Placebo responders versus non-responders: a cross-sectional cohort study for psychological determinants

Introduction

Placebo effects refer to the reduction of symptoms caused by the psychosocial context rather than the properties of the treatment itself [59]. At the individual level, placebo studies have reported a great deal of variability from individual-to-individual (see systematic review, [34]) with varied placebo responsiveness status being documented in both clinical trials [22; 26] and laboratory studies[10; 60]. Some individuals experience substantial placebo effects while some experience even a worsening of pain and other symptoms [64]. The diverse pattern of placebo effects renders the answer to the question “who will benefit from placebos?” inconclusive.

In an attempt to address the individual differences in placebo occurrence and responsiveness, studies have examined various psychological factors and their impact on placebo hypoalgesia [29; 49; 55; 57; 63]. In particular, negative emotions and mood disorders such as anxiety [69], depression [72] and fear of pain [44] have been associated with lower placebo effects [69]. Dopamine-related factors including reward seeking/fun seeking [63] or pre-dispositional optimism [29; 49] have also been examined as predictors of increased placebo hypoalgesia. Social interaction factors (e.g., empathy [32], suggestibility [14]) have been linked to greater placebo effects. Distinct personality factors including neuroticism [55; 57], openness [76], and extraversion have also been linked to changes in placebo hypoalgesia. However, the results of these studies often conflict with one another, creating ambiguity in the impact of psychological factors on placebo effects [31]. Darragh et al. [18] then concluded that no single psychological factor is consistently and exclusively impactful for placebo effects. The aforementioned studies often focus on a few psychological factors without a cohesive framework, involve relatively small sample sizes, and lack a comparison between patients (e.g. chronic pain) and healthy participants.

To address these limitations, we determined the role of higher-level domains of functioning using the Research Domain Criteria (RDoC) framework [33; 61], a matrix with specified domains of psychological characteristics and behaviors, in influencing placebo effects. Placebo effects were induced in chronic pain and healthy participants using classical-conditioning-with-suggestion paradigm with heat thermal painful stimuli tailored to individuals’ pain sensitivities. We aimed to deconstruct psychological domains underlying expectations, learning and subsequent placebo effects to determine critical determinants of individual variability to placebos and the RDoC-based systems which would predict placebo formation phases and predict placebo non-responder phenotypes.

Method

Participants

The current study focused on the psychological determinants of placebo hypoalgesia in chronic pain participants with Temporomandibular disorder (TMD) and healthy controls (HC). The Internal Review Board (IRB) of the University of Maryland, Baltimore approved the study (Prot. HP-#00068315).

Participants within the age range of 18 to 65 years were prescreened over the phone to determine if they were a potential TMD participant or a healthy control. Recruitment took place at University of Maryland School of Nursing from August 5, 2016 to February 1^st, 2020. HC and TMD were recruited simultaneously and over the same time frame using a recruitment strategy with flyers, advertisements, and other recruitment strategies (see Supplementary Materials). All participants were compensated $100 for their time in participating the study. Part of the behavioral findings from a sub-cohort of the overall dataset have been reported in another publication [10]. A total of 834 participants have been screened since the beginning of recruitment, namely 421 TMD participants (102M, 319F) and 413 healthy controls (168M, 245F). 31 out of 834 participants were deemed ineligible due to life dependency on recreational drugs and alcohol, presence of major depression, bipolar disorder, schizophrenia, Attention-deficit/hyperactivity disorder, obsessive compulsive disorder, cancer, pregnancy, and/or use of antipsychotics. One withdrew during the screening procedure, and one was lost during post-screening follow-up, resulting in 402 TMD and 401 healthy controls for the enrollment (See Flow chart in Supplementary Material Fig. S1). In addition, 5 TMD participants and 4 healthy participants were excluded from further analysis because greater than 20% of the psychological questionnaire data was missing. The final analyzable dataset included 397 TMD participants and 397 healthy controls.

Inclusion and exclusion criteria

TMD participants.

