An analysis of factors that contribute to the magnitude of placebo analgesia in an experimental paradigm

Abstract

Placebo analgesia was produced by conditioning trials wherein heat induced experimental pain was surreptitiously reduced in order to test psychological factors of expectancy and desire for pain reduction as possible mediators of placebo analgesia. The magnitudes of placebo effects were assessed after these conditioning trials and during trials wherein stimulus intensities were reestablished to original baseline levels. In addition, analyses were made of the influence of these psychological factors on concurrently assessed pain and remembered pain intensities. Statistically reliable placebo effects on sensory and affective measures of pain were graded according to the extent of surreptitious lowering of stimulus strength during the manipulation trials, consistent with conditioning. However, all of these effects were strongly associated with expectancy but not desire for relief. These results show that although conditioning may be sufficient for placebo analgesia, it is likely to be mediated by expectancy. The results further demonstrated that placebo effects based on remembered pain were 3 to 4 times greater than those based on concurrently assessed placebo effects, primarily because baseline pain was remembered as being much more intense than it actually was. However, similar to concurrent placebo effects, remembered placebo effects were strongly associated with expected pain levels that occurred just after conditioning. Taken together, these results suggest that magnitudes of placebo effect are dependent on multiple factors, including conditioning, expectancy, and whether analgesia is assessed concurrently or retrospectively.

1. Introduction

Nearly all studies that have sought to explain the psychological basis of the placebo effect have focused on either a classical conditioning model or a single psychological factor such as anxiety, expectancy, hope, or faith in the treatment. Recently, the possibility that multiple or alternative psychological factors mediate placebo effects has been recognized (^{Price and Barrell, 1984; Bootzin, 1985; Evans, 1985; White et al., 1985; Montgomery and Kirsch, 1996a,b; Fields and Price, 1997; Price and Fields, 1997}). For example, ^{Montgomery and Kirsch (1996)} introduced an experimental manipulation that dissociated expectancy from conditioning to determine which mechanism best accounted for variance in placebo response. In this paradigm, a placebo cream is applied to the skin under the guise of a local analgesic and the painful stimulus strength is surreptitiously lowered only during conditioning trials. The placebo response is obtained during post-conditioning trials wherein the painful stimulus strength is applied at the same baseline levels prior to conditioning.

Montgomery and Kirsch modified this paradigm by informing one group of subjects about the lowering of painful stimulus intensity and not informing the other group. The uninformed group was thereby provided with an experience of placebo-induced analgesia during conditioning and, as expected, demonstrated placebo analgesia in the subsequent test trials. In contrast, expectations of analgesia were lowered in the informed group and no overall placebo analgesic effect occurred in this group. Furthermore, although conditioning trials significantly enhanced placebo responding, this effect was eliminated by adding participant's expectancy ratings to the regression equation, further indicating that the effect of pairing trials on placebo response was mediated by expectancy (^{Baron and Kenny, 1986}). This study demonstrated that conditioning is not sufficient for placebo effects and it provided evidence that expectation of a therapeutic effect is necessary for placebo effects. The results of this study constituted multiple lines of evidence against the Pavlovian conditioning model and supported the mediating role of expectancy.

These findings are consistent with response expectancy theory (^{Kirsch, 1990}). Response expectancy theory hypothesizes that: (1) expectations for non-volitional outcomes are sufficient to cause the expected outcome, (2) response expectancy effects are not mediated by other psychological variables, and (3) effects of response expectancies are self-confirming and seemingly automatic. For example, an expectation that one will be depressed for a long period of time is quite depressing and an expectation of being anxious before an exam can evoke anxiety. In the same way, expectations for pain reduction following placebo administration are hypothesized to directly cause subsequent pain relief.

Although expectancy appears to have a critical mediating role in producing placebo analgesia, it is unlikely to operate alone. Placebo analgesia often occurs under conditions wherein subjects want a therapeutic agent to provide pain relief. That is, an individual's desire for reduction in pain may also directly contribute to subsequent pain relief by providing a need to experience a treatment as effective. Similar to response expectation theory, the need to experience an effective analgesic treatment could have a direct effect on pain. The possible mediating role of motivation or desire for relief could account for several observations. It could indirectly account for the increase in placebo response with the severity of clinical pain (^{Beecher, 1955, 1959; Jospe, 1978; Levine et al., 1978}). It might also account for the greater placebo response for clinical pain versus experimental pain and for the observation that placebo responses in experimental pain increase as a function of its duration and severity (^{Jospe, 1978}). In other words, greater and more open-ended threats may lead to greater needs for relief and placebo responses.

