Can insights from placebo and nocebo mechanisms studies improve the randomized controlled trial?

3.1.1 Meta-analyses

The magnitude and the variability of placebo and nocebo effects will be established via meta-analyses. In short, a meta-analysis is a quantitative statistical analysis of several separate but similar experiments or studies. The effect size and variance for each study are calculated, and the weighted mean of these effect sizes is compared in order to estimate the magnitude of the phenomenon under study, here the placebo or the nocebo effect [38]. In the meta-analyses included in the present dissertation, the effect sizes will primarily be calculated and reported according to Cohen’s d, where a value of 0.2 is considered a small mean effect size, a value of 0.5 a medium mean effect size, and a value of 0.8 a large mean effect size [39] (Studies I–III). Some of these effect sizes will be weighted according to the number of participants in each of the individual studies either via d̅ (Study II) or Hedges’g (Study III) [40], [41], which is a variation of Cohen’s d that corrects for biases due to small sample sizes. These approaches give a weighted effect size with a weighting corresponding to that of a fixed-effect analysis. In a random effect model (Study IX), the average placebo effect is reported together with a confidence interval that also accounts for the heterogeneity between studies [38].

3.1.2 Experimental method

Placebo and nocebo effects are investigated in designs where an inactive placebo/nocebo treatment is tested against no treatment to calculate the placebo/nocebo effect (Studies IV,V,VII, VIII). Yet, as placebo and nocebo effects are related to the context and to patients’ perception of the treatment (cf. definitions below), it is also possible to investigate placebo and nocebo effects in studies in which no inactive treatments are given. This is done via the so-called open versus hidden design, where active treatments (e.g. lidocaine for pain reduction and capsaicin for pain increase) are given either in full view of the patient (open) or without the patient’s knowledge (hidden) and compared to a no-treatment condition [6], [31], [42], [43]. The difference between the open and the hidden administration of the agent represents the placebo component of the treatment, whereas the difference between the hidden administration and the no-treatment represents the efficacy of the pharmacological agent (Study VI).

Psychophysical tests in which a standardized and well-characterized exogenous stimulus is applied to the patient so that he/she can rate the perceived intensity of the sensation will be used ([44], Studies IV,VI). Evoked pain will be induced and based on standardized written statements, patients will be instructed how to rate their perception of pain on visual analog scales or numeric rating scales [44], [45], [46]. The visual analog scales will, for example, be anchored by “no pain sensation” at the left and “the most intense sensation imaginable” on the right, and the patient will be asked to mark the point on the line that corresponds to the perceived pain sensation [45]. The visual analog scales will be used to rate both pain intensity and pain unpleasantness as well as a range of psychological factors such as expectancy, desire, and anxiety, and the verbal anchors will be adapted accordingly [45], [46], [47].

3.1.3 Experiential method

In order to get a better understanding of patients’ perception of the treatment, the experiential method will be used (Studies V and VI). The experiential method combines basic phenomenology with psychometric methods [44], [46], [48]. Patients are trained to relive a brief episode of the treatment session and/or to notice brief episodes of direct experience, such as the experience of the description of the treatment agent or the emotional feelings in relation hereto, and to report these experiences without interpretation and explanation. These experiential reports are qualitatively sampled to determine the common elements within the experience and to characterize the interrelationships among them. The patients are also asked to rate the reports using psychometric methods (e.g. visual analog scales) to investigate whether the reports of the direct experience are in agreement with the quantitative experimental data [46], [49].

3.2.1 The placebo phenomenon

The placebo phenomenon is related to the patients’ or the experimental subjects’ “perception or direct experience of a treatment, i.e. seeing, smelling and hearing verbal information about the treatment as well as actively integrating this sensory information with memories of previous experiences and current expectations” [51], [55], p. 122].

3.2.2 Placebo agent

“All external aspects of the therapeutic intervention that can be perceived by the patient or experimental subject” [56], p. 207]. Please note that this includes sham treatments like sugar pills, but it is not limited to the physical agents [6]], [15].

3.2.3 The placebo analgesia effect

“The measured difference in pain across an untreated and a placebo-treated group or across an untreated and placebo-treated condition within the same group (as in crossover studies)” [6]], [15], [56], p. 207, 57], [58].

