ARTICLE IN PRESS

Journal of Psychosomatic Research xx (2008) xxx xxx

Consistency of the placebo effect

Ben Whalley a , Michael E. Hyland a,, Irving Kirschb
a

School of Psychology, University of Plymouth, Plymouth, UK b Department of Psychology, University of Hull, Hull, UK

Received 9 May 2007; received in revised form 4 October 2007; accepted 15 November 2007

Abstract Objective: The existence of reliable personality predictors of the placebo effect is controversial. For prediction to be possible, the response to placebo must be reliable. We tested the consistency of the placebo effect by assessing the response to four trials of placebo analgesic treatment. Methods: Two identical experimental pain stimuli were administered simultaneously to matching fingers on both hands. Pain sensation was compared between one finger, which was treated with a placebo cream and the other which was not treated. Two placebo creams were used, each with a different label. The procedure was repeated between 1 and 8 days later using the same creams and order of presentation. Two personality traits
Keywords: Placebo effect; Placebo responders; Pain; Testretest reliability

(acquiescence and absorption) and response expectancy were assessed as potential predictors of the placebo effect. Results: Placebo effects across trials were highly correlated (r=.60 and .77) when placebos bore the same name but were not significantly correlated when placebos had different names. Placebo effects were significantly associated with response expectancy but not with acquiescence or absorption. Conclusions: Context-specific predictions of placebo response (e.g., expectancy) are possible, but personality predictors will not be consistent across contexts. 2008 Published by Elsevier Inc.

Introduction When placebo effects are large, it can be difficult to demonstrate drug effects in clinical trials [1,2], and this has stimulated interest in identifying responders in order that that drugplacebo differences can be more easily detected [3]. One method for doing so is to use placebo run-in or washout periods in which all participants in a trial (including those assigned to verum) are given a placebo, usually for a period of 7 to 10 days. Following this period, participants who have responded to the placebo are excluded from the trial [4]. However, meta-analysis of antidepressant trials comparing those with a run-in period to those without

This study received ethical approval from the Faculty of Science Research Ethics Committee, University of Plymouth. Corresponding author. School of Psychology, University of Plymouth, PL48AA Plymouth, UK. E-mail address: m.hyland@plymouth.ac.uk (M. E. Hyland).

one suggests that this method does not increase drug placebo differences [5,6]. An alternative method is to identify self-report measures of individual differences that predict placebo responding. Although significant correlations between placebo response and the personality trait of acquiescence have been reported [711], the general consensus is that reliable correlates of the placebo response have not been found [12]. Both placebo washout periods and the search for personality correlates of placebo responding implicitly assume that the placebo response is stable over time without stability, reliable predictions would not be possible. Early reports suggested that individual differences in placebo responding are inconsistent [1315], but these studies were underpowered and did not use reliable measures of placebo response. Despite continuing efforts to identify placebo responders, no further studies of the consistency of placebo responses have been reported. In this study we addressed three important and unresolved questions. First, we tested the hypothesis that consistency in

0022-3999/07/$ see front matter 2008 Published by Elsevier Inc. doi:10.1016/j.jpsychores.2007.11.007

ARTICLE IN PRESS
2 B. Whalley et al. / Journal of Psychosomatic Research xx (2008) xxxxxx

placebo responding can be found under optimal conditions. To this end, responses to identical placebo treatments were assessed on two occasions. If consistency cannot be found with all other factors held constant, it is unlikely to be found at all. Second, we tested the hypothesis that an individual's response to placebo is reasonably consistent across contexts. To do this, we measured responses to two placebos that differed only in name. If consistency in the response to placebo can be disrupted by a simple change in name, then the search for a generic placebo responder would be impossible. Third, we tested the extent to which acquiescence, absorption, and response expectancy predicted placebo responding. Acquiescence and absorption are personality traits. Acquiescence has been reported to be associated with placebo responding [711], and absorption has been reported to be a correlate of imaginative suggestibility [16]. If there is stability in placebo responding they seem likely candidates in the search for predictors. In contrast to acquiescence and absorption, response expectancy, which has been identified as a central factor in placebo responding [17,18], is a context-specific variable. Unlike personality traits, which are presumed to be relatively stable across situation and time, context-specific variables vary greatly as a function of the context in which they are assessed. Because they are context-specific, response expectancies might predict placebo responding even where it is inconsistent over time and situation.

