Abstract
The aim was to determine the magnitude of the nocebo (adverse effects following placebo administration) in clinical trials for primary headache disorders. We reviewed randomized, placebo-controlled studies for migraine, tension-type headache (TTH), and cluster headache treatments published between 1998 and 2009. The frequency of nocebo was estimated by the percentage of placebo-treated patients reporting at least one adverse side effect. The dropout frequency was estimated by the percentage of placebo-treated patients who discontinued the treatment due to intolerance. In studies of symptomatic treatment for migraine, the nocebo and dropout frequencies were 18.45% and 0.33%, but rose to 42.78% and 4.75% in preventative treatment studies. In trials for prevention of TTH, nocebo and dropout frequencies were 23.99% and 5.44%. For symptomatic treatment of cluster headache, the nocebo frequency was 18.67%. Nocebo is prevalent in clinical trials for primary headaches, particularly in preventive treatment studies. Dropouts due to nocebo effect may confound the interpretation of many clinical trials.
Introduction
The term nocebo (‘I shall harm’) was introduced in contraposition to the term placebo (‘I shall please’) by Kennedy in the early 1960s in order to distinguish the noxious from the pleasing effects of placebo (1). Nocebo is, therefore, the antipode of placebo, and may result from the patient’s apprehension that medical treatment will harm instead of heal (2). Nocebo effects encompass both non-specific side effects that cannot be explained by the pharmacological action of a drug and symptoms that resemble those expected of the active drug (2,3).
More recently, these terms have been redefined to denote only those symptoms powered by psychological factors. If the positive psychological context that mediates the placebo effect is controlled or reversed, then the nocebo effect can be studied. Additional factors also contribute to symptom changes in patients treated with placebo such as the natural time course of the condition under investigation, co-interventions or bias. The nocebo effect results from the negative psychological context surrounding the treatment, and includes both pretrial suggestion and previous negative experiences during treatment. Recently, neurobiological investigations have begun to explore how this negative context influences both qualitative experience (the drug response) and neurological information processing (2). By recording fMRI signals during the evaluation of painful stimuli, Benedetti and his colleagues have begun to investigate the patterns of activity characteristic of nocebo-related hyperalgesia and placebo-related analgesia. These studies implicate an interaction between the stress response mediated by the hypothalamic–pituitary–adrenal axis (HPA) and limbic regions like the hippocampus, amygdale, nucleus accumbens, prefrontal cortex, and cingulate cortex involved in associative learning (4–10). Blockers of cholecystokinin (CCK) attenuated hypo-algesia in this nocebo model (5,7). Significantly, nocebo and placebo effects were associated with opposing changes in the activity of several neurotransmission systems, including dopamine, opioids, and β-endorphins (9).
Alternatively, little is known about how these nocebo effects influence therapeutic responses in the clinic. Reports from clinicians (11) indicate that the nocebo effects are very prevalent, but the exact magnitude remains elusive. It appears that nocebo side effects vary by disease, and that conditions characterized by chronic pain may potentiate the nocebo effect (7,10,12,13). In an extensive review of the literature, psychological characteristics like anxiety, depression, and tendency to somatize, together with prior adverse reactions to medication, were recognized as key predictors of nocebo (14). Since nocebo may increase treatment discontinuation, non-adherence, and treatment failure, it is essential to estimate nocebo’s exact size and strength in each disorder. Greater insight into nocebo could lead to the development of treatment strategies that ameliorate its effects in clinical practice and improved clinical trial design (14,15).
In the field of headaches, Reuter and colleagues found that up to one-third of migraineurs treated with placebo experience adverse side effects. In trials that tested the therapeutic efficacy of triptans for migraine, 21 ± 9% of control patients reported at least one side effect from the placebo. Symptoms were grouped into three categories: migraine-related (symptoms like nausea, photophobia, and phonophobia), drug-related (symptoms typical of the experimental compound like chest pressure in response to triptans), and non-specific or coincidental (symptoms like sleep disturbance). Thus, symptoms in the placebo group were related to the drug under study and to the symptomatology of migraine, whereas some others had no obvious relation to the condition or treatment (16). In another review aimed at estimating the placebo response in migraineurs treated with oral triptans, it was found that 23.40 ± 14.05% of participants treated with placebo reported side effects. Interestingly, studies performed in North America showed a higher nocebo frequency than those conducted in Europe (17). Recently, the Benedetti group published an extensive systematic review of nocebo in clinical trials for migraine (18). They investigated the adverse events after placebo in migraine trials testing non-steroidal anti-inflammatory drugs (NSAIDs), triptans, or anticonvulsants and found that nocebo mirrored the adverse effects expected of the active medication studied. In other words, nocebo in migraine trials arose from patients’ distrust (18).