402 TMD participants were eligible and enrolled in the experiment. An expert in orofacial pain at the Brotman Facial Pain Clinic at the University of Maryland, School of Dentistry provided in-person clinical examinations with the potential TMD participants who reported pain for a minimum of 3 months in their jaw, temple or ear areas of either side. The participants who met the Axis I Diagnostic Criteria for Temporomandibular Disorders (DC/TMD) [62; 77] were included in the current study. The exclusion criteria were as follows: presence of degenerative neuromuscular, cardiovascular, neurological, kidney or liver disease, pulmonary abnormalities, diffuse cancer within the past three years, any uncorrected impaired hearing, color-blindness, pregnancy, or breast-feeding.

Healthy participants.

401 Healthy participants were eligible and enrolled in the current study. Eligibility was identified according to the following exclusion criteria: suffered from any chronic pain condition, presence of pain disorders, presence of degenerative neuromuscular, cardiovascular, neurological, kidney or liver disease, pulmonary abnormalities, diffuse cancer within the past three years, any uncorrected impaired hearing, color-blindness, pregnancy, or breast-feeding. Volunteers who passed phone screening also underwent an in-person interview by trained personnel to ensure their eligibility.

For both potential TMD and HC participants, an external licensed psychiatrist screened cases of psychiatric problems based on Diagnostic and Statistical Manual of Mental Disorders-5 (DSM-V) [21]. Participants who had severe psychiatric conditions such as mania, dementia, bipolar disorder, schizophrenia major depression, Obsessive-compulsive disorder, or lifetime dependence on alcohol or drugs were excluded from the study.

Experimental procedure

The experimental procedure has been reported in a parent recent study [10]. This study employed a well-established conditioning paradigm [10] with 24 trials in the conditioning phase and 12 trials in the testing phase (Fig. 1a–b). Before starting the experiment, all participants gave written informed consent to this study. Individually-tailored heat pain stimulation was delivered using the ATS 30×30 thermode (PATHWAY System, Medoc, Ramat Yishai, Israel). To determine the temperature used for the conditioning and testing phase, pain sensitivity was assessed using the limits paradigm [28]. Specifically, participants had the ATS thermode applied to their forearms of the dominant hands and were asked to stop the heat stimuli using a remote controller when they felt warmth, minimum pain, and maximal tolerable pain. The temperature increased from 32°C with an increasing rate of 0.3°C/second for warmth and minimum pain detections, and 1°C/second for maximum tolerable pain. The temperature that participants stopped at when they felt warmth, minimum pain, and maximal tolerable pain were recorded as the levels of warm threshold, pain threshold, and pain tolerance limit, respectively.

Fig. 1. Conditioning paradigm to induce placebo hypoalgesia in TMD and HC participants.

(a) Participants were applied to an ATS thermode and a remote controller to stop the heat stimulation when they feel warm (warmth threshold), minimal pain (heat pain threshold) and maximum pain (heat pain tolerance limit). Warmth threshold, heat pain threshold and heat pain tolerance were assessed 4 times and the averaged values were used. 2°C lower than the average stimuli temperature for pain tolerance limit was initially set to be paired with the red screen. 6°C lower than the average stimuli temperature for pain tolerance limit was initially set to be paired with green screen. These two temperatures were further confirmed via subjective pain ratings in response to a 10-second heat stimuli to have 80 and 20 out of 100 VAS corresponding to red and green screens, respectively. Finally, 1°C lower than the stimuli temperature given during the red screen was selected as the level of pain stimulation during the testing phase.

(b) An ATS thermode was applied to the forearm of the dominant hand to deliver heat pain stimuli. A sham electrode device was placed to the same hand above the thermode. Participants were instructed that the electrodes would help to alleviate pain by delivering subthreshold electrical impulses that would reduce pain signals transforming to the central nervous system. After calibrating pain stimuli for high, low and moderate pain experiences, participants went through two conditioning sections (24 trials) and one testing section (12 trials). Each trial started with a red or green screen lasting for 10 seconds. Paralleling with the colored screen presentation, a heat-pain stimulus was delivered through the ATS thermode (10 seconds). After the offset of the colored screen and the painful stimulus, participants were asked to rate their pain experience on a VAS scale from 0=not pain at all to 100=maximum tolerable pain (8 seconds). The inter-stimuli-interval were jittered between 10 to 13 seconds. During the conditioning phase, red screens (high pain conditioned stimulus, CS+)were paired with high-pain stimuli, whereas green screens (low pain conditioned stimulus, CS−)were paired with low-pain stimuli. During the testing phase, both red and green screens were followed by moderate pain stimuli. Expectations about pain relief were measured at baseline, after the condition phase (reinforced expectations), and after the testing phase (overall expectation).