Unfortunately, the role of desire for a specific symptom change has yet to be tested for placebo analgesia and it has only rarely been tested for placebo effects in general. In one exceptional study of placebo manipulations suggesting possible sedative or stimulant effects, ^{Jensen and Karoly (1991)} assessed the separate contributions of motivation and expectancy to placebo effects. According to the authors, motivation was ‘the degree to which subjects desire to experience a symptom change,’ and expectancy was considered the subjects’ expectation of a symptom change. They found that motivation but not expectancy contributed a significant amount of variance to placebo responses that included perceived sedation or perceived stimulant effects. However, limitations of their study were that desire and expectancy manipulations and not direct measures of these factors constituted the independent variables of the study and the authors questioned the effectiveness of their expectancy manipulation.

One purpose of this study was to evaluate two factors for their possible contribution to placebo analgesia: (1) a desire for treatment to significantly relieve pain and (2) the level of expectation that pain will be significantly relieved. Based on the literature reviewed above, one would anticipate direct effects of both response expectancy and desire for pain reduction on the magnitude of placebo analgesia. Thus, we sought to determine the extent to which either of these two factors predicts placebo-induced reductions in ratings of sensory and unpleasantness dimensions of experimental pain. The evaluation of expectancy also represents an extension and replication of the study by ^{Montgomery and Kirsch (1996)}.

A second major purpose of the present study was to compare the magnitudes of placebo effects based on concurrent and retrospective ratings of pain and to determine whether the same factors of desire and expectation also influence placebo effects based on remembered pain. Concurrent placebo effects were based on ratings of pain that had just been evoked from heat stimuli applied within baseline control and placebo-treated areas of skin and were assessed during placebo treatments. ‘Remembered placebo effects’ were based on retrospective ratings of these same evoked pains approximately 2 min afterwards. The rationale for the evaluation of ‘remembered placebo effects’ is based on results of previous studies that show that memory distortion of pretreatment pain contributes to an exaggeration of self reports of pain relief (^{Linton and Melin, 1982; Jamison et al., 1989; Mathias et al., 1995; Feine et al., 1998}). This exaggeration was shown to result from inaccurate memory of pretreatment baseline pain as being more intense than it actually was (^{Feine et al., 1998}). Since reports of pain relief rely on memory of pretreatment pain intensity, exaggerations of pretreatment pain ratings and hence magnitudes of self reported pain relief may occur among numerous studies that use self reports of pain relief, including those resulting from placebo treatments. However, the possible enhancement of placebo analgesia by this memory mechanism has yet to be explicitly tested. Furthermore, mechanisms of such enhancement, such as mediation by expectancy or desire for relief, are also of interest.

In this report, we compare concurrent placebo analgesic effects with those based on remembered pain, both of which were produced by a placebo manipulation similar to that used previously (^{Voudouris et al., 1990; Montgomery and Kirsch, 1996}), as described above. Both types of placebo effects are analyzed with regard to their possible mediation by expectancy and desire for pain reduction.

2. Methods

2.1 Participants

Participants consisted of 40 undergraduate students (24 female and 16 male; mean age=19.3, range=18–22 years; 34 Caucasian, 1 African–American, 1 Asian, and 4 Hispanic). They volunteered to participate in exchange for partial course credit and reported an absence of medical conditions and medication use that might interfere with pain sensitivity or pose an increased risk of resultant tissue damage. These conditions and medications included high blood pressure, circulatory problems (e.g. Raynaud's syndrome or family history of the same), diabetes mellitus, heart disease or any other heart problems, asthma, seizures, frostbite, past trauma to hands, lupus erythematosus, arthritis, other large or small joint disease or injury, or use of psychoactive drugs, analgesics, antihistamines, and anti-inflammatory medications. Participants were randomly assigned to two groups (see below) with the restriction that there were the same proportions of males and females in each condition. Forty participants successfully completed all phases of the study protocol. Six participants were excluded from participation in the study because of medical reasons or because their calibrated pain stimulus levels would have been higher than 50°C.

2.2 Setting and materials

Pain was induced by means of a Peltier thermal probe. This device provides brief skin temperature increases within the noxious range (45–51°C) with up to a 10°C/s rise time from a 35°C adapting temperature. It employs a feedback controlled contact thermode whose contact surface (to skin) is 3 cm². An attractive feature of this device is that it provides a quick return to the adapting temperature via a water-circulating cooling mechanism. This prevents progressive warming of the skin. The controlling temperature of the thermal probe allows the experimenter or computer control of the temperature at the thermode-skin interface. Nociceptive thermal stimuli of 5 s duration were applied to the skin by means of the contact thermal probe. The thermode-skin temperatures rose rapidly (10°C/s) from a baseline of 35°C to a peak temperature preset on the stimulator and held at peak temperature for the remainder of the 5-second stimulus. All stimuli were 5 s in duration in order to minimize cues to participants that would lead them to expect differences in pain intensity based on stimulus duration. This was especially important for post-manipulation trials in which participants received the identical stimulus intensity and duration for all three areas (i.e. A, B, and C).