The aim of the present study was to determine the magnitude of nocebo in trials for all primary headache types, including all available drugs for both preventive and symptomatic treatment, in order to alert practitioners and trial designers to additional explanations for treatment discontinuation and treatment failures. We estimated both the nocebo frequency and its strength (dropout frequency) in clinical trials for all primary headache disorders published during the past decade.
Subjects and methods
We performed a systematic literature review of randomized, controlled, clinical trials for headache treatments that documented noxious side effects and dropouts in the control group. The MEDLINE database was searched for English–language articles from September of 1998 to 26 September 2009 using the Medical Subject Headings (MeSH) shown in Figure 1. We performed six consecutive searches for clinical studies focused on either acute or prophylactic (preventative) treatment of migraine, tension-type headache, and cluster headache. The search followed the PRISMA recommendations (19). Two independent reviewers (LIM and NGC) screened all references retrieved. In the last phase of filtering, all articles meeting previous criteria (placebo-controlled, randomized clinical trials investigating headache drugs) were fully reviewed and further processed for statistical analysis based on two specific inclusion criteria: headache classification according to IHS criteria (ICHD-I or ICHD-II), and report of CONSORT flow diagrams. Reviewers also included studies under the non-IHS classification in the case of chronic daily headache and transformed migraine. The former was classified as TTH and the latter as migraine. When patients with both headache subtypes participated in a study, the study was included in the analysis for both migraine and TTH. Studies with either cross-over or parallel design were included. On the other hand, studies were excluded when: (i) they did not report safety data in sufficient detail (no CONSORT flow diagram or no documentation of side effects); (ii) the data were from previously published sources; (iii) they were not written in English; and (iv) the Jadad score for quality of the article was <3. The reviewers agreed on inclusion assignment for all articles.
Flow diagram of the study selection process. A and B: search for migraine symptomatic and prophylactic treatment trials. C and D: search for tension-type headache symptomatic and prophylactic treatment trials. E and F: search for cluster headache symptomatic and prophylactic treatment trials.
Quality assessment
The quality of reports was classified using the Jadad scale because it is considered the most reliable (20). This scale includes only three dimensions – randomization, blindness (of participants, care-givers, and investigators), and adequate documentation of withdrawals and dropouts (21).
Data extraction
Data were extracted from each study in a structured coding scheme that included citation information and country or countries where the study was performed, headache subtype, sample size, median age and sex ratio of participants, drug dosage, treatment duration (in the case of preventive treatments), study design (parallel or cross-over), number of patients treated with placebo, number of patients treated with placebo who experienced side effects, and placebo-treated patients who withdrew due to side effects (information gathered directly from CONSORT flow charts). To estimate the frequency of nocebo in these trials, we calculated the percentage of patients treated with placebo who reported at least one side-effect. In this evaluation, any side-effect reported by patients treated with placebo was considered as nocebo. The frequency of nocebo dropouts was estimated by the percentage of patients treated with placebo who discontinued the treatment because of intolerance. These estimates were calculated separately for both symptomatic and preventative studies, and for each drug studied, headache type, root of drug administration, and year of publication. Each reviewer made a separate Excel data file and both files were compared and reviewed by a third reviewer (DDM) to produce the final data set.
Statistical analysis
StatsDirect statistical software was employed to perform the meta-analysis to calculate pooled estimates across selected trials. Pooling was conducted using the DerSimonian Laird method (random effects model) (22). The heterogeneity within trials was tested by Cochran’s Q-test based on inverse variance weights (23). The I2 statistic was also used to quantify the extent of inconsistency in outcomes across studies. In the current analysis, only random effects models which assume that the variability between effect sizes is due to sampling error plus variability in the population were considered. The effect of publication and selection bias on the summary estimates was tested by the Harbord–Egger bias indicator (24) and by the Egger indicator (25). Student’s unpaired t-test was used for comparing the nocebo frequency between North American and European studies, and for oral versus parenteral administration. The strength of the association between dropout rate in placebo groups and active drug groups in topiramate studies was measured by the Pearson product moment correlation coefficient (weighted by the population size). Finally, to compare placebo and active treatment dropouts, mixed effects models using treatment as fixed and study as the random factor were used.