We tested warmth threshold, pain threshold and pain tolerance limit, respectively for four times and averaged the four trial values (Fig. 1a). The average temperatures for heat pain tolerance limit were used to tailor the level of pain stimulation used in the placebo manipulations during the conditioning phase. In particular, 2°C lower than the average stimuli temperature for pain tolerance limit was initially set to be paired with the red screen. 6°C lower than the average stimuli temperature for pain tolerance limit was initially set to be paired with green screen. These two temperatures were further confirmed via subjective pain ratings in response to a 10-second heat stimuli to have 80 and 20 out of 100 VAS corresponding to red and green screens, respectively. Finally, 1°C lower than the stimuli temperature given during the red screen was selected as the level of pain stimulation during the testing phase.

After confirming the heat pain stimuli for red and green screens, both TMD and HC participants were informed that the goal of the experiment was to examine a pain modulation intervention consisting of a subthreshold electrical stimulus that could reduce pain. To achieve that, a sham electrode device was attached to the forearm of their dominant hand. Participants were instructed that the electrode device could activate the skin nerves for pain signal inhibition at the subthreshold level. In this way, participants would not sense the electrode device but they could refer to the colored screens (green or red) to tell whether the device was on or off. A green display screen indicated that the device was turned on, while a red screen was cue indicated that the device was turned off. The participants were not aware that individually-tailored minimal and maximal level of pain stimulation were paired with green and red screens, respectively, during the conditioning phase. 1°C lower than the stimuli temperature given during the red screen of conditioning phase was selected as the level of pain stimulation for both red and green screens during the testing phase. The participants used Celetritas Response System (Psychology Software Tools Inc, Sharpsburg, PA, USA) to rate their pain intensity for each of the heat pain stimulation repetitions on a 100 Visual Analogue Scale (VAS) ranging from 0=no pain at all to 100=maximal tolerable pain.

We assessed expectations about the effectiveness of the devices in mitigating pain experiences before the conditioning (baseline) and post-conditioning trials (i.e., reinforced expectations) by asking the question “How much do you think this procedure will reduce your pain?”. Reinforced expectation was operationalized as the expectations assessed at the post-conditioning phase. After the testing phase, we also assessed the overall perceived effectiveness of the device by asking the question “How much pain relief have you perceived?”. Expectations and perceived effectiveness were measured using 0–100 VAS scale (0=no pain relief to 100=maximal pain relief).

Since deceptive information was used in the study procedure, participants were debriefed about the nature of the study (i.e., to examine conditioning induced placebo hypoalgesia) and the involvement of the deception (i.e., the electronic device was not open and did not influence pain perception) after completing the study protocol. Participants were provided the option to withdraw their data from analysis after debriefing (see Debriefing form – Supplementary Materials), but none of the participants withheld their data use approval.

Psychological questionnaires

Research Domain Criteria (RDoC)

Research Domain Criteria (RDoC) framework [33; 61] provided a matrix with functional constructs representing specific domain of psychological characteristics and behaviors. We adapted the three Domains of the RDoC matrix that may have been closely linked to placebo hypoalgesic effects to determine features of placebo phenotypes in TMD and HC participants. In particular, considering pain as a stressor and aversive stimulation, the negative valence systems in charge of the response to aversive situations may play a role in influencing subjective pain experiences [20; 41]. Positive valence systems, which are primarily responsible for the responses to rewarding motivational contexts such as Pavlovian conditioned responses and reward learning, may facilitate placebo analgesic effects [63]. Social process systems that mediate interpersonal responses may also contribute to the formation of placebo effects given the involvement of interpersonal activity in the placebo procedure [52]. Therefore, the above three domains were adopted and associated psychological questionnaires were measured in the cohort of TMD and HC participants.

Negative valence systems.

