Differences in placebo responsiveness have long been acknowledged and the topic of much empirical scrutiny. Since the onset of dedicated placebo research, the question of who does or does not respond to sugar pills has been discussed (for reviews, see Bootzin and Caspi, 2002; Kaptchuk et al., 2008). For some time, responding to placebos was considered a stigma for various reasons (Parkhouse, 1963; Brody and Brody, 2000). However, since the grounding of placebo responses (PRs) in a less mentalistic and more physiological framework—e.g., with the seminal work of Levine et al. (1978), involving the endogenous opioid system in placebo analgesia –, this stigmatization has partially subsided, at least in academia (Thompson et al., 2009). In recent years, especially with the widespread use of technology enabling the monitoring of brain activity, the responder question has been reconciled with mechanistic approaches. Consequently, interest in prediction of PRs was revived (Geers et al., 2005a) and since sparked a number of dedicated studies. Until recently however, the evidence supporting PR prediction was still considered disparate (Kaptchuk et al., 2008; Benedetti, 2009). A better understanding of placebo mechanisms, and therefore predictability of PRs, will allow for both the maximization and minimization of PRs. In established treatments, maximizing the additional PR is desirable. However, in early phases of pharmacological research, minimization and control of PRs is an important issue, in order to isolate and prove that a new treatment is effective beyond placebo (Enck et al., 2013).

This review is concerned with collating studies of sufficient methodological quality that address prediction of the PR. Terminological uncertainties notwithstanding (Hróbjartsson, 2002), we regard the PR as a symptom improvement caused by a placebo treatment and the context in which it is delivered. This causal relation is considered to be mediated mostly by expectancy and related mechanisms (Finniss et al., 2010; Enck et al., 2013). The core question to be addressed is: “Are the determinants of placebo responsiveness located within or outside of the individual?”

Placebo responsiveness could constitute a stable trait (uni- or multidimensional, e.g., a personality variable or genetic polymorphism). It would affect PRs independently of situational variables. People could therefore be classified on a dimension ranging from “responder” to “non-responder.” This point of view has been especially prevalent during early investigations into placebo responsiveness, but results have been inconclusive (Gryll and Katahn, 1978; Bootzin and Caspi, 2002).

To the contrary, the working model of most clinical trials—but also of many dedicated placebo studies—pragmatically emphasizes the fluctuating, situational determination of PRs. In fact, this is the implicit default assumption when determinants of PRs are not further considered in a study. Situational determinants can be generated outside (e.g., experimental manipulation) or inside the individual (e.g., mood). In this model, everybody is a potential responder, placebo responsiveness is ubiquitous and activated by situational influences (Bootzin and Caspi, 2002).

These simple explanations have merit to a certain degree: On the one hand, temperament, personality traits etc. by definition produce consistent behavioral output even in changing environments (Rothbart, 2007). On the other hand, sufficiently strong situations do have the ability to normalize a range of behaviors (Snyder and Ickes, 1985). However, most psychological characteristics conform to a more interactional view (David et al., 1997). This is reflected in the third basic explanation, which emphasizes how different trait variables interact with different situational variables to explain the extent of placebo responsiveness (Geers et al., 2007). Frequently, the focus of these approaches in placebo research lies on the therapeutic relationship and the quality of the interaction (Shapiro, 1971; Frank and Frank, 1973; Gryll and Katahn, 1978).

The latter concept can be elaborated by pointing out how PRs feed back to influence their determinants in future instances. This emphasizes possible changes in responsiveness over time and thereby the relevance of medical experience and socialization (the long-term dispositional and situational requirements to be met for placebo responsiveness to develop) (Moerman, 2000).

A general overview of the 15 articles is provided in Tables 1, 2. Sub-studies are indicated by consecutive numbers as they appeared in the respective publications, e.g., Heatherton et al. (1989-2) for the publication's second sub-study. Categorization into stable/situational/interactional predictors is tentative, since most psychological phenomena meet the criteria of more than one category. Nine studies concerned with the identification of PR predictors used experimental or clinical pain as a dependent variable (Table 1). Three studies used cardiac parameters; two studies each used affect, auditory discomfort, calorie intake and sleep parameters, while one study each used caffeine symptoms and motion sickness (Table 2). Overlap exists due to multiple outcome variables.

The prediction of PRs has been an endeavor of investigators from the onset of dedicated research into the phenomenon. This review attempted to assess the state of knowledge by applying conservative methodological criteria for study inclusion. Surprisingly few original papers prevailed—all in all, only 20 (sub-)studies met the criteria. While it is possible that the search was overly conservative, the goal was to identify only those studies from which variables that differentially predicted symptom change in the placebo group vs. symptom change in the control group were clearly distinguishable.