gloves. Expectancy ratings were taken after the cream was applied but before the pain stimulus. The study was conducted over two identical experimental sessions, between 1 and 8 days apart (mean 2.5 days, S.D.= 2.0). Each session consisted of two trials. On each trial, participants simultaneously received two pain stimuli, each to a finger of a different hand. One finger was treated with a placebo analgesic cream; the other finger was untreated. For one trial, the placebo was labeled Trivaricaine; for the other, it was labeled Ibuprofen, and order of presentation was balanced across the sample. Different fingers were used for each trial within a session. For half the trials, placebo was administered to a finger on the right hand and for half to a finger on the left hand. After the pain stimulus was applied, pain intensity was recorded. At the end of their final session, participants completed the two personality measures and were thanked and debriefed. Measures and pain stimulus Two identical Forgione-Barber Strain Gauge Pain Stimulators were used. The device consists of a doughnutshaped weight (900 g) at the end of a movable bar (231 g) that pivots vertically from a stand at one end. The subject's finger is placed in a grooved notch on the top of a 50-mm stand so that the bar can be lowered gently onto the finger. The bar tapered at the point of contact and is approximately 2 mm wide. The pain stimulator delivers approximately 2041 g of force to a finger. Two sets of the apparatus were adjusted for the right or left hand, so that the remaining three fingers could rest on the platform between the stand and the bar attachment. Measurements of pain and pain expectancy were taken on 11-point scales, with separate ratings taken for each hand. End points were marked from no pain at all (0) to the worst pain you can imagine (10). For pain ratings, participants were asked How intensely did the lever hurt [with/without] the cream? and for expectancy measurements participants were asked How intensely do you think the lever will hurt [with/without] the cream? For each drug (Trivaricaine and Ibuprofen) and at each time point (Session 1 and Session 2), we calculated the following variables for analysis: placebo scores, subtracting the score for untreated finger from the score for the treated finger, and expectancy scores, subtracting expected pain intensity for the untreated finger from expected intensity for the treated finger. Acquiescence was measured with the 56-item Bass Acquiescence scale [19]; valid responses to each item are yes, no, and uncertain. Bass scale scores are calculated by summing yes responses. Absorption was measured with the 34-item Tellegen absorption scale [20]; valid responses are yes (1) or no (0). Absorption scores in this sample (mean=19.1, S.D.=5.9) were near identical to previous college-based samples [21]. Internal reliability for both scales was acceptable: Tellegen =.80, Bass =.87.

Method Participants and procedure Students and nonfaculty staff at the University of Plymouth were recruited via poster for the study; most of the students received partial course credit in return for their participation, but the remainder of the participants received no compensation. All participants had to be aged over 18 years, and exclusion criteria were any preexisting skin condition, any previous trauma to the fingers to be tested, or any medical condition affecting bones or joints (e.g., osteoporosis, arthritis). A total of 81 participants were recruited, of whom 71 (88%) completed both test sessions and are included in the analysis below (six participants did not give consent to begin the first trial and four failed to return for the second test session). Our sample was 67% female, and the mean age was 20.6 years (S.D.=5.4). During the consent procedure, participants were told that the experiment was to compare the reliability of two analgesic creams: a well-known over-the-counter remedy (Ibuprofen) and a medication used in hospitals for the relief of pain (Trivaricaine). Both creams were placebos but were presented in realistic nonbranded chemists' packaging and consisted of white aqueous cream. To administer the creams, the experimenter donned surgical