The placebo cream was mixture of iodine, oil of thyme, and water that produced a brownish, medicinal smelling effect when applied topically. The placebo was placed in two medicinal looking bottles labeled, ‘Trivaricaine-A’ and ‘Trivaricaine-B’: Approved for research purposes only’. These placebo agents were applied to areas on the ventral part of the forearm, which had corresponding labels of A and B. A third area, labelled C, was used for control trials in which water from a bottle labelled C was applied. This procedure provided a baseline control, which included effects of vehicle solution on pain stimuli. An office in the University Health Center Infirmary was used as the setting for the experiment to alleviate skepticism on the part of participants that might have been caused by prior associations with the psychology department

2.3 Response measures

In each experimental session, well-validated visual analogue scales (M-VAS) were used to measure pain sensation intensity and degree of unpleasantness evoked by the nociceptive thermal stimuli described above (^{Price et al., 1994}). Subjects were instructed how to rate both pain sensation intensity and unpleasantness according to standardized written statements described in detail elsewhere (^{Price et al., 1985, 1994}). As in previous applications of these scaling methods, verbal anchors served to establish the distinction between these two pain dimensions. The VAS sensory scale was anchored at the left by the descriptors ‘no pain sensation’ and at the right by ‘the most intense pain sensation imaginable’. Likewise, the VAS-unpleasantness was anchored by the descriptors ‘not at all unpleasant’ and ‘the most unpleasant imaginable’. Expectancy measures were obtained by asking participants ‘What do you expect the pain intensity to be with the Trivaricaine-A?’ in area A and ‘What do you expect the pain intensity to be with the Trivaricaine-B?’in area B and ‘What do you expect the pain intensity to be without the Trivaricaine in area C where only the wetting agent is applied?’ Pain expectancy ratings were made on the same VAS as those on the pain intensity scale. Participants rated desire for relief on VAS just after they were instructed about the number of trials they would endure during the post-manipulation trials. This VAS was anchored as ‘no desire for relief’ at the left and as ‘the most intense desire for relief imaginable’ at the right.

2.4 Procedure

Participants were greeted in the lobby of the University Health Center, escorted to an office by a research assistant, and introduced to the experimenter, who wore a white lab coat and was described as a ‘Behavioral Medicine Researcher’. Although the experimenter was not blind with respect to procedures and treatments given, all instructions to participants that are described below followed a standardized script given to all participants. Participants were told that a new, topical, local anesthetic was being tested for its pain reducing effects. They were told the drug's name was ‘Trivaricaine,’ and that it has been proven effective in reducing pain in preliminary studies at other universities. The number of pain trials was described to each participant as part of the informed consent procedure and they were told that they could discontinue participation at any time, without negative consequences. Those participants agreeing to participate then completed the medical screening form. Acceptable participants were randomly assigned to one of two experimental manipulation groups (see Section 2.8): (1) those receiving instructions presumed to elicit high desire for pain reduction (n=20) and (2) those receiving instructions presumed to elicit low desire for pain reduction (n=20).

Participants in the two groups were given 4 blocks of pain trials that included: (1) familiarization trials (n=4), (2) calibration trials (n=8), (3) manipulation trials (n=30), and (4) post-test trials (n=6) identical for both groups, as shown in Fig. 1. Order of administration of baseline and placebo trials was randomized during manipulation and post-manipulation trials.

Fig. 1: Time course of experimental conditions and experimental design.

2.5 Familiarization trials

One trial each of 44, 45, 47, and 49°C stimuli were given in ascending order to familiarize participants with the range of stimulus intensities to be used in the experiment and to quickly identify those few participants unwilling to participate in the study.

2.6 Calibration trials

To control for individual differences in pain perception, a calibration procedure similar to that of ^{Montgomery and Kirsch (1996)} was utilized. Two trials each of 44, 45, 47, or 49°C stimuli were randomly presented during the calibration session to establish a stimulus-response curve (based on linear regression analysis). Care was taken to ensure that subjects could not determine which of the four temperatures would be presented on a given trial. Nociceptive thermal stimuli extending from 44 to 49°C were chosen because the stimuli chosen span a broad range of intensity, extending from a level close to pain threshold (44°C) to a level that is close to but nearly always less than pain tolerance {>50°C (^{Price et al., 1980, 1994})}. Pain trials were separated by approximately 15–20 s and participants rated pain intensity and unpleasantness immediately on VAS following each trial. At the conclusion of calibration trials, a separate regression equation was calculated for each participant by inputting the VAS ratings and their corresponding stimulus intensity levels in°C. This was used to calculate the stimulus intensity levels corresponding to each individual's ratings of ‘6’ (applied to area C, control solution), ‘5’ (applied to area B, ‘weak’ placebo), and ‘2’ (applied to area A, ‘strong’ placebo) as predicted by the regression equation. For brevity, this number was used to indicate the corresponding stimulus intensity level, although this level differs between individual participants according to their personal perceptions of pain.