Results are heterogeneous, but imply that predictors are rather not found among “classic” trait personality variables (such as extraversion or neuroticism) or emotional response dispositions (such as anxiety). Instead, they are more on the side of cognitive constructs such as self-efficacy, locus of control, and “emotionalized” contingency expectations. From a theoretical perspective, this is intuitive, considering how closely related these concepts are to expectancy (Bandura, 1977). Expectancy is considered a major PR mechanism (Finniss et al., 2010; Enck et al., 2013).

An important cluster seems to revolve around an internality-externality dimension, such that participants follow placebo instructions more readily when having an external locus of control. For example, results by Heatherton et al. (1989) have led the investigators to hypothesize that restrained eaters lack sensitivity to internal states and are therefore more reliant on external cues to regulate behavior. The “permission” implicit in the placebo instructions could provide such a cue. We propose that this readiness is actually tied to low self-esteem (Brockner, 1988), which would explain our own finding concerning predictors of PRs on motion sickness (Horing, 2013). This interpretation is supported by self-esteem moderation of PRs on sleep onset reported by Brockner and Swap (1983).

Many of the mentioned constructs also draw attention to potential sources of bias in repeated symptom measurements and emphasize the necessity of careful experimental design when investigating PRs. This notion is especially prevalent when considering results by Handley et al. (2011) involving the belief in expectation biases in consequent placebo analgesia.

Due to its broad psychological ramifications, sex seems a less tangible predictor of PRs. Debate is still open what relevance it has for placebo responsiveness (Aslaksen et al., 2011). Results presented here are mixed. Both psychological and biological explanations for the positive predictive value of sex are offered by Bjorkedal and Flaten (2012), albeit the paradigm used is rather unusual (if intriguing), limiting comparability. In general, it seems too convenient to restrict oneself to sex for psychologically codetermined response variables like pain. Instead, it seems worthwhile to include measures of (psychological) gender to explore sociocultural influences.

It has been noted that there is not just one PR affecting all symptom modalities, but many (Benedetti, 2009), potential overlap notwithstanding (Petrovic et al., 2005). This could explain some of the heterogeneity—for example, the involvement of optimism in pain and sleep quality (Geers et al., 2007, 2010), but not in stress or motion sickness (Horing, 2013; Darragh et al., 2014). This cautions one not to overgeneralize findings in single symptom modalities, but instead try to describe the qualitative characteristics of the investigated conditions. So far, the data base for a symptom-specific assessment is thin. Pain would be the most promising candidate for an isolated analysis.

As far as the characterization of placebo responsiveness as a trait or state variable is concerned, this review offers little to resolve the issue. We feel that a pattern is discernible which warrants further investigation of stable characteristics such as personality variables. This is in line with other studies in the field (Pecina et al., 2013) and does not contradict more interactional oriented approaches (Kong et al., 2013).

We are confident of the reliability of the selection method. Yet, this review has several limitations that need to be addressed. First, the sample of studies examined for inclusion can be criticized. Research has been mostly constrained to PubMed, resulting in a probable bias for medical literature; more psychologically oriented literature, as e.g., accessible via PsycInfo, was not included. This shortcoming has to be eliminated by a more thorough, dedicated research amounting to a systematic review or meta-analysis. Secondly, publication bias is an important issue in placebo research (Hróbjartsson et al., 2011) and certainly affects this review, as well. Presumably, negative results in particular will be underrepresented. Thirdly, inclusion criteria were rather conservative. Especially the exclusion of cross-over designs proved to be a severe restriction, prohibiting consideration of pertinent studies (e.g., Aslaksen et al., 2011; Pecina et al., 2013).

Yet, this concession is congruent with most authoritative sources conceding a scarcity of reliable data concerning predictors of placebo responsiveness (Kaptchuk et al., 2008; Benedetti, 2009). For example, Hall et al. (2012) implicated the reward system in placebo responsiveness. While they corroborated previous evidence (Leuchter et al., 2009), the latter publication could not be included due to lack of an adequate control. Another article carried similar implications by reporting an association of gray matter density in reward sensitivity related brain areas and PRs (Schweinhardt et al., 2009). Unfortunately, the study used a cross-over design which was not considered for inclusion in this review. Likewise, other studies did not have sufficient validity regarding reward sensitivity (de la Fuente-Fernandez et al., 2001; de la Fuente-Fernandez and Stoessl, 2002; Scott et al., 2007). Most relevantly, de la Fuente-Fernandez et al. (2001) and de la Fuente-Fernandez and Stoessl (2002) only assessed post-instruction associations of placebo responding with reward sensitivity. As to the first study (de la Fuente-Fernandez et al., 2001), sample size was very low (N = 6), thereby not granting sufficient protection against chance results; furthermore, carry-over effects seem likely, as far as the procedure could be ascertained. The second study did use a control group to assess novelty effects of first exposure to the stimulus material, but no control incorporating the effects of measurement repetition (de la Fuente-Fernandez and Stoessl, 2002). Scott et al. (2007) reported a correlation between Nucleus accumbens-activity in placebo analgesia, and after monetary reward. While the results are highly interesting and were well received, the investigators did not include an adequate control group. Therefore, it cannot be precluded that simple measurement repetition would have the same (or at least aligned) effects.