3.2.4 The placebo analgesia response

“A reduction in pain in an individual that results from his or her perception of the therapeutic intervention” [56], p. 207]. The experience of receiving a treatment is assumed to be different within an individual across natural history and placebo administration conditions [37].

Recently, however, the placebo response has also been defined according to a treatment response, which is “the change that one observes following the administration of the treatment” [58], p. 102]. Hence, the placebo analgesia response is seen as the change in pain following the administration of an inert placebo agent [58]. This terminology is introduced to differentiate between the well-controlled placebo effects, where the placebo treatment is compared to a no-treatment group/condition, as it is typically done in placebo mechanism studies, and the uncontrolled placebo responses observed following a placebo treatment as typically investigated in RCTs [58]. For the sake of clarity, this distinction and terminology will be used in the present dissertation, although the former definition of the placebo response has been used in some of the studies that are part of the dissertation.

3.2.5 The nocebo phenomenon

This phenomenon was first named by Kennedy [59]. It was originally introduced to describe the negative effects of a placebo treatment [60] and may still be used in this manner [61], [62]. Yet, today it is primarily conceptualized as an independent phenomenon that mirrors the placebo effect [6]], [63]. Accordingly, the nocebo effect is seen as the effect that follows the administration of an inert treatment along with behavioral procedures and/or verbal suggestions that tend to worsen symptoms [64]. Like placebo effects, nocebo effects are differentiated from the natural history of pain [6]], [17]. The definitions of the nocebo phenomenon, agent, effect, and response mirror the definitions listed under placebo.

3.2.6 Pain

“An unpleasant sensory and emotional experience associated with actual or potential tissue damage, or described in terms of such damage” [65].

3.2.7 Nociplastic pain

“Pain that arises from altered nociception despite no clear evidence of actual or threatened tissue damage causing the activation of peripheral nociceptors or evidence for disease or lesion of the somatosensory system causing the pain” [66].

3.2.8 Neuropathic pain

”Pain caused by a lesion or disease of the somatosensory nervous system” [67].

4.1.1 Is the placebo powerless?

In 2001, Hrobjartsson and Gøtzsche published a meta-analysis entitled “Is the Placebo Powerless?”, which involved 114 randomized clinical trials covering 40 different diseases, one of them being pain [68]. All trials included a no-treatment condition in order to allow for a calculation of the placebo effect. Across most of the conditions, no placebo effects were found, but in the 27 studies involving continuous outcome measures of pain, a significant effect of placebo was found as indicated by a pooled standardized mean difference of −0.27 (95% CI: −0.40 to −0.15). Since pain is a subjective experience, and therefore subjectively reported, the authors suggested that the significant placebo analgesia effect was caused by response bias. This meta-analysis roused attention and was by some interpreted as if placebo effects were non-existent [69], [70]. Yet, the meta-analysis and the conclusions were subsequently questioned and debated, for example, the number and the quality of the included studies [71], the interpretation of the data [72], [73], and especially the conceptualization of placebo effects [74], [75], [76], [77], [78]. Hróbjartsson and Gøtzsche conceptualized placebo as “an intervention labeled as such” [68], p. 1595], primarily including inactive agents such as sugar pills, saline injections, and turned off vibrators. Interestingly, the doctor-patient relationship and the patients’ expectations were not seen as part of the placebo effect [68], p. 1599].

4.1.2 Does the design of the study influence the magnitude of the placebo analgesia effect?

In 2002, Vase and colleagues conducted two meta-analyses of placebo analgesia effects published in one article (Study I) to test whether the design/purpose of the study influenced the magnitude of placebo analgesia. In these meta-analyses, all studies also included placebo-treated and no-treatment groups/conditions. The first meta-analysis included 23 double-blind RCTs, and, in line with the findings by Hrobjartsson and Gøtzsche, a small but significant placebo analgesia effect was found as indicated by an effect size of 0.15 (range: −0.95 to +0.57, Cohen’s d). In contrast, the second meta-analysis included 14 placebo mechanism studies and found a large and significant placebo analgesia effect of 0.95 (range: −0.64 to +2.29, Cohen’s d). Noteworthily, the magnitude of placebo analgesia effects was significantly different in the two types of meta-analyses, showing that the magnitude of the placebo analgesia effect is markedly larger in studies that investigate placebo mechanisms than in studies in which placebo is used as a control condition.

