Is it personality or genes? – A secondary analysis on a randomized controlled trial investigating responsiveness to placebo analgesia

2.1 Design

The current exploratory analysis was part of a randomized controlled trial with an 8 × 2 between–within mixed design. Participants were randomly assigned to one of eight groups (between-subject condition), in which a participant either learned through different (combinations of) psychological learning mechanisms that a placebo device (an inactivated TENS) could alter their pain, or did not (control group). The implemented learning mechanisms were verbal suggestions, classical conditioning, and/or observational learning, which are often utilized to induce placebo effects in research [6,23,25,26,27,28,29]. The experiment consisted of two phases: a learning phase, which served to induce expectations, and a test phase, in which placebo-analgesic effects were assessed. In the learning phase, participants were made to believe through three different experimental manipulations related to different learning mechanisms (i.e., a verbal suggestion, a conditioning paradigm, or a social-observation video), or a combination of two or three of these manipulations, that the placebo device could lower their pain perception. The experimental verbal suggestion manipulation consisted of a standardized message informing the participants that when the placebo device was activated, they could expect their pain perception to be lowered as compared to when the placebo device was not activated. In addition, a mock calibration procedure was adopted to convince participants even more that the device was indeed showing activity. However, after the mock calibration procedure, the device was, unbeknownst to the participant, completely reduced in electrical activity. During the experimental conditioning paradigm manipulation, participants repeatedly received low heat pain stimuli along with the word “ON” on a computer screen with a colored border (purple or yellow) and moderate heat pain stimuli along with the word “OFF” on the same screen with a colored border. The heat pain stimuli that were given while participants were shown “ON” were considered active trials, while heat pain stimuli that were given with participants being shown “OFF” were inactive trials. The social-observation video manipulation displayed a 4-min mock conditioning paradigm with an actress, who consistently reported less pain when the placebo device was active, and more pain when it was inactive. The participants assigned to groups that did not receive all three of the different experimental manipulations were also exposed to control manipulations in order to prevent systematic differences between groups in terms of design differences other than the learning mechanisms. A detailed overview of all experimental and control manipulations per study group is shown in Table 1. In the test phase, individual placebo effects were tested by looking at a systematic difference in experienced pain intensity between the first three active and first three inactive trials (within-subject condition) while participants received heat pain stimuli that were similar in intensity. A more elaborate description of the experimental design was published elsewhere [24].

Table 1

Overview of specific experimental or control manipulations per study group aimed at inducing placebo analgesia by means of different learning mechanisms, or merely serving as control

Manipulation	Group
	CTRL	VS	CC	OL	VS + CC	VS + OL	CC + OL	VS + CC + OL
Verbal suggestion
Conditioning paradigm
Observational video

CC = classical conditioning, CTRL = control, OL = observational learning, T = temperature, VS = verbal suggestion.

2.2 Participants

The study investigated healthy individuals who were aged 18–35 years and were English or Dutch speaking. Participants were excluded if they experienced pain on the day of testing (either “yes” or “no”) or had a severe physical or mental disability; used long-term medication that might interfere with pain perception (e.g., anti-epileptics) or used prescription analgesics 24 h prior to the day of testing or over the counter analgesics within 12 h before testing; used >2 alcoholic units or soft drugs within 12 h before testing or any hard drugs (e.g., amphetamines) within 48 h prior to testing; had experience with placebo research; were pregnant or currently breastfeeding; had injuries on the hands, wrists, or arms at the location of the thermode or TENS electrode; had a pacemaker or implantable cardioverter-defibrillator; and were unable to feel a minimal difference in pain score ranges (low-mid) after calibration of the heat pain, which makes application of different pain stimuli impossible.