2.7 Manipulation trials

Manipulation trials began after the conclusion of the pretest calibration trials. All participants in high and low desire groups received ten stimulus trials with ‘strong’ placebo (Tivaricaine A), ten stimulus trials with ‘weak’ placebo, and ten stimulus trials with the vehicle solution (C). These three agents corresponded to bottles labelled A, B, and C, respectively, and were applied to three corresponding forearm regions also labelled A, B, and C (marked with labels/band-aids). The locations of the three forearm regions were randomly selected across the 40 participants. Participants were informed that bottle C was a control wetting solution and that bottles A and B contain different strengths of the local analgesic Trivaricaine. They were further informed that the study was double blind with neither the participants nor the investigators knowing whether bottle A or B contained the highest strength local analgesic. The conditioning procedure represents an extension of that used by ^{Voudouris et al. (1990) and Montgomery and Kirsch (1996)}. Participants were given the stimulus strength corresponding to levels 5 and 6 (based on individual regression equations) for areas B and C, respectively. They were given a stimulus strength corresponding to level 2 for area A in order to provide participants with an experience of pain reduction associated with a strong local analgesic (A) in comparison to baseline control (C). In effect, trials in areas B and A provided a basis for evaluation of effects of weak and strong placebo manipulations, respectively. Similar to previous use of this type of paradigm, participants were not informed that stimulus intensity levels were being altered on placebo trials in which stimuli were presented to areas A and B. The assessment of expectancy and desire for pain reduction was made just prior to the post-manipulation trials.

2.8 Instructions designed to alter desire for relief

After the manipulation trials and just prior to post-manipulation trials, all participants were informed about the level of pain they would receive and the maximum number of trials they would be given in the posttest trial condition. Participants in the high desire group were instructed that they would receive 30 trials of the heat stimulus at the highest pain level they had experienced thus far and that this may be difficult to endure if the local analgesic was not working. Further, they were told that they would receive ten trials applied to area A, ten trials to area B, and ten trials to area C. Participants in the low desire group were instructed that they would receive six trials at a stimulation level that may be at the highest level they had experienced so far, but since there were only six trials, most participants had found this easy to endure. They were then told that two trials each would be applied to areas A, B, and C. All participants received 6 post-manipulation trials, two trials each for areas A, B, and C. Those in high desire groups knew this only at the end of the six trials. Post-manipulation testing began 3 min after the conclusion of the manipulation trials and after the instructions for the number of trials to be endured. For all participants, the procedures for the post-manipulation trials were similar to that of the manipulation trials except that the intensity level remained at the same level (i.e. 4) for all three areas (A, B, and C). Stimulus trials applied to areas A, B, and C were delivered in random order (Symbol). Desire for relief and expected pain levels for each of these three agents were assessed prospectively just prior to the beginning of the post-manipulation condition.

Symbol

2.9 Post-experimental ratings of pain

Approximately two min after the post-test trials, participants rated pain they remembered experiencing during these trials. The following instructions were given and questions were asked: We are very interested in your experience during the last six trials in which you received the test stimuli. Please take a few minutes and answer the following questions to the best of your ability. A VAS was provided for each of the following questions.

1. What was the pain sensation intensity without the Trivaricaine-A or B?
2. What was the pain sensation intensity with Trivaricaine-A? Trivaricaine-B? (2 ratings).
3. What was the pain unpleasantness without Trivaricaine-A Trivaricaine-B? (2 ratings).
4. What was the pain unpleasantness with Trivaricaine-A? Trivaricaine-B? (2 ratings).

3. Results

The experimental paradigm employed herein represents a mixed model containing a within-subjects repeated measures factor (i.e. an expectancy manipulation in the form of control, ‘weak’ placebo, and ‘strong’ placebo analgesic solutions provided to each participant) and a between-subjects factor (i.e. a desire manipulation in which participants were assigned to either high desire or low desire conditions). Therefore, the main analytic strategy consisted of 2×3 (desire by expectancy) mixed model analyses of variance (ANOVA) on expected, concurrent and retrospective ratings of pain intensity and unpleasantness. Parametric analyses were utilized because the specific VAS used has been shown to have ratio scale properties (^{Price et al., 1994}).

3.1 Effects of experimental manipulations on ratings of pain during manipulation trials

Based on the regression analysis of calibration trials, the stimulus intensities chosen were predicted to evoke pain intensity ratings of ‘6’, ‘5’, and ‘2’ for areas C, B, and A, respectively.

