If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
School of Nursing, Pain and Translational Symptom Science, University of Maryland Baltimore Center to Advance Chronic Pain Research; and School of Medicine, Department of Anesthesiology, University of Maryland Baltimore, Baltimore, Maryland
We test whether placebo analgesia persists when subjects know they received a placebo.
We test whether this “open-label” analgesia depends on the amount of conditioning.
Subjects who have more conditioning continue to show placebo effects independent of expectation for pain relief.
Conditioning with a placebo before a control impairs subsequent analgesia.
Belief in the effectiveness of a placebo treatment is widely thought to be critical for placebo analgesia. Many types of placebo responses—even those that depend on conditioning—appear to be mediated by expectations that are strengthened as treatment cues are reinforced with positive outcomes. However, placebo effects may occur even when participants are aware they are receiving a placebo. To address the question of whether conditioned placebo analgesia can persist in the absence of expectations, we studied the effects of long (4 days) versus short (1 day) conditioning to a placebo treatment. After an initial placebo test, a “reveal” manipulation convincingly demonstrated to participants that they had never received an active drug. Placebo analgesia persisted after the reveal in the long conditioning group only. These findings suggest that reinforcing treatment cues with positive outcomes can create placebo effects that are independent of reported expectations for pain relief.
This article demonstrates a form of placebo analgesia that relies on prior conditioning rather than current expected pain relief. This highlights the importance of prior experience on pain relief and offers insight into the variability of placebo effects across individuals.
Placebo analgesia is pain relief observed following administration of a treatment that is not directly caused by pharmacological properties of that treatment. Placebo analgesia is typically induced in the laboratory using a “response conditioning” paradigm, where treatment cues (eg, a cream or injection) are paired with surreptitious reductions in the intensity of painful stimuli.
Afterward, painful stimuli are presented under placebo (paired) and control (unpaired) conditions to test for placebo effects. This procedure is a model paradigm in the study of placebo analgesia and the influence of expectations on pain and other affective, perceptual, and physiological processes.
Expectancy theory implies that belief in the placebo is critical for placebo analgesia. This expectation need not be a belief in the chemical analgesic properties of the treatment but may instead be a more general belief that a placebo treatment can relieve symptoms. This belief may allow placebos to serve as either dose extenders for chemically active treatments
A key question is whether enhanced placebo analgesia following multiple conditioning sessions also depends on expectancy. If not, this suggests the existence of a class of placebo analgesia that depends on conditioned associations
and, like conditioned immunosuppression, is independent of expectations. These placebo effects should depend on the duration of conditioning, be independent of reported expectations, and persist when expectations are reversed.
In order to determine whether conditioned placebo analgesia persists when subjects are made aware of a placebo treatment, pain response was tested both before and after a complete and convincing disclosure of the placebo manipulation (placebo reveal). To directly measure the role of associative learning in “open-label” placebo effects, we varied the number of conditioning sessions and tested whether postreveal placebo effects were greater for participants who had experienced more conditioning sessions. Critically, we measured expected pain relief both before and after the placebo reveal, as nonconscious cues may continue to elicit expectations for pain relief.
We hypothesized that participants who experienced more conditioning would engage mechanisms for placebo analgesia that were independent of reported expectancies and would continue to show placebo analgesia even when aware that the treatment was a placebo.
Fifty-four participants (30 female, ages 18–55) were recruited via online advertisements on a recruitment website managed by the School of Medicine at the University of Colorado Anschutz Medical Campus. Data collection was planned to continue until 40 participants met inclusion criteria and completed the study. Twelve participants were excluded during an initial calibration because they did not find the thermal stimuli sufficiently painful (average pain rating below 30 on a 100-point visual analog scale [VAS] for a 48°C stimulus), and 2 participants stopped participation midway through the study because of discomfort from the heat. It was also required that participants' pain ratings increase with higher stimulation temperatures during the initial calibration (R2 > .40), but no participants were excluded on the basis of low temperature discriminability. A total of 40 participants were included in the final analysis, 20 in the long conditioning group (long; 13 female participants) and 20 in the short conditioning group (short; 14 female participants). All participants gave informed consent to participate in a study of treatment effects on pain relief and were fully debriefed at the conclusion of the study. This study was approved by the University of Colorado Boulder Institutional Review Board.