The results of the two meta-analyses have been debated and recalculated [79], [80], [81], [82], [83], but the finding that magnitudes of placebo analgesia effects are high in placebo mechanism studies has remained consistent [16], [81], [82]. When the mean effect size in the placebo mechanism studies was weighted according to the number of participants [79], the effect size was 1.14 (Cohen’s d) [81], and when the magnitude was recalculated based on the ambiguities stated by Hrobjartsson and Gøtzsche [80], the effect size was 0.97 (Cohen’s d) [82]; i.e. both above a Cohen’s d value of 0.8, which is considered a large effect [39].

4.1.3 Updated meta-analyses

Both Hróbjartsson and Gøtzsche and Vase and colleagues have conducted new and updated meta-analyses as new studies have become available ([83], [84]; Study II). In 2004 and 2010, Hrobjartsson and Gøtzsche updated their meta-analyses with a total of 44 and 66 clinical trials, respectively, with continuous measures of pain, leading to small but significant pooled standardized mean differences of −0.25 (95% CI: −0.35 to −0.16) and −0.28 (95% CI: −0.36 to −0.19), respectively. In 2009, Vase and colleagues conducted a new meta-analysis of 24 placebo mechanism studies, which yielded a large average magnitude of the placebo analgesia effect as indicated by an effect size of 1.00 (range: 0.12–2.51, Cohen’s d) and an effect size of 0.81 d̅, when the average effect size was weighted by the number of subjects in each study (Study II). Hence, the meta-analyses consistently show that the magnitude of the placebo analgesia effect is low in clinical trials and large in placebo mechanism studies. In addition, a recent meta-analysis of placebo mechanism studies has shown an average effect size of 1.29, Hedges’ g (95% CI: 1.07–1.51), thereby corroborating that the magnitude of the placebo analgesia effect is large in placebo mechanism studies [85].

4.1.4 Meta-analysis of nocebo hyperalgesic effects

With the emergence of well-controlled nocebo studies and reviews hereof [63], [86], [87], [88], [89], [90], it has become possible to conduct a meta-analysis of nocebo mechanism studies to test whether the magnitudes of nocebo and placebo effects are comparable (Study III). Ten studies involving a nocebo and a no-treatment condition were included, yielding a lowest effect size of 0.62 (0.24–1.01, Hedges’ g) and a highest effect size of 1.03 (0.63–1.43, Hedges’ g), so the magnitude of the nocebo hyperalgesia effect appears to be moderate to large, which is roughly similar to the magnitude of placebo analgesia effects. These effect sizes were also highly variable.

4.1.5 Discussion of the magnitude of placebo analgesia and nocebo hyperalgesia

The meta-analyses by Hrobjartsson and Gøtzsche [68], [83], [84], and, Vase and colleagues (Studies I and II) indicate that when placebo is used as a control condition in clinical trials, the average magnitude of placebo analgesia effects is low, whereas it is high in placebo mechanism studies. This finding is not surprising. Clinical trials aim at controlling for placebo effects, not at optimizing placebo effects. For example, participants are typically told via the informed consent that they may receive “either an active pain medication or an inactive placebo agent”, which may lead patients to have uncertain expectations about treatment effects. On the other hand, placebo mechanism studies often focus on the interaction with the healthcare provider, and participants are typically given explicit verbal suggestions to the effect that “this agent has been shown to powerfully reduce pain in some patients”, thereby optimizing patients’ potential for developing expectations of low pain levels. Experimental studies have supported this interpretation of the data (see below).

The magnitude of nocebo effects is moderate to large in nocebo mechanism studies (Study III) and overall the magnitude and the variability of placebo and nocebo effects are roughly similar (Studies I–III). In addition, the magnitude of both placebo and the nocebo effects appears to be largest when the effects are induced via a combination of verbal suggestions and conditioning (Studies I and III).