The overall mean ratings of pain intensity collapsed across the 10 manipulation trials for areas C, B, and A were 6.2 (SD=1.3), 4.5 (SD=1.1), and 2.5 (SD=0.9), respectively, and reasonably close to levels predicted on the basis of participants’ regression analysis of calibration trials.

3.2 Effects of experimental manipulations on expectations and desires for pain reduction

The experimental manipulations that were designed to produce differences in expectation and desire for pain reduction succeeded in doing so. Means and standard deviations of expected pain and desire for pain reduction are presented in Table 1 and Fig. 2. A series of 2×3 (desire by expectancy) mixed model analyses of variance (ANOVA) on expectancy scores indicated a significant main effect for the conditioning manipulation on both expected intensity, F(2,76)=168.94, P<0.001, and expected unpleasantness, Symbol, P<0.001. Post-hoc contrast tests, with alpha set at P<0.05 indicated that participants expected less intense pain at area A (mean=2.95, SD=1.38) than at area B (mean=4.30, SD=1.18), and less intense pain at area B than at area C (mean=7.25, SD=1.37). Similarly, post-hoc contrast tests revealed that participants expected less unpleasant pain at area A (mean=2.57, SD=1.48) than at area B (mean=4.06, SD=1.33), and less unpleasant pain at area B than at area C (mean=7.13, SD=1.71). The analysis also revealed a significant effect of the desire manipulation on desire ratings, Symbol, P<0.05. Participants who were told that they would have to endure 30 more trials at a high intensity level had a mean desire rating of 6.5 (SD=2.4), whereas those who were told that they would receive only 6 more trials had a mean desire rating of 4.5 (SD=2.4).

Symbol

Symbol

Table 1: Mean expected pain ratings as a function of desire/expectation

Fig. 2: Pain sensation intensity and pain unpleasantness ratings of 40 participants for three areas of skin (A, B, and C). Area C was untreated with conditioning trials, area B was treated with weak placebo conditioning trials, and area A was treated with strong placebo conditioning trials. These ratings are for: (1) expected pain within areas C, B, and A; (2) pain just after placebo conditioning trials for these same areas, that is, concurrent pain; and (3) remembered pain about 2 min after post-test trials for these same areas.

3.3 Effects of manipulations on ratings of pain in placebo-treated and untreated areas

Means and standard deviations of concurrent ratings of pain sensation and pain unpleasantness are displayed in Table 2. Similar to pain ratings during manipulation trials, the mean rating of pain intensity within the control area C (mean=4.56, SD=1.7) was reasonably close to actual stimulus intensity (4). A series of mixed model ANOVAs on concurrent pain ratings did not yield a significant effect of the desire manipulation or the interaction of the desire and expectancy manipulations. However, these analyses indicated a significant effect for the conditioning manipulation on both intensity, Symbol, P<0.001, and unpleasantness, Symbol, P<0.001. Post-hoc contrast tests, with alpha set at P<0.05, showed that the mean rating of stimulus intensity in area A (mean=3.32, SD=1.46) was significantly lower than that for area B (mean=4.00, SD=1.68) and that the mean rating for area B was significantly lower than that for area C (mean=4.56, SD=1.70). Therefore, a definite placebo effect was obtained in the case of pain sensation intensity for areas A and B. In contrast, post-hoc contrast tests revealed that the mean rating of unpleasantness at area A (mean=2.52, SD=1.33) was lower than that for area B (mean=3.43, SD=1.51) and for area C (mean=3.81, SD=1.74). However, the difference in unpleasantness between areas B and C was not significant. Thus, a placebo effect on ratings of unpleasantness occurred only for area A.

Symbol

Symbol

Table 2: Mean pain ratings during placebo treatment and during post-trial questioning

3.4 Effects of manipulations on remembered pain levels in placebo-treated and untreated areas

Means and standard deviations of remembered pain intensity and unpleasantness are displayed in Table 2. A series of 2×3 (desire by expectancy) mixed model analyses of variance (ANOVA) on ratings of remembered pain failed to yield a significant effect for desire or interaction of desire and expectancy but did indicate significant effects for the conditioning manipulation on both intensity, Symbol, P<0.001, and unpleasantness, Symbol, P<0.001. Post-hoc contrast tests, with alpha set at P<0.05, indicated that participants remembered less intense pain at area A (mean=3.02, SD=1.60) than at area B (mean=4.72, SD=1.60), and less intense pain at area B than at area C (mean=6.82, SD=1.52). Thus, definite placebo effects were obtained in the case of remembered pain intensity for areas A and B. Similarly, post-hoc contrast tests revealed that participants remembered less unpleasant pain at area A (mean=2.46, SD=1.44) than at area B (mean=3.84, SD=1.67), and less unpleasant pain at area B than at area C (mean=6.44, SD=1.65). Thus, unlike concurrent ratings of pain intensity and unpleasantness, strong placebo effects on ratings of remembered pain intensity and of unpleasantness occurred for both areas A and B.