Results
Literature search and nocebos
The results of our MEDLINE search are summarized in Figure 1 and the results of our analysis are summarized in Table 1. After repetitive filtering, only 56 articles studying acute symptomatic treatment for migraine were analyzed. Statistical analysis revealed that the nocebo effects and nocebo dropouts frequencies for all trials were 18.45% (95% CI 14.90–22.23%) and 0.33% (95% CI 0.19–0.53%), respectively. In trials studying triptans, nocebo effects and nocebo dropout frequencies were 20.93% (95% CI 16.46–25.78%) and 0.36% (95% CI 0.18–0.61%). In trials with oral drugs, nocebo and dropout frequencies were 19.82% (95% CI 15.84–24.12%) and 0.33% (95% CI 0.17–0.55%). In the migraine prevention era, 45 trials were analyzed (Figure 1). Nocebo effects and nocebo dropout frequencies were 42.78% (95% CI 34.73–51.36%) and 4.75% (95% CI 3.28–6.45%). Nocebo frequencies between trials for botulin toxin type A (BTX) and topiramate did not differ significantly except that nocebo dropouts in trials for BTX were significant lower than the average for all other prophylactic anti-migraine treatments (0.22% vs 4.755%; Table 1). Only four trials were retrieved that tested drugs for TTH prevention; in these trials, nocebo effects and nocebo dropout frequencies were 23.99% (95% CI 4.61–52.20%) and 5.44% (95% CI 1.32–12.12%). We did not find sufficient data to analyze trials for symptomatic treatment of TTH. In the case of cluster headache, only a few studies fulfilled the search and review criteria (Figure 1). For symptomatic cluster headache treatment, four trials were analyzed and the nocebo frequency was estimated at 18.67% (95% CI 10.65–28.33%); insufficient data were gathered to calculate the nocebo dropouts. All relevant proportion meta-analysis plots (random effects) for prophylactic treatments are presented in Figures 2–4, together with bias assessment plots, Cochran Q values, inconsistencies (I2), pooled proportions, Egger biases, and Harbord biases. Plots for symptomatic treatments are not presented.
Relative frequency of patients experiencing any side-effect while treated with placebo (nocebo frequency) in trials for migraine prophylaxis. (A) Proportion meta-analysis plot (random effect). (B) Bias assessment plot. Relative frequency of dropouts due to adverse events (nocebo dropout) in placebo-treated patients in trials for migraine prophylaxis. (A) Proportion meta-analysis plot (random effect). (B) Bias assessment plot. Relative frequency of patients experiencing any side-effect while treated with placebo (nocebo frequency) in trials for tension-type headache. (A) Proportion meta-analysis plot (random effect). (B) Bias assessment plot. Nocebo frequency and dropouts in trials for primary headaches Nocebo frequency: percentage of patients treated with a placebo who reported at least one side effect. Nocebo dropouts: percentage of placebo-treated patients who discontinued treatment due to intolerance.
Stratified analyses
In migraine studies, nocebo effects and nocebo dropout frequencies were higher in preventive trials than in symptomatic trials (P < 0.001). Additionally, nocebo frequency varied by year of publication in trials for symptomatic treatment of migraine, decreasing from 22.05% (95% CI 16.46–28.21%) for trials published within 1998–2004 to 14.39% (95% CI 10.81–18.39%) for trials published within 2005–2009 (P < 0.001). Nocebo did not change with route of drug administration and no differences were found between studies performed in North America compared to Europe. In studies for migraine prevention with topiramate, we compared dropouts in the placebo group versus dropouts in the active treatment group; nocebo dropouts were lower in the placebo group (mean difference 7.09%; 95% CI 4.1–10.1%; P < 0.0001). In addition, dropout rates were strongly associated in both treatment groups (placebo and topiramate (r = 0.824; P < 0.0001).
Discussion
Synopsis of results
Our meta-analysis revealed that approximately 20% of migraineurs assigned to the placebo group in symptomatic drug trials experienced adverse side effects (nocebo), but that only a small fraction (less than 1%) discontinued treatment. Patients suffering from cluster headache showed similar patterns. On the contrary, in chronic treatment studies, the nocebo effect was much more prevalent and powerful; almost half of headache sufferers, whether from migraine or TTH, reported nocebo side effects and about 5% withdrew from the study. Nocebo rates did not vary with the drug tested, with headache type, or by continent. Only in the symptomatic migraine treatments did nocebo change over time.