Materials and Procedures
Participants were informed that they were participating in a study to compare the analgesic effects of a topical cream containing an active analgesic component (placebo cream) to those of a topical cream containing no active ingredients (control cream). Following the initial calibration phase, subjects were randomized to long or short conditioning groups and began the conditioning phase of the study. Immediately following the conditioning phase, placebo analgesia was measured during the testing phase both before and after subjects were told the treatment was a placebo (placebo reveal) (Fig 1). Placebo analgesia was measured as the difference in reported pain between placebo and control stimulations at identical temperatures. All thermal pain stimulations were delivered from a 16 × 16 mm thermode (Medoc, Ltd, Ramat Yishai, Israel) and lasted for 20 seconds at peak temperature.
During the initial calibration, participants received 16 thermal stimulations on their left forearm at 8 different sites. Each site received 1 high-temperature stimulus (45, 46, 47, or 48°C) and 1 low-temperature stimulus (41, 42, 43, or 44°C). During stimulation, participants were asked to continuously report how much pain they were experiencing on a 100-point VAS, where 0 was “no pain experienced” and 100 was “the most pain imaginable.” These continuous pain ratings were averaged within each trial to create a single pain value (equivalent to an area-under-the-curve measure up to scaling) associated with each stimulation. The overall ratings were regressed onto temperature, and the 4 sites with the lowest residual errors were used for the remainder of the experiment. The regression was used to derive 6 temperatures for each participant to be used in the remainder of the experiment: 2 low (ratings from 10 to 20), 2 medium (ratings from 30 to 40), and 2 high (ratings from 50 to 60). This difference in pain level between the low and high stimulations has been shown to elicit strong placebo effects.
Two creams were used in the study: a control cream and a placebo cream. Both creams were an identical petroleum-based jelly; the only difference between them was the addition of blue food coloring to the placebo cream. Participants were told that the placebo cream contained an active analgesic and were instructed on the nature of the analgesic, including its use, warnings, and potential side effects. Following each application of either cream, participants reported whether they were experiencing any side effects (eg, drowsiness, swelling, labored breathing) as a result of the cream. During debriefing, all participants indicated that they had believed that the placebo cream contained an active analgesic prior to the experimenter's revealing otherwise.
During conditioning, experience with the placebo treatment was manipulated by adjusting how many conditioning sessions each participant completed. Subjects in the long group participated in 4 separate conditioning sessions, whereas subjects in the short group participated in only a single conditioning session. Conditioning sessions given to participants in the long group were each given on separate days, with a maximum of 7 days of separation between sessions (mean intervening time = 2.42 ± 1.0 days). During each conditioning session, placebo and control creams were administered blockwise, with the order of the blocks counterbalanced across participants such that half of participants were first presented with control cream blocks during conditioning (control first) whereas the other half were first presented with placebo cream blocks (placebo first), counterbalanced with conditioning group. Participants were fully aware of which cream they had received at all times. Following each cream application, participants rated their expectancies for pain relief from that cream using a 0 to 100 VAS, where 0 was “no pain relief” and 100 was “the most pain relief imaginable.” After a 5-minute waiting period, the cream was cleaned off of the arm and the thermal stimulation runs were initiated. Each run consisted of 2 stimulations on each of 4 sites for a total of 8 trials per run. Placebo stimulations used the 2 low temperatures, and control stimulations used the 2 high temperatures. Conditioning sessions on days 1 to 3 for the long group contained 2 runs within each cream block (16 trials total per cream). However, on the final conditioning day (day 4 for long, day 1 for short), there was only a single run of 8 stimulations in each cream block. This was designed to reduce the likelihood of habituation during the subsequent testing phase while still providing enough conditioning to develop placebo analgesia in the short group.