According to the RDoC framework [17; 33; 61], negative valence systems are in charge of responses to aversive contextual environment. Those responses may include depression, anxiety, stress, and negative emotional states. Individual psychological questionnaires were then selected based on the definitions of this RDoC domain. In line with this domain, we measured depression and anxiety symptoms using the Depression Anxiety Stress Scale (DASS [67]), Beck’s Depression Inventory (BDI [5]), State-Trait Anxiety Inventory-Trait (STAI-Trait [3; 70]), as well as Mood and Anxiety Symptoms Questionnaire (MASQ [35]). Moreover, Positive and Negative Affect Scale (PANAS [46]) and Life Orientation Test (LOT-R [50]) were used to measure the negative emotional states as well as dispositional optimism and pessimism in the current study.

Positive valence systems.

According to the RDoC framework [17; 33; 61], positive valence systems are mainly responsible for reward seeking or reward/habit learning – a set of psychological and behavioral responses to positive motivational contexts. To assess this domain, we used the Behavioral Avoidance/Inhibition Scales (BISBAS) [7] to assess individual differences in the sensitivity to the behavioral avoidance system (BIS), which measures motive to move away from something aversive and punishment, and the behavioral approach system (BAS), which measures motive to move toward something desired.

Systems for social processes.

Systems for Social Processes are primarily responsible for interpersonal activities such as understanding and interacting with others’ actions. To test the constructs of this domain, we used the Interpersonal Reactivity Index (IRI) [19] to assess reactions of one person to the observed experience of another. It contains 4 subscales that assessed different aspects of interpersonal activities including Perspective Taking (the tendency to spontaneously adopt the psychological point of view of others), Fantasy (taps respondent’s tendencies to transpose themselves imaginatively into the feeling and action of fictitious characters in books, movies, and plays), Empathic Concern (assesses “other-oriented” feelings of sympathy and concern for unfortunate others), and Personal Distress (measures “self-oriented” feelings of personal anxiety and unease in tense interpersonal settings).

Finally, given that personality factors such as neuroticism and openness to experiences may impact the development and maintenance of chronic pain [51] as well as placebo analgesic effects [55; 76], we used the Neuroticism, Extraversion, Openness Five-Factor Inventory (NEO-FFI) [15] to provide information on five domains of personality: neuroticism, extraversion, openness to experience, agreeableness, and conscientiousness. A recent review has proposed big Five personality such as neuroticism is aligned with the negative valence system but is not equivalent to the definitions of the RDoC domains [74].

In addition to the RDoC framework of human behavior domains, we also assessed psychological responses specific to pain including fear of pain (Fear of Pain Questionnaires, FPQ, [48]) and pain catastrophizing (Pain Catastrophizing Scale, PCS, [66]). Psychological questionnaires information and abbreviations are provided in Suppl Materials Table S1.

To reduce the number of constructs of the survey items, we conducted principal component analysis (PCA [6]) to capture the major psychological characteristics of the study participants.

Data analysis strategy

Missing data imputation

Five out of 402 TMD participants and four out of 401 healthy participants were excluded because greater than 20% of the psychological questionnaire data from those seven participants was missing, resulting in 397 TMD participants and 397 HC for data analysis. In the 397 TMD and 397 HC cohorts, the missing data of psychological questionnaires was minor and randomly distributed (TMD: ranges: 0% to 0.6%, Little’s MCAR test [43] χ²=75.29, p=0.180; HC: ranges: 0% to 0.5%, Little’s MCAR test χ²=86.67, p=0.100). Therefore, we imputed the missing data using the mean of the subscales to retain the cases for further analysis [65].

Principal component analysis on psychological questionnaire subscales

A correlation matrix of the psychological questionnaire subscales indicated that the TMD and HC shared similar underlying psychological constructs for survey subscales (Fig. 2a–b). Thus, we performed PCA in a combined dataset of TMD and HC to compare the psychological profiles between TMD and HC participants controlling for the socio-demographic variables (i.e., age and sex). Varimax rotation was employed for PCA [1; 6]. Elbow scree figure and eigenvalues were used to determine the number of components [1] (Supplementary Materials, Fig. S2). Components with an eigenvalue over than 2 were retained for further analyses. The normalized sum of the identified factors was calculated to represent the level of each component. After determining the psychological constructs of the included survey subscales, univariate analysis of covariance (ANCOVA) was conducted to compare the psychological factors profiles between TMD and HC participants controlling for the socio-demographic variables (i.e., age and sex). Furthermore, multiple linear regressions were performed to explore the associations between psychological profiles and TMD related chronic pain severity and pain interferences, separately.