Meta-analyses of the nocebo effect
Our findings confirm and extend the finds of several recent meta-analyses of nocebo. In the meta-analysis of Amanzion et al. (18), nocebo dropout in triptans studies was only 0.39%, which is comparable to the 0.36% derived by our analysis. Furthermore, the nocebo dropout frequency for preventive anticonvulsants (topimate, valproate, and gabapentine) was 7.71% compared to 6.75% in the topiramate studies that we analyzed. In trials with oral triptans, nocebo frequency was estimated at 23.40% (95% CI 9.35–37.45%) by Lober and colleagues (17), similar to the 20.93% (95% CI 16.46–25.78%) reported by our analysis. Interestingly, in Lober’s review for trials published between January 1991 and March 2002, studies conducted in Europe reported lower side-effect rates than studies conducted in North America (17). These findings emphasize the need for more accurate evaluation of number needed to treat (NNT) and number needed to harm (NNH) for efficacy and safety comparisons across different trials. Due to cultural differences, study location may influence placebo effects, particularly in anxiety-mediated disorders like blood hypertension and ulcer disease (26). Our meta-regression analysis, however, did not confirm this finding.
Implications for clinical trial design
Clinical trial data and basic human research utilizing nocebo models like hyperalgesia clearly indicate that both baseline psychological profile (level of anxiety or depression due to past treatment failures) (27) and pretrial suggestion can increase negative side-effects (28). We found that nocebo effect with concomitant trial dropout was common in preventative drug trials while dropout was relatively rare in symptomatic trials. This difference may relate to the psychological profile of chronic headache sufferers as discussed below. The length of treatment per se may also be significant, as preventative studies obviously require longer treatment periods. There is also considerable instability of nocebo effect over time, but whether this reflects changes in trial design or the participants’ behaviour over time remains unclear. Stratified analysis demonstrated that nocebo did not vary with trial phase (although the majority of studies examined were phase-IV). Participants may become more familiar with the drug family (e.g. triptans) even in cases where they were naïve to the particular drug investigated. Old drugs in new combinations may also explain this familiarity. The new anti-migraine calcitonin gene-related peptide receptor (CGRP) antagonists did not differ significantly in nocebo, although the ratio was smaller arithmetically (the nocebo frequency was 36.2% and 30.9% in two studies included in analysis vs 42.78% for all prophylactic treatments). Thus, these recent trials may contribute to understanding the effects of trial duration in limiting nocebo.
It is believed that CGRP-antagonists exhibit an improved safety profile compared to triptans (29). The efficacy of the active drug often correlates with the placebo effect (positive outcome in the control group) (30), suggesting that better safety of the active drug may result in a lower nocebo effect. To investigate this issue, we performed a post-hoc analysis in all studies of the anticonvulsant topiramate. We found that dropouts in the placebo arm were indeed associated with dropouts in the active drug arm. This is in line with recent findings from experimental human studies indicating that nocebo may arise from verbal suggestions of negative outcome (4,29). In clinical trials, investigators deliver safety information to all participants in the some way; thus, all subjects are equally aware of the possible side-effects. Cued and contextual conditioning may also increase nocebo frequency, although the effects of these forms of pretrial conditioning appear to augment the placebo effect (analgesia) more than the nocebo effect (hyperalgesia) (4).
We also investigated the role of drug delivery. In trials with BTX, some investigators reported that symptoms could be a consequence of the injection itself (e.g. swelling and redness at the site of the injection). These side-effects have no known connection to administration of placebo. For this reason, we compared injection and oral placebo, and found that placebo injections caused fewer side-effects and fewer dropouts than oral placebo. However, this difference may be related with the treatment frequency (once every three months for the placebo injection compared to every day administration for the oral placebo).
Since the nocebo effect may play an important role in clinical trial outcomes, new recruiting policies are essential to avoid biased selection, particularly in studies of behaviour-modifying drugs or when comparing behaviour and drug-based treatments (31). Nocebo should be taken into account in sample size calculations and data safety reports (32), but more importantly, investigators should scan trial participants for characteristics predisposing to nocebo even before randomization to prevent high dropout rates. Thus, while patient numbers in the placebo group should be minimized, this ethical concern must be balanced against the scientific rationale underpinning study design.
