that treatment effect of an active drug in studies on acute therapy of migraine is lower in placebo-controlled studies than in those not using a placebo-controlled design. From 522 eligible studies on acute therapy of migraine cited in Pubmed database which were published between 1964 and 1997 we randomly selected 100 studies for evaluation. We excluded five studies because they did not include a quantitative measurement of pain intensity. Of the 95 studies included in the analysis, 61 used placebo control. Response to active drug was significantly lower (P < 0.05) in placebo-controlled studies (61% [lower and upper quartiles 44% and 75%] vs. 71% [56% and 75%]). We did not find any significant effect of quality of study design on net effect of verum. This observation should be taken into account for future planning of controlled clinical studies in acute migraine.
Introduction
The standard design for evaluation of treatment effects is a placebo-controlled trial. This design allows a separation of the active pharmacological effects of a drug from its placebo effect (1).
Placebos have been reported to improve subjective and objective outcomes in pain therapy in about 30–40% of patients with a wide range of responses (2). Explanations for the mechanism of action of placebo in pain research include natural history and regression to the mean as well as non-specific influences. These non-specific influences include physician attention, the sense of interest and concern provided by a healing setting, as well as patient and physician expectations of treatment effects (3). Recent studies have focused on the impact of non-specific influences of placebo in pain therapy (2, 4). The extent to which the patient or clinician believes in the effectiveness of a therapy may influence the power of non-specific effects (5). When using placebo control in studies on pain therapy expectations of success of therapy by both clinicians and patients anticipating a potentially powerless placebo may be diminished. Therefore we tested the hypothesis that the total drug effect is lower in those studies using a placebo when compared with those trials designed without placebo control in studies on acute therapy of migraine.
Methods
Literature search and selection of studies
We searched in the Pubmed database for trials published between 1964 and 1997. The search was performed using the keywords placebo, acute therapy, and migraine. From 522 eligible studies on acute therapy of migraine cited we excluded studies performed on children, adolescents, or with subjects suffering from menstrual migraine, and randomly selected 100 studies for evaluation.
From that pool we excluded five studies which did not include quantitative assessment of pain intensity. Therefore, 95 studies were included in this prospective analysis. Sixty-one studies used a placebo, defined as an intervention labelled as such in the report of a clinical trial (2), while 34 did not.
Extraction and synthesis of data
The main outcome criterion of our study was the response to active treatment, defined as the decrease of pain intensity (percent of control). If more than one drug was tested in a study of acute migraine therapy, we only analysed the response to the first drug applied for comparison. Data are presented as median [quartiles]. The appropriate two-sided statistical null and alternative hypotheses were:
H01: P11 = P12
H11: P11 ≠ P12
where P11 is the proportion for the effect of active drug in studies using placebo and P12 is the proportion for the effect of active drug in studies without a placebo-controlled design.
The global level of significance in a two-sided hypothesis was α= 5%. According to the protocol, the hypothesis was to be proven with the non-parametric Mann–Whitney U-test.
Furthermore, statistical measures and exploratory P-values were used to investigate the influence of the quality of the study design on the verum response rate. We used the following criteria which we considered critical to the quality of the study design, and assigned 0s and 1s as points: not-randomized (=0) vs. randomized (=1) study, not blinded (=0) vs. blinded (=1), unicentre (=0) vs. multicentre (=1) design, and the number of patients included (n < 50 = 0, n ≥ 50 = 1). From these data we built a quality score with 0 indicating the worst quality, and 4 indicating the best quality observed. We used the χ2 test for comparison of the dichotomous variables blinding and placebo control and we used the non-parametric Mann–Whitney U-test to compare with the verum response rate the dichotomous determinants of quality of study design. Finally, we tested the relationship between the quality score and the verum response rate by means of non-parametric correlation (Pearson).
Results
Response with active treatment was significantly lower (P < 0.05) in placebo-controlled studies of acute migraine therapy compared with those studies not using placebo control (61.5 [quartiles 45 and 73.5]% vs. 71 [56 and 82.25]%, respectively; Fig. 1).
