Placebo effects in allergen immunotherapy: an experts’ opinion

Placebo effects are common in medicine. Randomised clinical trials help us to understand their magnitude in different therapies. There are particular problems with placebo effects in allergen immunotherapy (AIT) as it is difficult to blind the active treatment and the endpoints are largely subjective. This may explain why large placebo effects are often found in AIT trials. Patients receiving open label AIT get the benefit of the active and placebo components but it can be difficult to say how much benefit is due to the active component. The use of active placebos has been proposed but brings its own problems (ethical and scientific). An EAACI Task Force has been established to address these issues. Here we review the current literature on the placebo effect in general, with a special focus on AIT trials, and indicate what we believe to be important considerations and unmet needs in AIT trial design.


Introduction
Placebo e ects are common in medicine and the underlying complex neuropharmacological and -anatomical mechanisms involved have mainly been investigated in the eld of pain and Alzheimer's disease [1]. Moreover, placebo e ects play a major role in allergic diseases which also have been demonstrated to be much a ected by this phenomenon [2]. Especially in trials on allergen immunotherapy (AIT) by both the sublingual (SLIT) and subcutaneous (SCIT) route, placebo e ects of large magnitudes have been reported and several hypotheses for these e ects have been suggested [3,4]. ose factors undoubtedly have an important bearing on the design, execution and interpretation of results of clinical trials in AIT. e purpose of a current Task Force (TF) initiative therefore is to review this phenomenon with a number of stakeholders in the eld of AIT comprising clinicians, methodologists, reg-ulatory and patient representatives and to develop an EAACI Position Paper on this topic. e following review overviews the current literature on the placebo e ect with a special focus on AIT trials, but also indicates important considerations and unmet needs in the view of experts' opinions.

Psychological and sociological mechanisms of the placebo e ect
e placebo e ect has been known for many years and goes right back to the writings of Hippocrates. Literally, "placebo" means "I will please", and encapsulates the psychological and physical bene ts of seeking advice and treatment for a problem, independent of the pharmacological e ects of the prescribed therapy [1]. ere is therapeutic value in

Abstract
Placebo e ects are common in medicine. Randomised clinical trials help us to understand their magnitude in di erent therapies. ere are particular problems with placebo e ects in allergen immunotherapy (AIT) as it is di cult to blind the active treatment and the endpoints are largely subjective. is may explain why large placebo e ects are o en found in AIT trials. Patients receiving open label AIT get the bene t of the active and placebo components but it can be di cult to say how much bene t is due to the active component. e use of active placebos has been proposed but brings its own problems (ethical and scienti c). An EAACI Task Force has been established to address these issues. Here we review the current literature on the placebo e ect in general, with a special focus on AIT trials, and indicate what we believe to be important considerations and unmet needs in AIT trial design.
simply talking to someone about your problem, which helps you to debrief and put the problem into perspective. We are social animals: sympathy from a friend (or a physician), and sharing our problems is a deep-seated human instinct and is psychologically helpful. Empathy -the sense that someone else cares about you, and that your concerns are both real and matter to someone else -can be therapeutic. Reassurance that you are not alone helps people to cope. And then sometimes a treatment is o ered which may or may not work, but the whole process of problem sharing and then obtaining advice and treatment is all rolled into the perceived value of each medical consultation [1].
Many diseases, including allergic conditions, show natural variation over time. One of the reasons humans have been successful is that we are good at making connections between events and possible causes.
is is useful and helps us to develop strategies for reacting to similar events in the future. However, it also leads to false attributions of cause and e ect. In crude terms our brains take "I did X and Y happened" and we conclude "Y happened because I did X". And then we continue to do X in the hope that Y will happen, and it can take quite a long time before we nd that the two events are in fact unconnected, especially if the action and desired e ect are turned into a form of superstition or a formal religious belief.
So in our clinical practices we o en meet people who have uctuating conditions, like urticaria and irritable bowel syndrome, who incorrectly attribute ares of their condition to food allergy. But the ability to structure their diet and feel that they can control their condition may be useful, as it gives them back a sense of control, which is psychologically bene cial. Similarly, if a condition uctuates randomly it is more likely that you will seek advice when the condition is aring up, and then as the condition spontaneously improves, you may attri-bute that improvement to any treatment you have just received. If you always take the same treatment (which actually does nothing) whenever the condition ares up and then you spontaneously get better, you will come to believe that the treatment is actually driving the improvement.