Is the nocebo effect common in clinical practice?
Our data clearly demonstrate that the nocebo effect is a confounding factor in clinical drug trials, but the prevalence of an analogous nocebo effect may even greater in the clinic. Indeed, patients participating in clinical trials may not accurately reflect the patient population treated in daily clinical practice. Patients who are reluctant to receive novel medical treatments due to anxiety or general mistrust might avoid participating in clinical trials completely. For this reason, a nocebo effect, whereby patients experience adverse side effects seemingly unrelated to the pharmacological activity of the medication, may be common. Indeed, nocebo may be more frequent among headache sufferers in the general population. Co-morbidity with anxiety disorders, neuroticism, somatization, hypochondria, and depression increases the negative expectation for possible drug side-effects (6,27,33) and these traits are characteristic of chronic headache sufferers (34–36). This may also explain why the nocebo effect was more prevalent and stronger in studies of preventive headache treatments than in trials of symptomatic treatments; preventative or prophylactic trials focus on chronic sufferers.
It was recently demonstrated that the adverse events in placebo arm are qualitatively similar to the adverse effects of the active medications (18). For example, nocebo in the control groups of anticonvulsant trials was characterized by anorexia and memory loss, a side-effect profile similar to the treatment group. This is a key finding and contrasts with the previously held belief that nocebo effects were simply non-specific and unexpected side-effects following medical treatment.
Since nocebo is associated with the safety profile of the active drug, one should expect more nocebo side-effects when administrating drugs with poor safety profiles. Based on our clinical experience, nocebo patients are those that have experienced failure with pharmacotherapy, and so may have lower expectations that they will respond to a new drug. Reading the drug brochures or Internet information on drug safety without proper re-assurance from their physician may increase nocebo. It is important, therefore, for practitioners to search for characteristics predisposing to nocebo and tailor strategies for delivering safety information (28). Discussing the likely risk for nocebo and explaining the phenomenon to the patient may help. In addition, face-to-face follow-up could help to minimize patients’ fears and maximize their understanding of the pharmacotherapy benefits.
Limitations of the meta-analysis
The heterogeneity observed in the frequency of adverse side-effects and dropouts due to intolerance among the selected trials limits the accuracy of pooled estimates for nocebo. As in all meta-analyses, our study was also prone to selection bias. Although meta-regression and stratified analyses uncovered some potential sources of bias, it should be stressed that the type of data used for the nocebo analysis necessitated significant culling of the literature. More standardized reporting of clinical trials may reduce the number of trials that must be excluded, and so reduce these sources of bias. In addition, new methodologies to assess nocebo within trials are warranted. To apply a proper comparator for determining nocebo, side-effects could be compared with those observed in a ‘no intervention’ group (no drug or placebo). Additionally, some placebos like saline injection are not truly inert and should be avoided.
To pool homogeneous data, this study was limited to the past 10 years. Older preventive medications that are still currently prescribed (like propranolol) were under-represented while newer drugs with limited clinical use may be over-represented. Finally, this study provides evidence-based information regarding nocebo in clinical trials, not in clinical practice. Documented knowledge to prevent nocebo in trial designing and clinical practicing is missing, although we believe that the observed side effects in the placebo arm of clinical trials underestimates the occurrence of nocebo when active pharmacological treatments are given.
Conclusions
Nocebo may be a serious confounding factor in clinical trials for primary headaches, particularly in studies of preventive treatments. Clinicians should be aware that drug intolerance and treatment failure may be caused by nocebo. Nocebo does not change with the type of headache, but co-morbidity with affective disorders typically increases nocebo. Nocebo includes both unexpected (non-specific) side-effects and symptoms that are specific to the tested drug. Thus, the proper delivery of drug safety information is crucial and clinicians should use tailored strategies to prevent nocebo. In the field of clinical science, the potential risk for nocebo supports the use of non-treatment arms in trials for headaches. This study also underscores the need for precise and homogeneous safety reports, allowing improved meta-analyses that will further clarify factors contributing to this phenomenon. Single trials that control contextual variables and assess prior patient history are required in order to identify the specific factors contributing to nocebo. Nocebo limits the therapeutic effects of medical treatment in headaches, while placebo increases them. To improve headache therapeutics, nocebo warrants greater attention, both in clinical practice and in clinical trial design.