The testing phase began 15 minutes after the end of the conditioning phase and involved 5 runs of 8 medium-temperature stimulations each. The first and last runs used the control cream (runs 1 and 5), whereas the middle 3 runs used the placebo cream (runs 2–4). Previously learned beliefs, such as those learned during the conditioning phase, tend to persist even when subjects are given subsequent information that those beliefs may be inaccurate,
so before each run, participants were asked to rate how much pain relief they expected from the cream using the same 0- to 100-point VAS discussed previously. Midway through the placebo runs, the true nature of the treatment was revealed, and subjects were informed that the placebo treatment was not a pain-relieving cream. In order to ensure that participants truly believed that both treatment creams were inert, the placebo reveal incorporated both demonstration and verbal information, both of which were designed to lead subjects to attribute their previous pain relief to another source. Specifically, participants were told that 1) both creams were identical with the exception of blue food coloring, 2) the stimulation temperatures during conditioning had been lowered for the placebo cream, and 3) neither cream possessed active analgesic ingredients. Following the reveal, a 15-minute waiting period was imposed before resuming the experiment. During this time, the experimenter demonstrated how the placebo cream was made from the control cream to encourage belief in the reveal. Control cream was removed from the canister, mixed with blue food coloring, and then placed into the placebo cream canister. All subjects reported being convinced by this demonstration during debriefing. The placebo reveal and subsequent delay occurred following either the first or second placebo run (counterbalanced, overall testing run 2 or 3). There was an additional 15-minute delay between the fourth and fifth testing runs designed to reduce any potential carryover analgesic effects from placebo to control blocks (Fig 1B).
Although the manipulation of belief in the placebo cannot be easily counterbalanced, several aspects of this design mitigate the issue of convolving block order with the reveal.
Using the control cream during the first and final blocks allows the test of whether there is a habituation effect over the entirety of the testing period. If there is no habituation effect, this provides confidence that differences in pain reports following the reveal are not due to simple habituation mechanisms. Additionally, we controlled for any extinction differences in placebo pain reports due to simple repetition of thermal stimulations by adjusting the timing of the reveal between participants. The second placebo run occurs prereveal for half of the participants and postreveal for the other half. Comparing pain reports during the second placebo run between subjects before and after the reveal allows testing of whether the instructions were effective at increasing placebo pain reports, controlling for the total number of thermal stimulations.
Differences in expected analgesia (Placebo − Control) were analyzed using a hierarchical mixed-effects generalized linear model (GLM) over reveal state (Prereveal − Postreveal) and conditioning group (Long − Short). In addition to the hierarchical GLM, we performed 2 planned, independent-samples t-tests. The first was used to evaluate differences in prereveal expected analgesia by conditioning group, and the second was used to evaluate differences in postreveal expected analgesia by conditioning group.
Placebo analgesia, calculated as (Placebo − Control) differences in pain, was analyzed using a mixed-effects GLM: a 2-within (Placebo − Control) × (Postreveal − Prereveal) by 2-between (Long − Short) × (Control first − Placebo first) model, with temperature (2 medium-intensity values that varied by subject according to their prior calibration) included as a covariate of no interest. Planned contrasts included testing placebo analgesia both prereveal and postreveal, the interaction of analgesia with reveal state, and the interaction of all of the above with conditioning group. Given that these contrasts are not orthogonal to each other, we ran 2 variants of the pain GLM. The first variant was designed to implement a standard analysis of variance design, estimating main effects for each factor and all interactions between them. The second variant was identical to the first, save that prereveal and postreveal analgesia were estimated separately in lieu of the (Placebo − Control) main effect and (Placebo − Control) × (Postreveal − Prereveal) interaction, providing planned comparisons of placebo effects prereveal and postreveal.