Relevance for study designs in the age of evidence-based medicine
Science is meant to be the way that we explore these phenomena and nd out whether there really is a causal relationship. Science works by disproving things -a successful experiment is one where you design a test which shows convincingly that two things that you thought were connected are in fact only occurring together randomly. Unfortunately, in medicine we tend to like studies that con rm our prejudices. We talk about "positive results" and "experiments that worked" when we nd a connection, when in fact an experiment only really works when it disproves the hypothesis [5]. And when it comes to marketing drugs and treatments, there is a natural tendency for companies to promote data which supports their marketing strategy, rather than data which casts doubt on the product's e cacy. is "publication and presentation bias" has the unintended but entirely predictable e ect of making us feel that our treatments are more e ective than they truly are.
To address this we need clinical trials -these help us to assess the true e ectiveness of interventions in terms of who bene ts, how much they bene t by, and which outcomes are a ected. Trials also allow us to assess the harmful potential of interventions and they lead on to decisions on licensing and whether to fund treatments through insurance and state-funded healthcare systems [5]. In some cases the trials show little or no e ect compared to a control group. e history of medicine is littered with treatments for which those with enthusiasm have no controls, while on the other hand those with controls have no enthusiasm.
Double-blind placebo-controlled clinical trials (DBPCCT) are the gold standard for assessing the e ectiveness of therapies [5,6]. e main aim of a DBPCCT is to reduce the scope for bias due to physicians choosing who gets active treatment, bias in reporting outcomes (by the participant and the observer) and bias in reporting (or ignoring) sidee ects. But there is a very real question whether the criteria for a DBPCCT can be met for trials of allergen immunotherapy (AIT).
To perform a DBPCCT properly, you need to have patient groups that are randomly allocated to receive active and control therapy, with blinding of allocation, intervention, observation of outcomes, and analysis of data. You also need a su cient level of symptoms or endpoints in the control group to avoid the possibility of missing an e ect due to lack of signal (i. e. insu cient room to see a reduction in the chosen endpoints). is can partly be addressed by a power calculation, but that presumes that you can estimate the likely e ect size. e power of a study is also a ected by any heterogeneity of response, in terms of both proportion and scale -the greater the heterogeneity of response, the lower the power of a study to show any e ect [5]. e robustness of trial endpoints varies between di erent clinical areas. Some endpoints are relatively immune from subjective in uences. So death is a robust endpoint -it either happens or it doesn't. Measurements of blood cholesterol or renal function are relatively robust -once the sample is taken, the lab guarantees that the measurement is accurate. Blood pressure is slightly less robust and can be a ected by suggestions from the observer, as well as by the beliefs of the patient. Other parameters used in clinical trials are much more open to subjective in uences, including symptom scores, medication scores, measures of quality of life (QoL) and physicians' global assessment which are all dependent on the subject and the investigator applying their judgement to score and grade their internal feelings [6].

Placebo e ects in AIT
In his seminal paper on subcutaneous AIT (SCIT), published in 1914, John Freeman noted that AIT led to "a distinct amelioration of symptoms" and cautioned that "there is a constant tendency to detect such improvement in adventitious uctuations of health." He also noted that "a patient inoculated with one pollen became immune to another", but he was not clear whether this was a true bystander e ect or an indirect consequence of people having less symptoms during the pollen season and therefore becoming more tolerant of other exposures [7]. e rst randomised clinical trial of SCIT was conducted by Frankland and Augustin in the early 1950s. ey did a controlled trial comparing crude grass-pollen extracts with the isolated main protein component in the prophylaxis of summer hay-fever and asthma [8]. Ten years later, Lowell et al. published the rst North American RCT of SCIT [9]. ey matched patients on multiple SCIT in pairs in terms of their symptoms in the ragweed season and then removed the ragweed from one of each pair's vaccine and replaced it with caramelised sugar solution. e study was reasonably blinded as patients still had local reactions due to the other components of their vaccines. By this method they con rmed the allergen-speci city of AIT and that you needed to receive ragweed in your vaccine to get protection against natural exposure to ragweed. ere are some issues in conducting DBPCCT of AIT that are speci c to this treatment. Unlike most of the tablets tested in clinical trials, AIT injections can elicit local and systemic reactions. is makes it di cult to blind subjects to which treatment they are receiving. Most of those who receive active therapy will have some local side-e ects during the trial and, although you can ask patients which treatment they are receiving, it is indisputable that experiencing side-e ects is likely to impact on the perception of symptoms. is in turn may lead those in the active group to report their symptoms di erently.
ose who do not get side-e ects and think they are in the control group may be disappointed not to have received active treatment and the supervising trial sta may pick up who had active treatment and who did not. It is not necessary for all participants to be aware of which arm they are in for the integrity of the DBPCCT to be a ected. With sublingual immunotherapy (SLIT), where the allergen is applied to the oral mucous membrane, local sidee ects are extremely common, a ecting over half of the patients receiving active treatment [10], which further complicates our ability to achieve blinding of allocation in clinical trials of SLIT. Another important bias in AIT trials is the level of motivation of patients to participate and to bene t from treatment [11].
With these caveats, clinical trials of AIT do demonstrate e cacy -with clear di erences between the experience of active and control groups even in the rst season of grass pollen AIT [12,13]. However, these parallel group trials do not tell us how much of a placebo e ect may be operating. Other trials with run-in seasons have shown very large placebo e ects -for example a 2-year trial of SLIT in children aged 6-14 years de ned improvement as a decrease of > 30 % in symptom scores.
is was achieved in 85 % of the active group but also in more than half of the placebo group, but there was no e ect on medication use, which serves as a surrogate marker of e cacy [14]. Other clinical trials of SCIT also designed with an initial run in (baseline) period have demonstrated placebo e ects of 6 to 52 % for di erent extracts (reviewed in [4]).
Conversely, a clinical trial of AIT for cat allergy found very stable thresholds for conjunctival provocation tests in those who received placebo treatment while the actively treated group showed a dramatic reduction in sensitivity [15]. e paediatric SLIT study mentioned above showed a strong placebo e ect on CPT thresholds a er one year of treatment but by the time of the nal assessment a er two years treatment, the placebo e ect had disappeared [14]. Another interesting example of a huge placebo e ect are data recently published on CAT-SPIRE (synthetic peptide immuno-regulatory epitopes) for SCIT in cat allergic patients (CATA-LYST; [16]).
ough results from a preliminary phase II (dose-range nding) trial performed in an allergen exposure chamber were very promising [17], the subsequent eld (pivotal) trial failed in demonstrating a meaningful clinical additional treatment e ect to the placebo e ect itself which already improved patients' symptoms by approximately 60 % [16]. Explanations for this high placebo e ect may be seen in the high motivation of the participating patients, a possible disadvantage of inated baseline symptoms or spontaneous remission of symptoms ("natural desensitization" under daily allergen exposure), but these should be further investigated and predictors of a high placebo response evaluated [1,11,16,18].
A Cochrane review of SLIT trials, which conrmed the e cacy of SLIT in a systematic assessment of SLIT studies, commented that there were some limitations to applying meta-analysis techniques to the older studies because many of the early SLIT studies were rather small, with variable dosage, variable endpoints and limited baseline data, so o en the active and placebo groups were ill-matched. But taking everything into consideration, there were large placebo e ects in most published studies of SLIT [19], whereas another analysis has controversially found only a minimal impact of the placebo e ect in SLIT trials [4].