Symbol

Symbol

3.5 Associations among ratings of desire and expected, concurrent, and remembered ratings of pain in placebo-treated and untreated areas

Consistent with the ANOVA and post-hoc tests for possible group one (high desire) and group two (low desire) effects on pain ratings, none of the correlations of desire ratings with expected, concurrent, and remembered ratings of pain intensity and unpleasantness for areas A, B, and C were significant. Table 3 shows correlations among expected, concurrent and remembered ratings of intensity and unpleasantness by experimental area. Consistent with ANOVA analysis of the experimental manipulations designed to induce differences in expectancy, correlations between expected and concurrent ratings of intensity were highly significant for areas A and B, but not for area C. Likewise, correlations between expected and concurrent ratings of unpleasantness were significant for areas A and area B, but not area C.

Table 3: Correlations among expected, concurrent, and remembered pain intensity and unpleasantness ratings for placebo and control solutions

Associations between expected and ratings of remembered pain evidenced a slightly different pattern. Expected and remembered ratings of intensity were highly correlated for areas A and C, but not for area B. Corresponding ratings of unpleasantness were significantly correlated for A and C, but not for B. Finally, concurrent and remembered ratings of pain intensity and unpleasantness for areas A, B, and C were inconsistently correlated at no more than modest levels of strength.

3.6 Comparisons between concurrent and remembered placebo effects

Table 4 compares concurrent and remembered placebo effects across areas A and B and across pain intensity and pain unpleasantness dimensions. Whereas concurrent placebo effects for areas A and B were small (0.37 to 1.28 VAS units), those of remembered placebo effects were consistently over three times larger (2.10 to 3.98). A series of one-way repeated measures analyses of variance (ANOVA) on the four arithmetic difference scores (i.e. area C minus area A; area C minus area B) for intensity and unpleasantness ratings, thereby representing the placebo effect for concurrent and remembered pain, indicated a significant effect of time (i.e. concurrent versus remembered) for intensity, Symbol, P<0.001, and unpleasantness, Symbol, P<0.001. Post-hoc contrast tests, with alpha set at P<0.05 revealed that at area A, participants obtained a significantly larger placebo effect for the remembered intensity of pain (delta=3.81) than concurrent pain (delta=1.24), as well as for the remembered unpleasantness of pain (delta=3.98) than concurrent pain (delta=1.28). Likewise, post-hoc contrast tests showed that at area B, participants experienced a significantly larger placebo effect for intensity of remembered (delta=2.10) versus concurrent pain (delta=0.56), as well as for the unpleasantness of remembered (delta=2.60) versus concurrent pain (delta=0.37). This appears to result mainly from the fact that remembered baseline levels of pain intensity and unpleasantness were much larger than concurrent ratings of these dimensions, as shown in Fig. 1.

Symbol

Symbol

Table 4: Concurrent and remembered placebo effects

4. Discussion

Expectancy but not desire for relief was confirmed as contributing to the magnitude of placebo analgesia. A placebo effect was shown by statistically reliable reductions in pain ratings after conditioning trials had taken place. This result indicates that the pairing of surreptitiously lowered pain stimulus intensity with placebo administration results in subsequent placebo analgesia. The effect was graded according to the extent of surreptitious lowering of stimulus strength during the manipulation trials, a result that also is consistent with conditioning. This placebo effect also was manifested both within sensory and affective dimensions of pain. However, expectancy contributed to a large proportion of the variance in sensory and affective pain ratings within areas treated with placebo cream, providing further evidence that the conditioning effect is mediated by expectancy. The results did not confirm the hypothesized factor of desire for relief as having a mediating role in placebo analgesia.

Placebo effects assessed by ratings of remembered pain intensity and unpleasantness were over three times greater than concurrent placebo effects assessed during the post-test stimulation trials. Although placebo effects based on ratings of remembered pain were much greater than concurrent placebo effects, they were also strongly associated with expectancy ratings (Table 3). These combined results further establish expectancy as a causal factor in placebo analgesia and extend previous work by showing that expectancy has a role in both placebo effects based on concurrent and remembered pain.

4.1 Further evidence that placebo analgesic effects are mediated by expectancy

Similar to ^{Montgomery and Kirsch (1996)}, a strong association occurred between expectancy levels and pain ratings of stimuli presented to areas treated with placebo cream (i.e. placebo-treated areas B and A). These results further support Montgomery and Kirsch's conclusion that classical conditioning may occur, but that the proximal mediator of placebo analgesia is that of expectancy. The present study further extends their conclusions in three ways. First, it demonstrates that the expectancy hypothesis is confirmed across different types of experimental pain, iontophoretically induced pain in their study and heat evoked pain in the present one. Second, it demonstrates that increasing the strength of the conditioning manipulation, the degree of surreptitious lowering of stimulus intensity during manipulation trials, produces increasing levels of expectation of pain reduction and increasing magnitudes of placebo analgesia. Third, it provides evidence in support of the hypothesis that placebo effects based on actual pain as well as placebo effects based on remembered pain are both mediated by expectancy.