A predominant finding of the meta-analyses ([68], [80], [83], [84]; Studies I–III) is that the magnitude of placebo and nocebo effects is highly variable (see ranges of effect sizes above). Therefore, instead of debating the exact magnitude of placebo or nocebo effects, it may be more valuable to investigate the factors that contribute to the observed variability. Although the meta-analyses have highlighted some of the factors (e.g. that verbal suggestions combined with conditioning are associated with larger placebo and nocebo effects (Studies I and III) and that long-duration stimuli are associated with large placebo analgesia effects (Study II)), experimental studies are needed to further understand this relationship. Importantly, the meta-analyses indicate that it is relevant to test this, not only in nociceptive but also in hyperalgesic pain states (Study II).

4.2.1 Experimental studies of nociplastic pain

In 2003, Verne and colleagues conducted a clinical trial with the aim of testing whether lidocaine was a valid treatment for patients suffering from irritable bowel syndrome (IBS) [91]. Ten IBS patients were exposed to rectal balloon distention under rectal lidocaine, rectal placebo, and no treatment in a crossover fashion, and in the two treatment conditions the patients were told that they would receive either an active or a placebo treatment. In this study, there was a significant effect of lidocaine over placebo and a significant effect of placebo over no treatment [91].

At the same time, Vase and colleagues conducted the first placebo and nocebo mechanism study in chronic pain patients with the aim of understanding the psychological mechanisms of placebo and nocebo effects (Study IV). Thirteen IBS patients were exposed to no treatment, rectal placebo, rectal lidocaine, and rectal nocebo in a crossover fashion while receiving verbal suggestions for pain reduction (rectal placebo and lidocaine conditions) or pain increase (rectal nocebo condition) and reporting expected pain levels and desire for pain relief. A large placebo effect was found, as indicated by a significant difference between the placebo and no-treatment conditions, and it increased over the 50-min interval of the study (average Cohen’s d=1.79). The pain levels were higher in the rectal nocebo condition than in the no-treatment condition, but there was no significant nocebo hyperalgesia effect. Also, there was no significant effect of lidocaine. Expected pain levels and desire for pain relief accounted for 77% of the variance in pain levels following placebo administration, and, 81% of the variance in pain intensity levels following lidocaine administration (Study IV).

These results show that large placebo effects can be obtained in a chronic pain population. They corroborate the findings of the meta-analyses reviewed above, in so far as the magnitude of the placebo effect was low in the RCT [91], whereas it was high in the placebo mechanism study (Study IV). These experimental studies further substantiate the meta-analyses by showing that verbal suggestions given for pain relief are likely to contribute to the difference in the magnitudes of placebo effects across the two types of studies. The finding that explicit verbal suggestions for pain relief are related to large magnitudes of placebo effects has been confirmed by subsequent experimental studies [15], [32], [33], [49], [52], [92], [93], [94], [95], [96], [97], [98], [99], [100].

Subsequent placebo mechanism studies have also demonstrated significant placebo effects in chronic nociplastic pain conditions like IBS [95], [96], [97], [100], [101]. For example, Kaptchuk and colleagues conducted a study of placebo acupuncture, in which 262 IBS patients were tested under the following conditions: no treatment (waiting list), placebo acupuncture with limited interaction with the healthcare provider, and placebo acupuncture with augmented interaction with the healthcare provider. Both placebo conditions yielded significant placebo effects in relation to no treatment, and the augmented placebo condition yielded a larger effect than the limited placebo condition [97]. Hence, today it is well established that large placebo effects exist in chronic nociplastic pain.

4.2.2 Qualitative studies of nociplastic pain

In order to further understand how patients’ perception of a treatment and of the verbal suggestions given for pain relief influences the treatment, 25 IBS patients were interviewed about their direct experience of receiving a treatment immediately after a placebo treatment or a lidocaine treatment (Study V). All of the patients perceived the treatment as an active treatment, and most of them mentioned the relationship with the doctor, whom they generally described as likeable, as central to their perception of the treatment. The doctor gave the same verbal suggestions to all patients, but the patients perceived this description of the agent differently. The majority of the patients went along with the suggestions and generated positive expectations of treatment efficacy, whereas a minority of the patients did not use the verbal suggestions to guide their expectations of treatment efficacy but were waiting to see if the treatment worked. The placebo effects increased during the session, and at the beginning of the treatment, many patients reported listening to the doctor’s verbal suggestions and matching them with their bodily sensations, whereas toward the end of the treatment they were focused on things outside their body. The majority of the patients felt calm throughout the treatment.