ARTICLE IN PRESS
B. Whalley et al. / Journal of Psychosomatic Research xx (2008) xxxxxx Table 1 Means and (S.D.'s) for pain intensity scores (N=71) Placebo name Trivaricaine Ibuprofen Time 1 Untreated 6.56 (1.80) 6.61 (1.75) Treated 5.32 (1.89) 5.69 (1.86) Time 2 Untreated 6.83 (1.78) 6.68 (1.73) Treated 5.46 (1.95) 5.78 (2.03) Expectancy scores Trivaricaine Time 1 Ibuprofen Time 1 Trivaricaine Time 2 Ibuprofen Time 2 P.05. P.01. Table 3 Correlations of placebo and expectancy scores Placebo scores Time 1 Trivaricaine .25 .47 .59 .27 Ibuprofen .20 .18 .22 .61 Time 2 Trivaricaine .14 .36 .47 .68 Ibuprofen .12 .15 .33 .58 3

Data analysis Statistical analyses were performed with SPSS 14. A within-subject analysis of variance (ANOVA) was used to test for the presence of a placebo effect and differences in its magnitude as a function placebo name and experimental session. Because the intertrial interval varied widely, we tested whether the length of interval had effect by entering interval is a covariate in the above ANOVA. Pearson correlations were calculated to assess consistency of individual differences in the placebo effect and the prediction of the placebo effect by personality variables initial response-expectancy ratings. The mediating effect of expectancy ratings was analyzed with multiple regression analyses in which prior placebo effect ratings were controlled statistically. Results Did the creams generate a placebo effect? Mean pain ratings with and without placebo on each trial are presented in Table 1. We carried out repeated-measures ANOVA (drug label, session, and treatment) to evaluate the effects of the creams as placebos. A significant main effect was found for placebo treatment [F(1,70)=43.61, P.001, 2=.38, S.M.D.=0.60], indicating that the placebo creams had a moderate beneficial effect when compared with no cream. No other significant main effects or interactions were found, suggesting that the two labels were equally effective and they were equally effective at Sessions 1 and 2. Repeating this analysis with intersession interval as a covariate
Table 2 Correlations of personality measures with placebo and expectancy scores Tellegen absorption Placebo responses Trivaricaine Time Trivaricaine Time Ibuprofen Time 1 Ibuprofen Time 2 Expectancy scores Trivaricaine Time Trivaricaine Time Ibuprofen Time 1 Ibuprofen Time 2
a a

indicated that the delay between test sessions did not significantly affect the reliability of placebo responses. Were responses temporally consistent? First, we tested whether a person who responded to a placebo at Time 1 was more likely to respond to the same placebo at Time 2. Large correlations between placebo scores for the same placebo at Times 1 and 2 indicate a strong temporal relationship for responses to the same placebo at different times (Trivaricaine r =.60, P.001; Ibuprofen r =.77, P.001). Were responses consistent across contexts? We tested whether a person who responded to one placebo would be more likely to respond to a second, differently labeled placebo. Given the high correlations between responses to the same placebo across sessions, we combined scores for each placebo, creating a mean placebo Trivaricaine score and a mean placebo Ibuprofen score. There was no significant relationship between mean scores for the two placebos: r =.10, P N.41. An analysis using individual scores from each trial (e.g., correlating response to Trivaricaine at Time 1 or Time 2 with response to Ibuprofen at Time 1 or 2) produced similar results. Did our personality measures correlate with the placebo effect? Correlations between the personality measures used and placebo and expectancy responses are shown in Table 2. No correlations were found with placebo responses, although one correlation was found between Time 1 Trivaricaine expectancy and absorption. No further correlations were found with placebo or expectancy scores when collapsing across drug, session, looking at trials in the order of administration, or when correlating the maximum placebo response exhibited in any of the trials. A follow-up analysis, testing for a moderating effect of personality on responses across context, indicated that neither acquiescence nor absorption affected consistency of response across contexts.