The results also confirm those of a second study by ^{Montgomery and Kirsch (1996)} that provided evidence that psychological mediation of placebo analgesic effects involves much more specific mechanisms than the simple reduction of anxiety or other global effects on emotions. They demonstrated that the application of a placebo in the guise of a topical anesthetic produced reduction in pain at a body site at which the placebo anesthetic agent was administered but not at a control body site. Controlled mechanical pain stimuli were administered simultaneously to treated and untreated fingers for one group of participants and sequentially for another. For both groups, reduction in pain occurred on the finger that was treated with the placebo analgesic but not on the untreated finger, thereby indicating a spatially restricted mechanism. Similarly, the present study showed placebo effects for skin areas treated with ‘strong’ placebo cream (A) and not nearby skin areas (C). Results of both studies suggest that not all placebo effects are mediated by such global mechanisms as anxiety reduction or hormonal release of endogenous opioids. Such specificity in response would be consistent with a highly specific response expectancy on the part of the participant (^{Kirsch, 1990}).

It is interesting that the magnitude of the placebo effect was consistent with the experimental pairings. Trials in which the placebo was paired with a large decrease in pain stimulus intensity (i.e. the strong placebo condition) resulted in a greater placebo analgesic effect than trials in which the placebo was paired with a small decrease in pain stimulus intensity (i.e. the weak placebo condition). This finding is consistent with both response expectancy theory (^{Kirsch, 1990}) and stimulus substitution models of classical conditioning. Although it was not the purpose of this study to compare the relative contribution of classical conditioning factors and response expectancies to the gradient in placebo effectiveness, relations between the placebo conditions and participants’ expectations established are consistent with a response expectancy mechanism. Based on our previous work (^{Montgomery and Kirsch, 1996}), it seems likely that nature of the pairings (i.e. strong and weak placebo) contribute to participants’ expectations which in turn determine the resulting placebo analgesic effect. Simply put, it appears that greater expectations resulted in greater placebo pain reduction. However, one must acknowledge that a stimulus substitution explanation of the magnitude gradient found in placebo analgesia can not be formally ruled out.

Another possible limitation of this study is that the noxious stimuli, instructions to participants, and placebo manipulations were not given in a fashion unknown to the experimenter. Thus, it is possible that the experimenter's expectations could have been communicated to the participants. We attempted to control for this possibility by having the experimenter follow standardized experimental procedures (including a script) when giving instructions to the participants (see Methods). Although it is still possible that subtle cues were at least partly responsible for the expectancy effects, this does not critically influence the major conclusions drawn from this study. The purpose of the study was not to establish how expectancies are changed, but rather to investigate the effect of altered expectancies on concurrent and retrospective pain reports. In a sense, it doesn't matter whether the expectancies were changed by the manipulation or by the subtle cues from a non-blind experimenter, as also would be the case for placebo effects that occur under many clinical circumstances. Regardless of how expectancies were generated, evidence is provided that expectancy mediates placebo analgesia.

4.2 The possible role of ‘desire for relief’ in placebo effects

One of the major hypotheses, that desire for pain reduction contributes to the magnitude of placebo analgesia, was not supported by the results nor was there a trend in the direction of a direct positive influence of this factor. The manipulation designed to modulate desire for pain reduction resulted in two groups of participants with significantly different magnitudes of desire for pain reduction. Despite the effectiveness of this approach, a difference in placebo effect did not occur between the two groups nor was there a significant association between this factor and magnitude of placebo analgesia. Taken together, these results indicate that desire for pain reduction is not a contributing factor in placebo analgesia under these experimental conditions.

Although these results cast doubt on the possible contribution of desire for pain reduction to placebo analgesia, there are reasons that the hypothesis should not be completely rejected at this point. First, nearly all participants had some degree of desire for pain reduction and so it is possible that some minimal level of desire for pain relief may be necessary for placebo analgesia. Second, desire for reduction in pain may be a more critical factor in placebo effects during clinical pain studies and have much less application to placebo effects in experimental pain. This issue may be particularly relevant when the test stimulus is brief, as in the present study. Desire for relief may be much more of a factor when the pain is threatening or has a more uncertain duration and therefore likely to induce fear or anxiety, as would be present in many instances of clinical pain. It has been suggested that the generally larger placebo analgesic effects in clinical as compared to experimental pain studies may reflect a stronger desire for pain relief in the former (^{Price, 1988; Price and Fields, 1997}). This possibility needs to be evaluated in clinical studies that use measures of expectancy and desire for pain relief and that provide both placebo and natural history control conditions.