Fig. 2.

Correlation matrix of 33 psychological questionnaires subscales in TMD (a) and HC (b) participants. TMD and HC participants displayed similar psychological constructs underlying the psychological survey subscales. The color red indicated positive correlations, while the color blue indicated negative correlations. Details of the psychological questionnaires descriptions are included in the supplementary materials Table S1. Plac=Placebo hypoalgesia; CS=Conditioning Strength; RE=Reinforced expectations; MASQ_gdd=Mood and Anxiety Symptom questionnaire general distress depression subscale; MASQ_gda= Mood and Anxiety Symptom questionnaire general distress affect subscale; MASQ_aa= Mood and Anxiety Symptom questionnaire anxious arousal subscale; MASQ_adt= Mood and Anxiety Symptom questionnaire anhedonia depression subscale; PANAS_neg= The Positive and Negative Affect Schedule negative affect subscale; BDI=Beck Depression Inventory; STAI=State-Trait Anxiety Inventory; DASS_dep= Depression Anxiety Stress scales depression subscale; DASS_str= Depression Anxiety Stress scales stress subscale; DASS_anx= Depression Anxiety Stress scales anxiety subscale; LOT_pes= Life Orientation Test – Revised pessimism subscale; NEO_n= NEO Five-Factor Inventory neuroticism; MDAQ=Mood/Depression assessment questionnaire; BisBas_b= Behavioral avoidance/inhibition scales behavioral inhibition subscale; BisBas_f= Behavioral avoidance/inhibition scales fun seeking subscale; BisBas_d= Behavioral avoidance/inhibition scales drive subscale; BisBas_r= Behavioral avoidance/inhibition scales reward responsiveness subscale; NEO_e= NEO Five-Factor Inventory extraversion subscale; LOT_opt= Life Orientation Test – Revised optimism subscale FPQ_medical=Fear of Pain Questionnaire medical pain subscale; FPQ_minor=Fear of Pain Questionnaire minor pain subscale; FPQ_severe= Fear of Pain Questionnaire severe pain subscale; PCS_r=Pain Catastrophizing Scale rumination subscale; PCS_m= Pain Catastrophizing Scale magnification subscale; PCS_h= Pain Catastrophizing Scale helplessness subscale; IRI_ec= Interpersonal Reactivity Index empathetic concern subscale; IRI_fs= Interpersonal Reactivity Index fantasy subscale; IRI_pt= Interpersonal Reactivity Index perspective taking subscale; IRI_pd= Interpersonal Reactivity Index personal distress subscale; NEO_O= NEO Five-Factor Inventory openness to experience subscale; NEO_A= NEO Five-Factor Inventory agreeableness subscale; NEO_C= NEO Five-Factor Inventory conscientiousness subscale

Determining dynamics of conditioning, placebo hypoalgesia, and expectations

We operationalized conditioning as the average pain ratings differences between red and green trials during the conditioning phase, and the magnitude of placebo hypoalgesia as the average difference scores from red trials minus green trials during the testing phase. Larger differences scores during the conditioning phase and testing phase represented greater conditioning and placebo hypoalgesia, respectively. We determined the occurrence of conditioning and placebo hypoalgesia by conducting repeated measure ANCOVAs on the VAS pain intensity ratings with color of the scree set as a within-subjects factor. Age and sex were set as covariates. Given that the temperatures were tailored based on individual pain tolerance limit, the conditioning and placebo patterns may be influenced by the distinct temperature differences between individuals. To rule out the potential confounding effects of psychophysical painful stimuli in influencing the conditioning/acquisition pattern, we treated the levels of temperatures used for red and green screens as a covariate in the repeated measures ANCOVAs comparisons.

Additionally, we obtained the patterns of acquisition (conditioning phase) and extinction (testing phase) for each participant by calculating the slope of the 12 difference scores during the conditioning phase and the 6 difference scores during the testing phase. The slopes were calculated using regression coefficient analysis methods [58]. A positive regression coefficient indicated an increasing pattern, while a negative coefficient indicated a decreasing pattern in red-minus-green difference scores.

Repeated measure ANCOVA analysis was conducted with the three assessments of expectations and perceived effectiveness (baseline, post-conditioning, post-testing phase) as repeated variables and TMD vs. HC participants set as between-subjects factor. Age and sex were treated as covariates.