Bass acquiescence .05 .02 .02 .04 .13 .10 .07 .01

1 2

.06 .02 .06 .06 .28 .10 .03 .01

1 2

Pb.05. Intercorrelation between the Tellegen and Bass scales was .44.

ARTICLE IN PRESS
4 B. Whalley et al. / Journal of Psychosomatic Research xx (2008) xxxxxx

Did response-expectancy predict the placebo effect? To evaluate initial response expectancies prior to any experience with either placebo cream, we calculated each participant's initial placebo response and initial expectancy response, ignoring the name of the label. Initial placebo response and expectancy response were significantly correlated (r=.33, P.01). Table 3 shows the relationship between placebo response and expectancy for Trivaricaine and Ibuprofen in both sessions. Subsequent expectancies were significantly associated with prior placebo effects, suggesting that people's response expectancies were altered by previous experience. However, expectancies for Trivacaine [t(70)=1.10, ns] and Ibuprofen [t(70)=.07, ns] did not change between Sessions 1 and 2.

Discussion We found that with contextual factors held constant, responses to placebo were highly reliable indicating that, under these circumstances, there is temporal stability in placebo responding. However, participants' responses to two differently labeled placebos were not consistent, indicating that even very small changes in the context of presentation can affect individual differences in the placebo response and confirming Gelfand's early prediction that placebo correlates may vary according to the entire placebo situation [22]. These data suggest that the search for a generic placebo responderone who responds consistently to placebo across different situationscan never be successful, and our failure to find general associations between placebo responses and personality measures is therefore unsurprising. There may not be one placebo response but several. It would seem that there are multiple mechanisms involved and they differ as a function of the context in which the placebo is presented [2326]. For example, spirituality has been found to be a consistent predictor of a placebo when the placebo is contextualized as a spiritual therapy [26], and optimism has been found to predict for a positive but not a negative placebo [27]. Conceptualizing placebos as a meaning response [28] that acts through multiple mechanismsincluding motivation as well as expectancy and conditioningmay aid our understanding of these contextspecific effects. Response expectancies are context-specific, and this makes them more reliable predictors of placebo response than the generic trait measures included in this study. The best measure of the influence of expectancy is the association between initial expectancies and the initial response to placebo. This reveals a modest association, suggesting that, at best, expectancy only partially explains the placebo effect. These data suggest that there may be other context-specific characteristics of the person that determine the response to a placebo. Recent studies of placebos in therapeutic contexts have found just such a pattern, with context-specific traits

and response expectancies independently predicting variance in outcome [26]. Another possibility is that response to placebo is due to implicit, as well as explicit, response expectancies. Work in other fields has alerted researchers to the extent to which nonconscious processing can influence behavior without provoking verbal reports [29,30], and this has also been found in the effect of conditioning on placebo response [18]. It follows that verbal reports of expectancies may not exhaustively measure the extent to which people process and respond to features of a given placebo. Future research should explore this possibility. We found a strong association between experienced placebo effects and subsequent expectancies, suggesting that the experience of placebo effects altered participants' expectations for future responding. This may account for larger observed correlations between expectancies and placebo responses in later trials. A practical implication of these data is that initial response to a placebo, as might be obtained during the runin phase of a clinical trial, is likely to be the best way of identifying placebo responders. Placebo washouts are controversial, and they have been criticized both because they are ineffective and because they may have unexpected consequences on trial samples [31,32], although as Lee et al. [33] have recently noted, they continue to be widely used. Our data suggest that a general rejection of placebo washout periods may be premature. We note that much of the data on wash-out periods has come from trials of selective serotonin reuptake inhibitors (SSRIs) antidepressants for which a large component of the active treatment is in fact placebomediated [34], and under these conditions, washing out placebo responders would reduce effect sizes in both arms of the trial. Ironically, placebo washout periods may be effective only in trials of medications that do not generate large placebo effects. Expectancy predicts placebo responding to some degree and, although it is correlated with prior experience, it is partially independent. However, some caveats are in order. First, unlike the present study, placebo-controlled clinical trials contain double-blind instructions and participants are aware that they might receive placebo. In contrast, the instructions in our study were similar to those in comparative drug trials without placebo controls. Second, our population was primarily female and we cannot be sure that the results generalize to males as well. Third, we used an experimental pain stimulus. Thus, we cannot be certain that our results will generalize to clinical contexts. Future studies of consistency of response in clinical patients might employ a variant of the open-hidden paradigm [35], avoiding the ethical problems of withholding analgesics from patients in pain. Fourth, data on placebo analgesics might not generalize to other types of placebos. For these reasons, tests of consistency of placebo responding should also be conducted with other types of placebos known to be effective, for example anxiolytics or treatments for Parkinson's disease [35]. Additionally, the