Pertaining to the explanatory value of optimism, some precautions apply to the generalizability of these findings. On the plus side, the effect has been analyzed with excellent methodology. However, results come only from a single research group (Geers and colleagues) and should be replicated elsewhere. Unfortunately, while Morton et al. (2009) investigated optimism as well, they only performed an isolated analysis of the placebo group. It should also be noted that other investigators found conflicting results compared to Geers' group. For example, one study did not replicate the interaction of optimism with a placebo instruction, but only a main effect of optimism (Hanssen et al., 2014). Relatedly, the induction of state optimism was found to decrease pain (Hanssen et al., 2013), even though no placebo was applied—effectively, this amounts to an optimism effect in the control group, which did not show in the aforementioned experiments. Darragh et al. (2014) did not find an association of optimism with the PR on physiological stress parameters, nor did we find an association with the PR on motion sickness (Horing, 2013).

A more basic criticism can be leveled at studies presented by Geers and colleagues. Apparently, the basic efficacy of their placebo paradigms has not been established—in three of five publications, main effects of condition (e.g., placebo vs. control) were not significant (Geers et al., 2005b, 2007, 2010). Considering that the efficacy of placebo has been established in many prior studies, it can be argued that a significant mean difference should constitute a benchmark below which the symptom change in the placebo group cannot be considered a PR.

On a more general note, the majority of studies were performed with healthy undergraduate students, limiting generalizability in terms of age, medical condition, and other parameters. Of the 20 studies reporting placebo predictors, nine were performed by the same research group. Lastly, it is not guaranteed that whatever predicts placebo responsiveness on one occasion will be predictive at another—this applies to questions of reliability, but also to the correspondence of psychological predictors to culturally engrained medical concepts (Walach and Jonas, 2004). Apart from methodological issues, this might be the reason why a number of variables (e.g., neuroticism) which have been found to be involved in placebo responsiveness in the 1950s and 60s have rarely been implicated since (but cf. Darragh et al., 2014).

One of the most intriguing notions of the presented research is the role of the control group in placebo experiments. A number of the constructs identified here warrant a deeper engagement with the nature of this experimental aspect, as has been discussed before, e.g., on the topic of bias correction (Handley et al., 2011). As another example, the induction of optimism can lead to significant analgesia in groups not receiving any treatment at all (Hanssen et al., 2013). The same applies for self-efficacy in motion sickness (Eden and Zuk, 1995; Horing, 2013). It may be that the question of placebo responsiveness is malformed: That it is not about identifying who responds best to placebo, but how to create a paradigm that best reinforces aspects like optimism and self-efficacy, treatment given or not. When members of the control group are being told “you will not get any treatment before the retest…,” it might as well imply, “… so however you deal with the symptoms at retest is entirely up to you.” It seems worthwhile to not only look at what happens in the group receiving the placebo treatment, but to take the time to examine why specific symptoms change in the control group, in order to figure out why placebos are effective at reducing those symptoms at all. Not the least due to ethical constraints, this genuinely psychosomatic question has only rarely been addressed.

To conclude, despite the repeated and widely received emphasis on the necessity of adequate controls (Hróbjartsson, 2002; Kaptchuk et al., 2008; Hróbjartsson et al., 2011), studies are often ill-suited to provide answers concerning predictors of the PR. Control groups are omitted either by design, or by insufficient consideration in statistical analysis. In the latter case, many studies could be revisited to perform post-hoc analyses on response prediction, particularly involving readily available parameters like sex. Keeping in mind the complex interactions of the concrete experimental designs with the respective predictors, a number of variables have been repeatedly involved. It seems worthwhile to include the identified variables in future studies—especially, independent replication of some of the more frequently mentioned parameters is needed. Specifically, promising variables are goal-seeking, self-efficacy, self-esteem, locus of control, and optimism; desire for control and restraint; fun and sensation seeking, and neuroticism; participants' sex (or possibly gender); the val158met-polymorphism; suggestibility, beliefs in expectation biases, body consciousness and baseline symptom severity. At this point, it seems feasible to employ confirmatory analyses involving a priori defined constructs and appropriate error correction.