Although the interview data corroborate previous quantitative findings [33], they are to some extent at variance with the results of one of the few other quantitative studies of placebo analgesia effects [102]. This interview was conducted as part of the placebo acupuncture study of IBS patients outlined above [97]. Both this and our qualitative study found that the relationship with the healthcare provider is central. However, patients in the acupuncture study were concerned with whether they were receiving placebo or active treatment and they did not spontaneously report expectations of relief but of hope. There are several factors that may explain the partially different findings between the two qualitative studies; a central one being that the acupuncture study was conducted as an RCT, which may have drawn patients’ attention to the possibility of receiving an inactive treatment. Also, the patients did not know the acupuncturist beforehand, so expectations, which were not directly measured in the study, may have been lower in this study as compared to Study V. Yet, it should be noted that hope is known to be a combination of expectations and desires as it has been defined as a wish or a desire accompanied by an expectation of its fulfilment [103].

4.2.3 Placebo and nocebo effects in chronic neuropathic pain

Petersen and colleagues conducted the first combined placebo and nocebo mechanism study in chronic neuropathic pain patients (Study VI). Eighteen patients who had developed chronic neuropathic pain following thoracotomy were exposed to placebo and nocebo manipulation by an open versus a hidden application of lidocaine and capsaicin, respectively, compared to no treatment. The open administrations were accompanied by verbal suggestions for pain decrease and increase, respectively. Expected pain levels and emotional feelings were obtained, and the patients were exposed to quantitative sensory testing. For the first time, large and significant placebo effects were seen on spontaneous ongoing pain (Cohen’s d=1.01). There were also large placebo effects on evoked types of pain (Cohen’s d=0.89–2.01), which corroborates the findings of a previous study [99], but there was no significant nocebo effect. Expected pain levels significantly predicted the variance in spontaneous pain (53%) and evoked pain (24–35%), and patients had high levels of positive emotions and low levels of negative emotions during the open administration of lidocaine. Expected pain levels did not contribute to the nocebo effect, and during the open administration of capsaicin patients reported medium levels of both positive and negative emotional feelings.

4.2.4 Discussion of central mechanisms of placebo and nocebo effects in chronic pain

4.3.1 Placebo acupuncture under optimized blinding conditions

One hundred and one patients received active acupuncture, placebo acupuncture, or no treatment following third molar surgery. The patients rated expected and actual pain levels during the 30-min test, and they indicated perceived treatment allocation as well as their experience of de qi (in short, the experience of heaviness in relation to needle manipulation) (Studies VII and VIII). There was a large placebo effect (Cohen’s d=0.95, average of the placebo effect halfway through and at the end of the study), which increased over time (Cohen’s d=0.12, beginning; d=0.72, halfway; d=1.18, end of study). Expected pain levels contributed to pain levels following placebo administration, and expected pain levels accounted for larger amounts of the variance in the last 15 min (50–70%) compared to the first 15 min of the study (16–34%) (Study VII). Thus, under optimized blinding conditions, the magnitude of placebo acupuncture is roughly comparable to single-blinded, placebo mechanism studies in acupuncture [133]; Cohens’d=1.03, for global improvement, author’s estimate) and roughly similar to the placebo effects in pharmacological studies ([85]; Studies I and II; d=0.95, d=0.82, and g=1.29, Cohen’s d and Hedges g, respectively). Also, expected pain levels contribute to pain relief following both pharmacological and non-pharmacological medical treatments and in both types of treatment, expectations appear to have a self-reinforcing effect ([33], [95]; Study VII). It has been speculated that verbal suggestions for pain relief may lead to expectations of pain relief, which contributes to the actual experience of pain relief, which again further enhances the expectations of pain relief and so forth ([33]; Study VII).