The sample size for the study population was calculated for the primary analysis of the main fundamental study. A rough estimation for the number of participants required for this exploratory secondary analysis was based on the previous work by Colloca et al., as that study has a lot of similarities (e.g., same genes, experimental setup, similar learning mechanisms used, healthy individuals, placebo analgesia). Based on that study, we intended to collect approximately 25 samples of participants per condition (n = 200). Participants were randomized to the study groups by means of a block randomization scheme (block size = 8), stratified for sex (3:1 female/male ratio). The randomization was done by an independent researcher who also printed out the allotments and stored them in sealed envelopes.

2.3 Heat pain stimuli

Experimental pain was induced throughout the experiment with a thermal probe (TSA-II, Medoc, Israel) that was attached to the participant’s non-dominant lower arm. The heat pain stimuli were short-lived (4 s) and started from 32°C with a maximum possible temperature of 50°C. Peak temperatures for the heat pain stimuli were individually titrated to a low (numeric rating scale [NRS] ranged from 1 to 3) or moderate pain level (NRS ranged from 4 to 6) following a standardized protocol for quantitative sensory testing [30]. In the learning phase, a total of 18 low pain stimuli and 18 moderate pain stimuli were subsequently used to create an experimental conditioning paradigm or a control conditioning paradigm. The two temperatures corresponding to a low or moderate pain stimulus in the experimental conditioning paradigm were coupled with a color cue on a screen (purple or yellow) to induce a positive association that the placebo device could lower pain. The cue indicated whether the device was working (i.e., active trial) or not working (i.e., inactive trial). The two temperatures in the control conditioning paradigm were delivered at random, which prevented the formation of an association. In the test phase, a total of 36 moderate heat pain stimuli were delivered to observe any placebo-analgesic effects and their extinction, as reported by the participant. The thermal probe was moved every 24 pain stimuli to prevent any habituation or sensitization of the skin.

2.4 Placebo device

The placebo device was an inactive transcutaneous electric nerve stimulation (TENS) device that is commercially available (Beurer EM 80). The TENS was described as a “physical pain transducer” (PPT). Participants were connected to the PPT with two electrodes on the non-dominant lower arm. The (fake) activation (active trial) or deactivation (inactive trial) of the PPT device was displayed as a word “ON” or “OFF” along with a colored border (purple or yellow) on a computer screen through E-prime 3.0.

2.5 Measures

2.6 Procedure

Participants were recruited from August 2020 up until March 2022 through the university’s online participant recruitment system, social media, flyers around the university, and direct personal contacts. Interested participants were informed beforehand that the aim of the study was to investigate the influence of the mind-body interaction on heat pain. The experiment was conducted in a single experimental session that lasted for about 150 min. On the day of testing, participants received a detailed instruction about the experiment after which they provided written informed consent. Participants were explicitly informed about the DNA collection procedure and provided separate informed consent for the saliva collection. Next, participants filled out the online questionnaires in Qualtrics regarding their psychological characteristics. After this, the experimenter calibrated the heat temperature to pre-specified pain levels of the participant, with low pain corresponding to an NRS of 1–3, moderate pain corresponding to an NRS of 4–6, and high pain corresponding to an NRS larger than 7. Subsequently, the experimenter opened a randomization envelope to inspect the allocated intervention group, which was blinded to the participant. Participants then went through a learning phase, during which their lower arm was attached to the electrodes of the TENS device. They received three learning manipulations: a verbal suggestion, a video, and a conditioning paradigm, which, depending on the allocated group, could be an experimental or control manipulation (Table 1).

After completing the learning phase, participants went through a test phase during which they received only moderate heat pain stimuli (NRS: 4–6). Analgesia due to placebo effects was examined by looking at differences between inactive and active trials. At the end of the test phase, participants’ saliva cells were collected with a swab. More specifically, participants were instructed to swab the inside of their cheeks for 1 min. The sample collection procedure was conducted at the end of the experiment to make sure that participants did not consume any food or drinks for at least 30 min. The swabs were then stored in dedicated containers together with a preservation capsule and stored at room temperature (20°C). Then, participants were asked to answer the exit questions and subsequently debriefed about the true goal of the experiment. At the end of the experiment, participants received either a financial compensation (€18.75) or study credits and were thanked for their time.