ARTICLE IN PRESS
B. Whalley et al. / Journal of Psychosomatic Research xx (2008) xxxxxx 5

experiment was not double blind because the experimenter had access to previous trial data for each participant. To minimize the risk of unintentional bias, scores for each trial were recorded separately, and Trial 1 scores were not visible to experimenter or participant during the second trial. However, we cannot rule out the possibility of experimenter effects. We conclude that the search for predictors of placebo responses in particular contexts is realistic because context-specific placebo responding is as reliable as many trait measures. However, the search for a generic placebo-prone personality is unlikely to be successful. Our data suggest that focusing on context-specific predictions of responses to particular types of placebos would be a more productive enterprise. References
[1] Schatzberg AF, Kraemer HC. Use of placebo control groups in evaluating efficacy of treatment of unipolar major depression. Biol Psychiatry 2000;47:73644. [2] Walsh BT, Seidman SN, Sysko R, Gould M. Placebo response in studies of major depression: variable, substantial, and growing. JAMA 2002;287:18407. [3] Charney DS, Nemeroff CB, Lewis L, Laden SK, Gorman JM, Laska EM, Borenstein M, Bowden CL, Caplan A, Emslie GJ, Evans DL, Geller B, Grabowski LE, Herson J, Kalin NH, Keck PE, Kirsch I, Krishnan KRR, Kupfer DJ, Makuch RW, Miller FG, Pardes H, Post R, Reynolds MM, Roberts L, Rosenbaum JF, Rosenstein DL, Rubinow DR, Rush AJ, Ryan ND, Sachs GS, Schatzberg AF, Solomon S. National Depressive and Manic-Depressive Association consensus statement on the use of placebo in clinical trials of mood disorders. Arch Gen Psychiatry 2002;59:26270. [4] Quitkin F, McGrath P, Stewart J, Ocepek-Welikson K, Taylor B, Nunes E, Delivannides D, Agosti V, Donovan S, Ross D, Petkova E, Klein D. Placebo run-in period in studies of depressive disorders. Clinical, heuristic and research implications. Br J Psychiatry 1998:173. [5] Posternak MA, Zimmerman M, Keitner GI, Miller IW. A reevaluation of the exclusion criteria used in antidepressant efficacy trials. Am J Psychiatry 2002;159:191200. [6] Trivedi MH, Rush J. Does a placebo run-in or a placebo treatment cell affect the efficacy of antidepressant medications? Neuropsychopharmacology 1994;11:3343. [7] Fisher S, Greenberg RP. The curse of the placebo: fanciful pursuit of a pure biological therapy. In: Fisher S, Greenberg RP, editors. From placebo to panacea: putting psychiatric drugs to the test. New York: John Wiley, 1997. pp. 356. [8] McNair D, Gardos G, Haskell D, Fisher S. Placebo response, placebo effect, and two attributes. Psychopharmacology (Berl) 1979;63: 24550. [9] McNair DM, Barrett JE. Two Bass Scale factors and response to placebo and anxiolytic drugs. Psychopharmacologia 1979;65:16570. [10] McNair DM, Fisher S, Sussman C, Droppleman LF, Kahn RJ. Persistence of a drugpersonality interaction in psychiatric outpatients. J Psychiatr Res 1970;7:299305. [11] McNair DM, Khan RJ, Droppleman LF, et al. Compatibility, acquiescence and drug effects. Excerpta Medica Internat Congress 1966;129:536.