Challenges and considerations for the future
Nearly 20 years ago, reviewing the eld, we argued that the time for small studies was over, and that what we needed were large randomised DBPC studies, with carefully balanced groups of subjects, whose disease was moderately severe and whose sensitivity to the target allergen was clearly dened. As well as prolonged run-in periods collecting of baseline data to harmonise the groups before allocation to treatment, we proposed that better, more valid clinical endpoints were needed together with an economic evaluation and careful documentation of side-e ects. And ideally the trials should be conducted in a less intensive setting, more typical of where the treatment would be deployed as compared to the usual clinical trials unit [20].
But even when that strategy was followed, significant placebo e ects can be seen in AIT studies [4,21]. While it is possible to make statistical adjustments in order to look at the proportion of subjects who improve more than average, or to back up the symptom data with medication scores [6], the fundamental problem remains of side-e ects potentially a ecting the blinding of the trial. Similar considerations apply to studies using patients' or physicians' global assessments of e cacy. Because a substantial excess of subjects receiving placebo report overall improvement, the scope for showing an e ect of active therapy is reduced. Of course, if there is a true placebo e ect, then once the drug is used in open label studies, the recipients get both the true e ect and the placebo bene t, so the patients bene t from any placebo e ect, even if the funding body may be getting less true bene t per euro expended.
On the observation side, the main problem for DBPCCT design is the subjectivity of the symptom and treatment scores. e lack of agreed objective endpoints or measures of success contributes to heterogeneity of studies but also makes us more dependent on subjective parameters, as compared to other areas of medicine. Analysis of the data is not helped by the non-parametric nature of most scoring systems which use ordinal numbers (0, 1, 2, 3) and then add them and average them as if they were parametric variables, with a single step from 0-1, 1-2, or 2-3 equivalent in terms of clinical relevance. e frequency of visits and the interest of study team add to the potential for taking part in the trial to lead to an improvement in symptoms, without there being any therapeutic bene t from the treatment.
So, in summary, the endpoints for SCIT and SLIT trials are largely subjective [6,22], which is not a problem in itself, but becomes an important issue when the basic criteria for DBPCCT are not met, especially when there is di culty concealing allocation to the active and placebo groups. Although not universally true, subjects o en know which arm of the trial they are in, and those making observations are also likely to pick this up, which introduces potential bias in the assessment process. Some endpoints which are super cially objective, like conjunctival provocation tests, may also be a ected if people know which arm they are in. Remedies for this problem have been di cult to agree: the inclusion of histamine in the placebo treatment is one option, but that predictably causes side-e ects whereas the active treatment causes random side e ects and moreover, there is good evidence that histamine itself is immunomodulatory, through the presence of histamine receptors on the surface of T lymphocytes [23].

Conclusions
Placebo e ects play a fundamental role in the magnitude of e cacy of treatment options in all diseases. In the eld of allergen immunotherapy, several factors impact this e ect which should be taken into account when planning, performing and analysing clinical trials. is review gives a current overview of possible roles of di erent mechanisms a ecting the placebo response in AIT and discusses challenges and unmet needs. ese and other related issues are the subject of a current EAACI Task Force initiative which aims to report consensus-based de nitions and recommendations in a Position Paper.