2.7 Statistical analyses

Statistical analyses were conducted using RStudio version 4.0.1. (PBC, Boston, MA). Descriptive statistics were shown as counts and frequencies for categorical variables and means and standard deviations for continuous variables. Differences in demographic variables between SNPs were tested with chi-squared tests (for categorical variables) or one-way analyses of variance (ANOVAs) (for continuous variables) if their assumptions were met. The assumptions were inspected with QQ plots, residual plots, Shapiro–Wilk tests, and Levene’s tests. Non-parametric testing was applied when assumptions for normality or heterogeneity of variances were violated. Multivariate outliers were detected with Mahalanobis’ distance or, in case assumptions for the Mahalanobis were violated, with Z-scores above or below 3.29. The alpha level for all the analyses was set at 0.05, and for post hoc comparisons, a Bonferroni correction was applied [44].

The moderation of personality characteristics on the effect of the learning mechanisms on placebo analgesia was studied with linear multiple regression analyses. Optimism, neuroticism, worrying, empathy (four domains), and somatosensory amplification were analyzed by conducting distinct simple moderation models. The predicting variable was the different study groups, which were constructed with dummy coding where the control group served as the reference to all other groups [45]. The moderation models included an interaction term with the groups and the individual characteristics. As the predicting variable was multicategorical, i.e., eight different groups, the evidence for statistically significant moderation was obtained by testing the model fit of the omnibus F-test from the interaction model to the omnibus F-test from a multiple linear regression model without an interaction term [46]. If significant, standardized regression coefficients (β) were displayed as effect measures for the moderation analysis, and the interaction was subsequently probed with the “pick-a-point” method to visualize the moderation model [45]. The assumptions for the linear models were examined by inspecting the residuals with QQ plots, residual plots, Shapiro–Wilk tests, and Breusch–Pagan tests.

The relation between the SNPs, their possible interactions, the study groups, and placebo analgesia was tested with a linear multiple regression analysis with four fixed factors: three SNPs as predictors and the groups with one or more specific learning mechanisms as the fourth predictor (compared to the control group). The relation of the SNPs with placebo analgesia was explored in two steps: (1) between the individual SNPs and placebo analgesia and (2) between the SNPs, their interactions, and placebo analgesia. This step-wise approach was implemented to account for a smaller number of genotyped data. The group predictor was added to the model as a covariate and corrected for any differences in placebo effects due to the learning mechanisms. For the current analysis, two of the three SNPs were dichotomized as the OPRM-1 rs1799971 genotype G/A and G/G (G carriers) and the FAAH rs324420 genotype A/A and A/C (A carriers) seem to similarly impact placebo analgesia [17]. The other SNP (rs4680) was not dichotomized as all three variations could have a different influence on placebo analgesia [21]. Differences in placebo analgesia between SNPs, their interactions, and the interactions with the different groups were examined with post hoc pairwise comparisons using independent samples T-tests. Effect sizes were reported as standardized regression coefficients. The assumptions for the linear model were tested by inspecting residual plots, QQ plots of the residuals, and Breusch–Pagan tests.

3.1 Demographics

A total of 208 participants were enrolled in the experiment, of which two dropped out for technical reasons, and 197 (94.7%) provided genetic material for DNA analysis by means of saliva samples. From these 197 saliva samples, a total of 123 samples (62.4%) could eventually be genotyped because the quality of the remaining samples was not sufficient to run the PCR required for sequencing (n = 76, 38.6%). Further 15 samples (12.2%) were contaminated that not all SNPs could be obtained during the sequencing. This has led to 206 participants (99.0%) being included in the personality characteristics analyses and 108 participants (51.9%) in the genetic analysis of this study. The demographics and baseline data for all participants and the genotyped participants are provided in Table 2.