The lack of effects from our desire manipulation contrasts with ^{Jensen and Karoly (1991)} findings that desire to experience sedative effects from a drug resulted in greater placebo sedative effects. However, their desire manipulation was that of providing instructions to the effect that response to the drug meant that the participants had more favorable personality characteristics, whereas that of the present study relied more on participants’ intrinsic desire for pain relief. It may be that some types of motivations or desires are more effective than others and/or that ‘desire for pain relief’ has limited applicability to our experimental paradigm.

4.3 Placebo effects based on remembered versus actual pain

When participants were asked to remember and rate pain intensities and unpleasantness levels within each of the 3 areas during the last six trials, these ratings closely followed their expected but not actual pain levels associated with post-manipulation trials (Fig. 2). As shown in Fig. 2, participants remembered the untreated pain within control area C as being much more intense and unpleasant than it actually was. This distortion was less for area B and minimal for area A. Selective distortions, in turn, resulted in large placebo effects for remembered pain, effects that were over three times larger than that of concurrent placebo effects (Fig. 2; Table 4). Moreover, these large placebo effects occurred for both areas A and B and for both sensory and affective dimensions. However, similar to concurrent placebo effects, they were strongly associated with expectancy ratings. Another line of evidence that placebo effects based on remembered pain follow participants expectations of pain more closely than actual pain intensities is that systematically higher correlation coefficients were found between expected and remembered pain than between actual and remembered pain (Table 3).

The selective exaggeration of remembered pain intensity within untreated area C and the consequent enhancement of placebo analgesia are consistent with previous studies that show that memory distortion of pretreatment pain contributes to an exaggeration of self reports of pain relief (^{Feine et al., 1998; Mathias et al., 1995}). ^{Feine et al. (1998)} found that pain relief scores based on memory were over three times higher than those based on pre-treatment minus present pain VAS ratings, consistent with the results of the present study of placebo analgesia. This similarity exists despite the facts that the former study was based on data obtained from temporomandibular dysfunction (TMD) patients who received active treatments for 10 weeks and the present study was based on pain free volunteers exposed to brief experimental pain stimuli for about one hour. Feine et al. also found that overestimation of pretreatment pain was largest for patients whose pretreatment pain was low to moderate (<50 mm on 100 mm VAS). They point out that since pain associated with many chronic pain conditions is often low to moderate, it seems likely that most patients will report high levels of relief for small actual changes in pain intensity. Since the mean baseline pain rating of post-manipulation trials was about four on a scale of ten in the present study, a large distortion in remembered baseline pain would be consistent with conclusions of Feine et al. The present results extend their observations by providing evidence that their explanation for exaggerated estimates of pain relief apply to placebo treatments and in showing that, similar to concurrent placebo effects, remembered placebo effects are predicted by participants’ expected pain levels (Fig. 2, Table 3). Based on the combination of present and previous results (^{Montgomery and Kirsch, 1996}), it is likely that expectancy mediates both placebo effects based on actual pain as well as those based on remembered pain intensities.

As pointed out by ^{Feine et al. (1998)}, patients’ reports of relief following treatment are often used to establish the effectiveness of treatments. To the extent that such measures are used in clinical studies of pain treatments, estimates of magnitudes of analgesic effects from both placebo and active treatments are likely to be significantly enhanced when reports are made retrospectively. One has to openly wonder about the extent to which commonly quoted magnitudes of placebo analgesia (^{Beecher, 1955, 1959; Turner et al., 1994}) as well as reported magnitudes of analgesia from active treatments have been based on retrospective judgments of pain relief. To take a relevant example, decisions as to whether and to what extent patients with chronic non-malignant pain should rely on opioid analgesics are likely to be critically determined by whether their efficacy is assessed concurrently or retrospectively by clinicians who prescribe them and by researchers who study them.

The present study demonstrates the importance of response expectancies in determining both concurrent placebo analgesic effects as well as placebo analgesic effects based on remembered pain and provides further evidence that retrospective measurement of any treatment effect may exaggerate estimates of treatment efficacy. These findings are consistent with the growing literature on expectancy effects in clinical medicine (^{Montgomery et al., 1998}) and future research should determine whether the findings of the present study generalize to treatments of other clinical problems (e.g. nausea, distress, fatigue).

Acknowledgements

The authors gratefully acknowledge the invaluable assistance of Jennifer Lapinski, Joshua P. Redford, Andrea Sutherland, Erin Huber, Valerie Tenore, Stephanie Hays and Ray Almeida. We also are grateful to Dr. Charles Vierck for lending us his thermal stimulator.