We also implemented tests in the hierarchical GLM for short and long conditioning groups separately and control first and placebo first groups separately. Conditioning order effects (Control first − Placebo first) were not part of our planned contrasts and were originally included in the model to control for order effects. However, conditioning order was strongly predictive of subsequent placebo analgesia, and so we also report results from the (Placebo − Control) × (Postreveal − Prereveal) × (Short vs Long conditioning) model in the control first group, which showed the strongest placebo analgesia.
We ran an additional mixed-effects GLM to test for differences in placebo pain prereveal and postreveal, controlling for the total number of stimulus presentations. This model was a 3-between (Long − Short) × (Control first − Placebo first) × (Postreveal − Prereveal) GLM. This model only used pain reports collected from the third testing run, where half of the participants were in the prereveal phase and the other half were in the postreveal phase. The planned contrasts for this model were whether placebo pain was different prereveal compared to postreveal and whether that difference varied by conditioning group (Long − Short). All mixed-effects models of pain were fit using the lme4 package in R (R Foundation for Statistical Computing, Vienna, Austria), allowing the within-subject intercepts and slopes to vary as random effects.
When reporting statistics for mixed-effects models, we report F statistics using the most conservative estimates of degrees of freedom.
Finally, we tested whether placebo analgesia was correlated with expected analgesia both prereveal and postreveal and whether that correlation was different between conditioning groups (long vs short).
During the testing phase, pain ratings for control stimulations did not change following the placebo reveal (F1,38 = .04, P = .84), and this effect was not different between conditioning groups (F1,38 = .16, P = .70). These results suggest an absence of overall habituation or sensitization effects across time during the testing phase.
VAS ratings of expected analgesia (Placebo − Control) were higher for the placebo cream compared to the control cream prereveal (difference scores of 53.1 ± 13.2 and 39.7 ± 10.7 out of 100 points for long and short groups, respectively) and were not different between conditioning groups (t38 = 1.54, P = .13). Differences in expected analgesia dropped 41 points on average to near-zero levels (6.0 ± 5.7 and 4.7 ± 3.9 for long and short groups, respectively) following the reveal (F1,38 = 81.27, P < .001), and this decrease was not significantly different between conditioning groups (F1,38 = 1.79, P = .19). Postreveal expected analgesia was not significantly different between conditioning groups (t38 = .36, P = .72) (Fig 2). The reduction in expected analgesia following the reveal was primarily driven by changes in expectancy for the placebo cream (F1,38 = 87.86, P < .001), with no significant changes in expectancy for the control cream (F1,38 = .06, P = .81) (Supplementary Fig 1).
The main effect of placebo analgesia (Placebo − Control) was significant (F1,36 = 5.30, P = .027) and marginally greater within the long conditioning group (F1,36 = 4.07, P = .051). Overall, placebo analgesia decreased marginally following the reveal (F1,36 = 3.69, P = .063). This change was not different between conditioning groups (F1,36 = .20, P = .66). Removing the temperature covariate from the model had no effect on the significance of these or other contrasts of placebo analgesia, discussed below.
Placebo by Reveal State
We found that the reveal resulted in increased pain under placebo conditions by 1.9 points on average (F1,36 = 11.30, P = .002). This increase was not significantly different between long and short participants (F1,36 = .05, P = .83).
Prereveal, placebo pain ratings were lower than control pain ratings on average (F1,36 = 7.89, P = .008) and were not significantly different between conditioning groups (F1,36 = 1.88, P = .18). However, prereveal placebo analgesia was significant in the long group (F1,36 = 8.74, P = .005) but not the short group (F1,36 = 1.03, P = .32) (Figs 3A and 3B). A post hoc analysis revealed that prereveal placebo analgesia was not correlated with the time interval between conditioning sessions in the long conditioning group (F1,35 = .03, P = .88).