3.2 Personality characteristics

The descriptives for the personality characteristic analyses are provided in Table 3. The assumptions for the linear models used to conduct the moderation analyses were not violated. The results from the model fit comparisons indicated that none of the personality traits investigated in this experiment moderated the effect of the study groups on placebo analgesia (optimism – LOT-R: p = 0.489; neuroticism – EPQ: p = 0.546, empathy – IRI-PT: p = 0.614, empathy – IRI-EC: p = 0.770, empathy – IRI-FA: p = 0.637, empathy – IRI-PD: p = 0.913, worrying – PSWQ: p = 0.396, somatosensory amplification – SSAS: p = 0.835).

3.3 SNPs

The linear models of the SNP analyses showed that none of the SNPs, or their interactions, predicted the amount of placebo analgesia when the effect of the allocated groups was corrected for (Omnibus test for model 1: F _{11, 96} = 1.738, p = 0.074, R ² = 0.07, and model 2: F _{18, 89} = 1.51, p = 0.105, R ² = 0.08). The influence of the group predictor on placebo analgesia with this adjusted sample size (N = 108) was significant. Post hoc testing revealed that certain groups with combinations of learning techniques differed significantly from the control group in their respective placebo effects, similar to the original experiment [24].

3.4 Analyses of interaction effects, personality characteristics, and SNPs

As the current results did not indicate that any of the personality characteristics or any of the three different SNPs significantly predicted the amount of placebo analgesia, no further analyses were done to explore the interaction of both factors.

4.1 Clinical implications

Disentangling which individuals respond to placebo effects and which do not could improve patient-centered care in clinical practice. When a patient known to respond positively to placebo effects visits a doctor, the use of placebo effects in the dedicated treatment regimen for the patient could improve health outcomes [49]. Apart from improving the efficacy of the treatment itself, the profound placebo responses in the patient also create the opportunity to decrease medicine dosage and thereby lower side effects and costs [50]. A fitting example from the current clinical practice is patient-controlled infusion of analgesics (PCA). PCA reduces pain and the total dosage of required painkillers more effectively than provider-controlled infusion [51]. The effectiveness depends upon placebo effects, which are induced by the patients’ knowledge, control, and expectation of drug administration [49,52]. Hence, optimizing prediction models for placebo effects is a logical step to improve individualized medical treatments. A practical example could be a passport for every patient that shows their placebo responsiveness, similar to the existing pharmacogenetic passports for drug-metabolizing enzyme activity [53].

However, discovering a person’s responsiveness to placebo analgesia has repeatedly, also in this study, been shown to be difficult. A more recent view is that it depends upon a complex interplay between psychological characteristics, biological markers, and contextual factors [3]. Due to the extent of the complexity seemingly involved in predicting someone’s placebo responsiveness, one could question if it makes sense to assess the responsiveness profiles of every patient. In current-day medicine, physicians are already encouraged to adopt treatment strategies with placebo-evoking benefits such as empathic communication or PCA [54]. Regardless of the placebo effects that will be established in patients who are responsive, these contextual treatment strategies are also important to patients who do not benefit from placebo effects per se. It could therefore be argued that profiling every individual patient to obtain the maximum responsiveness to placebo effects yields only marginal extra benefit on top of the regular contextual treatment strategies. Also, undertaking this profiling, for example, with individual passports, is likely an extensive and costly undertaking, requiring thorough cost-effectiveness analyses before practical implementation [53]. Conducting future studies to evaluate the cost–benefit analysis of implementing placebo responsiveness profiling would be valuable to discover if this line of research is still worth exploring.

4.2 Limitations

Several limitations in this study require addressing. The current study did not examine the role of expectations in placebo responsiveness, which could be an important contextual factor considering its central mediating role in learning and placebo effects [3,24]. The sample size for the genetic analyses was smaller than anticipated, as a substantial amount of saliva samples could not be analyzed. Despite a thorough procedure, factors such as physiological variations in cell quality or contamination could have affected the genetic analysis [55]. Finally, the study population consisted solely of healthy, Caucasian students. This relatively homogeneous population could have limited the generalizability of the results to older, non-Caucasian, or patient populations [56].