Psychological factors influence conditioning, placebo hypoalgesia, and expectations

We built moderated multiple regression models [73] with psychological factors identified from PCA (i.e., emotional distress, reward seeking, pain related fear and catastrophizing, empathy and openness) treated as continuous predictors [73]. Group (coded as HC=−1, TMD=1) was treated as a categorical predictor. Interaction terms between each of the psychological factors and group were built in the regression models in order to determine if group would moderate the relationship between psychological factors and conditioning, placebo hypoalgesia, expectations, or perceived effectiveness. All psychological factors were person-centered. The moderated linear regressions were performed using the “lm” function in R studio i386.3.5.2 (R Studio, Inc., Boston, MA, USA). The standardized slope b was used as an index for effect size of regression following previous studies’ recommendations [2; 24]. The effect size slope b was calculated using G*power software vers. 3.1 [23].

We examined the relationship between expectations of pain relief and the magnitude as well as the extinction rate of the placebo effects. To achieve it, separated moderated linear regressions were conducted with magnitude and extinction rate of placebo effects as dependent variables. Baseline expectations and reinforced expectations were set as independent variables, respectively. The interactions between expectations and group (TMD vs. HC) were also modeled in the moderated linear regressions to determine if the expectation-placebo associations would be different between TMD and HC participants.

Identifying placebo responders

Finally, we identified placebo-responsiveness status via permutation tests as previously done [12; 68] on the 6 red and 6 green repeated measures of pain intensity ratings during the testing phase. For the permutation test, we first calculated the observed difference T_obs as the average pain ratings differences of the 6 red-minus-green trials. Next, we resampled 1000 times from the pooled pain ratings from red and green trials to create a new distribution of possible difference scores under the null hypothesis that pain ratings for red and green were exchangeable (i.e., came from the same distribution). The p value was calculated based on the observed difference T_obs and the new resampled distribution. Specifically, wherever the T_obs was contained within the 95% of the resampled possible differences scores, the null hypothesis was not rejected and permutation test was considered as non-significant. If the absolute value of T_obs was greater or equals to the 95% of the resampled possible difference scores, the null hypothesis was rejected and the permutation test was significant, indicating that pain ratings from red trials were different from pain ratings of green trials. The significant level was set as p=0.05. An individual was classified as placebo responder when the observed difference was positive (pain ratings of red > pain ratings of green) and the permutation test was significant.

A similar moderated logistic regression was conducted to determine the influences of psychological factors on the likelihood of placebo responders. Psychological factors, group and their interactions were modeled as predictors. The moderated logistic regressions were performed using the “glm” function in R studio i386.3.5.2 (R Studio, Inc., Boston, MA, USA). For all the analyses, the significance level was set as p<0.05.

Power Calculation

According to the parent on placebo effects [10], we expected to observe moderate to large effect size for placebo hypoalgesia (Cohen’s d=0.5) induced in TMD and HC participants. The power analysis indicated that a minimum N of 129 participants for each group would be needed to achieve 0.8 statistical power at the alpha level of 0.05.

In terms of the optimal sample size for principal components analysis, studies have posited that better outcomes came with larger N and greater ratios of participants to psychological questionnaires [13; 54]. A minimum ratio of 10:1 of participants to psychological questionnaire would be sufficient to obtain adequate statistical power for PCA [53], resulting in a minimal sample size of 330 participants for the current study.

Given the inconsistency of the previous studies in determining psychological predictors of placebo hypoalgesia, we expect to have a small effect size of psychological factors predictions and its interactions with group on placebo effects (Cohen’s f²=0.02). A minimum N of 759 participants would be needed to obtain 0.8 statistical power at the alpha level of 0.05. The power calculation was conducted using G*Power software [25]. Therefore, the current sample size of 794 participants provided sufficient statistical power for the PCA and the moderated linear regressions mentioned above.

Results

The TMD cohort was older and had a larger number of women compared to the HC participant cohort (Table 1). Given that age and sex were linked with placebo hypoalgesia (age: β=−0.11, p=0.005; sex: β=0.07, p=0.050) in the current study, these socio-demographic variables were treated as covariates in further analyses. Social demographic and clinical information of 397 TMD and 397 HC participants are presented in Table 1.