Plots of quartiles (boxes), median (line within box), 10th, and 90th percentiles (error bars, outliers given as open circles). Median of verum response is significantly (P = 0.022) lower in studies using placebo control (61 [45, 73.5]% compared with those studies not using a placebo 71.5 [quartiles 56, 82.25]%).
Quality scores were significantly higher (P < 0.01) in studies using placebo control (median of 3 vs. 1, respectively), but the quality did not correlate with the response rate.
Comparing response rates of active drug with the single variables determining the quality of study design Table 1), only the effect of blinding approached but failed to reach statistical significance (P = 0.068). Response rates with active drug in double-blind studies were 62.5%[47.1 and 75] compared with 72%[54 and 82] in those studies which did not blind to the test drug applied.
Quality of study design and success rate of verum
Distribution of the determinants of quality of study design and response rate reported. We compared with the verum response rate each dichotomous variable ((not) blinded, (not) randomized, (not) > 50 patients included, (no) multicentre study). Response rate of verum did not significantly depend on the variables determining the quality of the study design.
The variables blinding and placebo control were significantly related (P < 0.01). Therefore our data did not allow study of the effect of placebo control on verum response independently from the effect of blinding.
Discussion
The main finding of this analysis is that placebo control reduces response rates of the active drug in controlled clinical studies on acute migraine therapy.
Critique of methods
A high placebo effect in a study population may be useful to detect a significant influence of placebo control on the response to the active drug. In a recent meta-analysis the placebo effect was most distinct in studies on the treatment of pain (2). Other studies indicate that high patient compliance is also associated with a high placebo response (3). The compliance of patients to drug therapy is markedly higher during an acute migraine attack compared with migraine prophylaxis (6). Therefore, we speculated that studies on acute migraine therapy may be best to test our hypothesis.
From the whole sample of 522 studies on acute migraine therapy available, we randomly selected 100 studies reflecting a representative sample for analysis. Using this technique, no selection bias was expected and the power was sufficiently high to detect a difference in response to active drug between studies using a placebo and those which did not.
Methodological aspects of study design may affect therapeutic outcome (2, 7, 8). Therefore, we analysed the variables considered relevant to the quality of study design (2, 7) and their influence on the efficacy of the active drug. Since no scoring system for headache trials is established, we constructed a score without previous validation and also investigated the influence of each single determinant on response to the active drug. We found no relationship between the quality score and the response to active drug. The correlation between blinding and efficacy almost achieved significance (P = 0.068). As the variables placebo control and blinding were significantly related (P < 0.01), our data did not allow study of the influence of placebo control on active drug response independently from the effect of blinding.
Interpretation of results
A placebo effect is a change in a patient's illness attributable to a symbolic import to a treatment rather than a specific pharmacological or physiological effect (9). A placebo effect is expected in any therapeutic intervention and does not require a placebo tablet (3). There are two general reasons why the placebo effect may induce clinical improvement in a patient's condition: natural history, and regression to the mean. In addition, there are non-specific reasons including various psychological and social influences. In pain treatment it is well documented that the placebo effect exceeds natural history and regression to the mean (2), as measurable changes of biological variables are reported after placebo. In fact, endogenous opioids have been associated with the analgesic placebo response (10, 11). Moreover, recent findings indicate that effective placebo treatment induces changes in brain function that are different from those associated with drug medication (12). These measurable effects may be modulated by various psychological and social phenomena such as patient–doctor relationship, classic conditioning and learned behaviour, the expectation of relief, and the will and belief of both patient and physician (13).
Our data suggest that the placebo effect is less pronounced if the patient and the clinician anticipate a possible placebo treatment, which they may expect to be a less powerful treatment. Thus, placebo effect may be higher in those studies not using any placebo.
Due to ethical considerations, it is suggested that the use of placebo in studies on pain therapy is justifiable only in high-quality trials (14). Our data support this statement, since therapeutically outcome of verum is reduced in studies using placebo control. Therefore, when pointing out the specific risks of participation in a placebo-controlled study on migraine therapy, these observations should be taken into account.
Footnotes
Acknowledgements
We are grateful to W. Laughey and M. Eikermann for their useful comments and criticism.