[12] Brody H. The placebo response. New York: HarperCollins, 2000. [13] Wolf S, Doering CR, Clark ML, Hagans JA. Chance distribution and the placebo reactor. J Lab Clin Med 1957;49:83741. [14] Lasagna L, Mosteller F, Von Felsinger J. A study of the placebo response. Am J Med 1954;16:7709. [15] Shapiro AK. The reliability and validity of a placebo test. J Psychiatr Res 1979;15:25390. [16] Kirsch I, Braffman W. Imaginative suggestibility and hypnotizability. Curr Dir Psychol Sci 2001;10:5761. [17] Kirsch I. Response expectancy as a determinant of experience and behavior. Am Psychol 1985;40:1189202. [18] Stewart-Williams S, Podd J. The placebo effect: dissolving the expectancy versus conditioning debate. Psychol Bull 2004;130: 32440. [19] Bass BM. Development and evaluation of a scale for measuring social acquiescence. J Abnorm Psychol 1956;53:2969. [20] Tellegen A, Atkinson G. Openness to absorbing and self-altering experiences (absorption), a trait related to hypnotic susceptibility. J Abnorm Psychol 1974;83:26877. [21] Glisky ML, Tataryn DJ, Tobias BA, Kihlstrom JF, McConkey KM. Absorption, openness to experience, and hypnotizability. J Pers Soc Psychol 1991;60:26372. [22] Gelfand S, Ullmann L, Krasner L. The placebo response: an experimental approach. J Nerv Ment Dis 1963;163:37987. [23] Hyland ME, Geraghty AWA, Joy OET, Turner SI. Spirituality predicts outcome independently of expectancy following flower essence selftreatment. J Psychosom Res 2006;60:538. [24] Geers AL, Helfer SG, Kosbab K, Weiland PE, Landry SJ. Reconsidering the role of personality in placebo effects: dispositional optimism, situational expectations, and the placebo response. J Psychosom Res 2005;58:1217. [25] Geers AL, Weiland PE, Kosbab K, Landry SJ, Helfer SG. Goal activation, expectations, and the placebo effect. J Pers Soc Psychol 2005;89:14359. [26] Hyland ME, Whalley B, Geraghty AWA. Dispositional predictors of placebo responding: a motivational interpretation of flower essence and gratitude therapy. J Psychosom Res 2007;62:33140. [27] Geers AL, Handley IM, McLarney AR. Discerning the role of optimism in persuasion: the valence-enhancement hypothesis. J Pers Soc Psychol 2003;85:55465. [28] Moerman DE. Meaning, medicine, and the placebo effect. Cambridge: Cambridge University Press, 2002. [29] Bargh JA, Chartrand TL. The unbearable automaticity of being. Am Psychol 1999;54:46279. [30] Cleeremans A, Destrebecqz A, Boyer M. Implicit learning: news from the front. Trends Cogn Sci 1998;2:40616. [31] Reimherr FW, Ward MF, Byerley WF. The introductory placebo washout: a retrospective evaluation. Psychiatry Res 1989;30:1919. [32] Greenberg R, Fisher S, Riter J. Placebo washout is not a meaningful part of antidepressant drug trials. Percept Mot Skills 1995;81: 68890. [33] Lee S, Walker JR, Jakul L, Sexton K. Does elimination of placebo responders in a placebo run-in increase the treatment effect in randomized clinical trials? A meta-analytic evaluation. Depress Anxiety 2004;19:109. [34] Moncrieff J, Kirsch I. Efficacy of antidepressants in adults. BMJ 2005;331:1557. [35] Colloca L, Lopiano L, Lanotte M, Benedetti F. Overt versus covert treatment for pain, anxiety, and Parkinson's disease. Lancet Neurol 2004;3:67984.