Table 1.

Demographic, social economic and clinical characteristics of TMD and HC cohorts.

	TMD (n=397)		HC (n=397)		t/χ2
	Mean/N	SEM/%	Mean/N	SEM/%
Age (mean±sem)	41.38	0.71	29.37	0.52	t=13.61, p<0.001
Race a (N/%)					χ 2 =50.57, p<0.001
Native American	1	0.3%	2	0.5%
Asian	29	7.3%	98	24.7%
African-American/Black	138	34.8%	89	22.4%
White	202	50.9%	190	47.9%
Mixed race	27	6.8%	18	4.5%
Sex b (N/%)	304	76.6%	237	59.7%	χ 2 =26.04, p<0.001
Educational level c (N/%)	106	26.7%	152	38.3%	χ 2 =12.15, p<0.001
Annual income d (N/%)	231	58.2%	230	57.9%	χ2=0.01, p=0.943
Blood pressure (mean±sem)
Systolic	131.05	3.43	119.77	0.67	t=3.39, p=0.001
Diastolic	80.13	0.50	73.89	0.50	t=8.85, p=0.001
Heart rate (mean±sem)	74.01	2.39	69.76	0.60	t=1.72, p=0.085
BMI (mean±sem)	28.40	0.36	25.21	0.26	t=7.15, p<0.001
Mood disorder
Depression (BDI) d					χ 2 =63.46, p<0.001
Minimal range [0–13]	278	70.1%	365	91.9%
Mild range [14–19]	50	12.6%	18	4.5%
Moderate range [20–28]	45	11.3%	8	2.0%
Severe range [29–63]	24	6.0%	6	1.6%
Anxiety (STAI-Trait) e					χ 2 =46.78, p<0.001
No or low anxiety [20–39]	204	51.4%	193	48.6%
Moderate or high anxiety [40–80]	297	74.8%	100	25.2%
Clinical Pain Characteristics
TMD Type (N/%)
Myalgia	377	95.0%	-		-
Myofascial pain with referral	35	8.8%	-		-
Right arthralgia	295	74.3%	-		-
Left arthralgia	286	72.0%	-		-
GCPS [0–4] (mean±sem)	2.03	0.06	-		-
Pain duration [in months] (mean±sem)	142.40	6.48	-		-
Pain medicines (N/%)					-
NSAIDs	236	59.4%	-		-
Muscle relaxants	10	2.5%	-		-
BDZ	15	3.8%	-		-
TCA	5	1.3%	-		-
SSRI	24	6.0%	-		-
SNRI	18	4.5%	-		-
SARI	8	2.0%	-		-
NDRI	11	2.8%	-		-
Oxycodone	19	4.8%	-		-
Vicodin	4	1.0%	-		-
Morphine	3	0.8%	-		-
Fentanyl	1	0.3%	-		-
Percocet	5	1.3%	-		-
Narco	2	0.5%	-		-
Hydrocodone	8	2.0%	-		-
Cocaine	3	0.8%	-		-
Cannabis	38	9.6%	-		-
Other pain disorders			-		-
Ehlers Danlos Syndrome	3	0.8%	-		-
Knee pain	47	11.8%	-		-
Shoulder pain	47	11.8%	-		-
Low back pain	136	34.3%	-		-
Osteoarthritis	80	20.2%	-		-
Fibromyalgia	20	5.0%	-		-

Significant t-tests and chi-square tests were marked in bold.

^aSex: the proportion of women was reported for each group.

^bEducational level: college graduation or higher vs. some college or lower. The proportion of college graduation or higher was reported for each group.

^cAnnual income: more than $60,000 vs. less than $60,000. The proportion of $60,000 or higher was reported for each group.