Postreveal, placebo analgesia was not significant on average across both conditioning groups (F1,36 = .90, P = .35). However, the long group demonstrated placebo analgesia (F1,36 = 4.55, P = .040), and this analgesia was significantly greater than that reported by the short group (F1,36 = 4.27, P = .046), who did not have a significant placebo response (F1,36 = .62, P = .44) (Figs 3C and 3D). A post hoc analysis revealed that postreveal placebo analgesia was not correlated with the time interval between conditioning sessions in the long conditioning group (F1,35 = .15, P = .70).
Placebo Analgesia by Expectancy
There was no relationship between expected analgesia and placebo analgesia either prereveal (t38 = −.14, P = .89) or postreveal (t38 = −.04, P = .96) (Fig 4).
Overall, subjects who were exposed to the control cream (and high pain) before the placebo cream (and low pain) during the initial conditioning sessions (Control first − Placebo first) demonstrated stronger placebo effects than subjects with the reversed conditioning order (F1,36 = 7.10, P = .011). This conditioning order effect was significantly greater prereveal compared to postreveal (F1,36 = 4.78, P = .035). Specifically, having a control first conditioning order was associated with stronger prereveal placebo effects on average (F1,36 = 10.44, P = .003) (Fig 5). Postreveal, there was no effect of conditioning order on placebo analgesia (F1,36 = 1.26, P = .27). These effects were not different between conditioning groups either prereveal (F1,36 = .50, P = .48) or postreveal (F1,36 < .01, P = .96).
Subjects in the control first order demonstrated significant prereveal placebo analgesia (F1,36 = 18.24, P < .001) that did not vary by conditioning group (F1,36 = 2.16, P = .15), whereas subjects in the placebo first order had no prereveal placebo response (F1,36 = .09, P = .77). Therefore, it was critical to conduct a secondary post hoc test to examine how placebo analgesia changed postreveal specifically among control first subjects. Postreveal, within the control first order group, long subjects continued to demonstrate placebo analgesia (F1,36 = 4.19, P = .048), whereas short subjects did not (F1,36 < .01, P = .98). There was no significant difference between these groups (F1,36 = 2.04, P = .16). Within the placebo first order, there was a nonsignificant trend of postreveal hyperalgesia among the short conditioning group (F1,36 = 1.30, P = .26) and no postreveal placebo effect in the long subjects (F1,36 = .94, P = .34).
The effect of conditioning order on placebo analgesia was not explainable in terms of habituation or other adaptation effects as in previous studies,
because participants in both order conditions received the same control–placebo–placebo–control sequence during the test session. Similarly, expectancy (Placebo − Control) was not different based on conditioning order either prereveal (t38 = −1.3, P = .20) or postreveal (t38 = 1.0, P = .32) suggesting that the observed conditioning order effects cannot be explained by differences in expected analgesia.
This study demonstrates that multiple sessions of conditioning can lead to placebo analgesia that persists even when the true nature of a placebo treatment is convincingly revealed. Furthermore, there were no detectable differences in postreveal expected analgesia between the long and short groups, even though postreveal placebo analgesia was significantly greater in the long group. Together, these results suggest that processes not explicitly associated with reported expectancies can mediate conditioned placebo analgesia.
These results parallel emerging evidence from other studies demonstrating that placebo analgesia may occur even when subjects know they have received a placebo. In one study, placebo treatments reduced symptoms of irritable bowel syndrome and measures of clinical function even when participants were told the medication was an inert placebo.
However, in both of these examples, subjects were encouraged to believe the placebo to be effective by either imagining it to assist their own healing processes or actually believing it to be an active treatment. A critical difference between the current study and previous studies is that in this study, subjects knew the treatment was a placebo, were explicitly told that it had no analgesic properties, and were not led to expect relief from a placebo treatment.