Yet, 61–68% of the patients and 83–81% of the acupuncturists were able to correctly identify whether the active or the placebo treatment was given, thereby indicating that the penetrating and non-penetrating needles were not fully blinded (Study VII). Still, for patients, it should be noted that the distribution was not far from 50/50. The experience of de qi appeared to be the main factor leading to the correct treatment identification and thereby breaking of the blinding. The circumstance that the penetrating needle gives a sensation of the needle and may elicit de qi makes it resemble an active placebo (i.e. eliciting adverse events similar to the active treatment). The degree of blinding in this acupuncture study is in line with the degree of blinding in pharmacological studies involving active placebos [134]. Yet, the employment of active placebos does not seem to solve the problem of unintentional blinding, and therefore it may be helpful to think of new ways of accounting for patients’ perception of the treatment and for the contribution of placebo factors to the overall treatment outcome.

4.3.2 Additional and/or alternative ways of accounting for perceived treatment

A previous study of acupuncture has shown that the perception of treatment allocation (active or placebo) influences pain levels to a higher extent than the actual treatment allocation [131]. Thus, it is relevant to analyze the data not only according to actual treatment allocation but also according to perceived treatment allocation (Study VII).

In the actual treatment allocation, there was a large effect of placebo acupuncture over no treatment, but there was not an effect of active acupuncture over placebo acupuncture. Yet, when analyzing the data according to perceived treatment allocation, there was a significant effect of perceived active acupuncture over perceived placebo acupuncture as well as an effect of perceived placebo acupuncture over no treatment. Thus, the patients’ perception of the treatment greatly influences the treatment outcome, in so far as active acupuncture only had an effect if the patients believed that they were receiving the active treatment. These findings are in accordance with previous acupuncture studies of pain and surgical studies of patients with Parkinson’s disease [131], [132], and they corroborate these findings by showing that the effects remain after the inclusion of a no-treatment condition. Interestingly, the perceived treatment allocation also seemed to influence the temporal development of the pain levels. Patients who believed that they received active acupuncture experienced significantly decreased pain intensity and pain unpleasantness over time, whereas patients who believed that they received placebo acupuncture experienced increased pain unpleasantness over time. This finding is consistent with surgical studies showing that the effect of the perceived treatment allocation may last up to 1 year after the surgical intervention [132]. Thus, in contrast to previous comprehensions of the placebo response as a short-lasting effect [135], recent studies indicate that the placebo response may be long-lasting and potentially increasing over time [136], [137]. Hence, in order to fully understand the results of a study, it may be helpful to directly ask patients about perceived treatment allocation and to take this information into consideration in the data analysis [17], [37], [131], [132]. In this manner, it may be possible to get an impression of the extent to which unblinding and the placebo component of a treatment may influence the outcome of a study.

4.3.3 Discussion of how unblinding and perceived treatment may influence test results

As outlined above, it is difficult to avoid unintentional unblinding in pharmacological and non-pharmacological studies, even under optimized blinding conditions and even when active placebos are used ([133], [138]; Study VIII). Unintentional unblinding is likely to influence the results. As illustrated by the acupuncture studies, patients who believe that they have received active acupuncture experience decreased pain levels over time, whereas patients who believe that they have received placebo acupuncture experience increased pain unpleasantness over time. In this manner, patients’ perception of which treatment they have received may influence the results independently of the active treatment being administered, thereby potentially leading to incorrect conclusions about the effectiveness of the active treatment. Yet, the influence of unintentional unblinding appears to be complex. In both the acupuncture study and in similar pharmacological studies, there was no effect of the active treatment, although the study was not fully blinded. Thus, it is important to carefully investigate this topic further. The outlined studies indicate that it may not be sufficient to optimize blinding and placebo conditions in order to improve the test of treatments; it may be essential to directly ask patients about their perception of the treatment. Just as placebo effects can be investigated without inactive placebo agents simply through manipulation of patients’ perception of the treatment (as seen in the open-hidden design; [6]], [31], [43]), the contribution of placebo factors to the overall treatment outcome may also be accounted for by tapping into patients’ perception of the treatment [17], [37].