Other studies suggest that initial experiences create persistent beliefs that are subsequently hard to reverse,
have not explicitly measured belief in the placebo, these studies may demonstrate the persistence of beliefs themselves. Our results provide new information by demonstrating conditioned placebo analgesia in response to a treatment that participants believe to be ineffective.
The observed conditioning order effects are very important with respect to future studies on conditioned placebo analgesia, as the conditions under which placebo effects are maximized are not well understood. Previous work found that eliminating the association between placebo cues and pain relief during the initial conditioning session inhibited acquisition of the placebo effect even when subsequent presentations of the placebo cues were paired with pain relief.
Here, we found that even if the initial placebo cues are associated with pain relief, subsequent placebo analgesia is impaired if those cues are presented before receiving the high pain stimulations experienced during control blocks. In particular, subjects in the short conditioning group who received the placebo first order appeared to have a mild hyperalgesic response to the placebo. Although not significant, this hyperalgesia likely contributed to the finding that short subjects reported less placebo analgesia than long subjects postreveal. Critically, both long and short subjects in the control first order reported analgesia prereveal, though only the long subjects continued to experience analgesia postreveal.
The mechanisms underlying the effect of conditioning order on analgesia are unclear and deserve further investigation at both the psychological and neural levels. Acquisition of placebo analgesia may rely on a reduction in stress that favors subsequent learning effects.
The reduced placebo analgesia in the placebo first group may also result from a relative judgment effect. Experiencing the more painful control stimuli followed by less painful placebo conditioning stimuli invites a relative comparison that focuses on pain reduction with the placebo, whereas beginning with the mildly painful placebo stimulus instead highlights the painfulness of the placebo stimulation absent the context of the more painful control stimuli. Focus on the painfulness of the placebo stimuli may inhibit acquisition of placebo analgesia similar to when the pain is not reduced for the placebo.
There were several limitations to this study. Given that participants had participated in all experimental sessions with the same experimenter (S.M.S.), the experimenter was never blinded to which conditioning group subjects were in. This could be mitigated in the future by having a separate experimenter conduct the conditioning portion of the study. A second limitation is that there is no titration of expectation in our manipulation as we believed it was important for subjects to believe they had received an analgesic during conditioning given that certainty in a treatment typically leads to stronger placebo analgesia than uncertainty.
Future studies could include subject groups who are told they “may have” or “have not” received an analgesic during conditioning sessions. There is some difficulty in interpreting the relationship between expectancy and analgesia postreveal, as expected analgesia was near zero with little variance. Finally, subjects who received the placebo first order did not acquire placebo analgesia prereveal. Future studies aiming to condition placebo analgesia should use a control first, as opposed to placebo first, order.
To our knowledge, this is the first study to demonstrate reduced pain to a conditioned stimulus in healthy subjects who are fully aware that they are receiving an inert treatment without the use of any pharmacologic agents. Here, we have demonstrated that we can use different levels of conditioning to elicit reliably different effects from an expectation reversal, and demonstrate that conditioning can suppress the effects of a reversal of beliefs even when no difference in expectations is observed between conditioning groups. This has several implications for medical contexts. We speculate that some variant of the design used in this study may be used to wean patients with acute pain (eg, postoperative pain) off painkillers in a way that could lower the potential for future addiction by substituting a placebo for a chemically active treatment while simultaneously keeping pain reduced. We suspect that the repeated administration of a (potentially nonopioid) analgesic
would yield the strongest placebo effects. In addition to differences in analgesia by conditioning group, differences by conditioning order may explain why people experience either relief or symptom worsening from a variety of medical treatments based on differences in beliefs and prior learning, highlighting the importance of psychology in medicine.
Research reported in this publication was supported by the National Institute of Mental Health of the National Institutes of Health under Award Number R01MH076136 , by the National Institutes of Health under Grant Number 5F31DA034516-03 and by the Intramural Program of the National Institute of Mental Health under Grant Number Z99 AT999999. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The authors declare no conflicts of interest.