^dBDI=Beck Depression Inventory

^eSTAI-Trait = State and Trait Anxiety Inventory-Trait subscale

Abbreviations: GCPS=Grades of Chronic Pain Scale; NSAIDs=Non-steroidal Anti-inflammatory Drugs; BDZ=Benzodiazepine; TCA=Tricyclic antidepressants; SSRI=Selective serotonin reuptake inhibitors; SNRI=Selective serotonin-norepinephrine reuptake inhibitor; SARI=Serotonin antagonist and reuptake inhibitors; NDRI= Norepinephrine–dopamine reuptake inhibitor

Conditioning response, placebo hypoalgesia, and expectations

During the conditioning phase, both TMD and HC participants rated green trials (mean=9.71, sem=0.33) significantly less painful than the red trials (mean=70.30, sem=0.53, main effect of the color: F_1,787=128.97, p<0.001) controlling for age, sex, and individual pain sensitivity. Noted that due to pain perception habituation, the observed average pain ratings for both green and red screens were lower than the pre-conditioning phase where 80 out of 100 pain ratings were paired with the red screen, and 20 out of 100 pain ratings were paired with the green screen. Conditioning patterns were significantly stronger in HCs (mean=62.62, sem=0.88) as compared to TMDs (mean=58.57, sem=0.88, F_1,787=9.42, p=0.002, Fig. S3a). We further compared the slope of 12-trial conditioning between TMD and HC participants, and there were no significant group differences in the slopes (TMD: mean slope=0.33, sem=0.02; HC: mean slope=0.31, sem=0.02; F_1,787=0.11, p=0.745).

During the testing phase, pain ratings for green trials (mean=30.71, sem=0.68) were significantly lower than in red trials (mean=50.18, sem=0.74) regardless of sex, age, or the heat pain sensitivity (F_1,789=6.14, p=0.013), suggesting occurrence of significant placebo hypoalgesia in both TMDs and HCs. Moreover, TMD and HC participants exhibited a similar magnitude of placebo effects (F_1,789=0.003, p=0.957, Fig. S3b). We then compared the extinction rate by calculating the slope of the 6-trial placebo hypoalgesia and found that in TMD the slope was flatter (mean slope=−0.26, sem=0.03) than in HC participants (mean slope=−0.35, sem=0.03, F_1,789=5.01, p=0.025), indicating placebo effects extinguished slower in TMD as compared to HC participants.

We further examined the expectations in TMDs and HCs measured at the baseline, post-conditioning, and perceived effectiveness post-testing phases. We only observed a significant group by time interaction (F_2,1580=4.12 p=0.027, Greenhouse-Geisser corrected) with TMDs exhibiting a higher level of post-testing perceived effectiveness than HCs did (p=0.031, Fig. S3c).

Psychological profiles of TMD and HC participants

The Principal component analysis (PCA) resulted in four orthogonal components (see Supplementary Materials Fig. S2), including i) emotional distress, ii) reward seeking, iii) empathy and openness, and iv) pain related fear and catastrophizing (Fig. 3a), corresponding to the RDoC framework negative valence systems (emotional distress), positive valence systems (reward seeking), and social process systems (empathy and openness). The fourth component fear and catastrophizing reflected pain-specific cognitive and emotional characteristics. These four components explained 54.58% of the total variances of the psychological subscales (Bartlett’s test of Sphericity KMO=0.910, p<0.001).

Fig. 3.

(a) Principal Component analysis resulted in 4 factors: emotional distress, reward seeking, empathy and openness, as well as pain related catastrophizing and fear. (b) TMD participants had greater emotional distress, reward seeking, and pain related fear and catastrophizing but no differences in empathy and openness when compared to HC participants emotional distress (ED), reward seeking (RS), pain related catastrophizing and fear (PC).

The dominant component was emotional distress which reflected the activation of negative valence system and assessments of anxiety, depression, stress, pessimistic life orientation, and personality neuroticism. The second component, reward seeking, reflected the activation of positive valence system by including measurements on sensitivity toward rewarding events (as measured by the Behavioral Avoidance/Inhibition Scales (BISBAS)[7]. Notably, the component of reward seeking also included reverse factor loadings for positive affect, extroversion, and optimism subscales (see Positive and Negative Affect Scale, PANAS; Neuroticism, Extraversion, Openness Five-Factor Inventory, NEO-FFI; and Life Orientation Test, LOT-R scales details in Suppl. Materials), indicating an individual’s motive for moving toward something desired was, to some extent, paralleled by a less positive emotional state and introversive personality in the current cohort of TMD and HC participants. The third component, empathy and openness, included measurements on the individual’s ability to understand other’s emotion/behavior (Interpersonal Reactivity Index, IRI[19] and openness to novel experiences (NEO openness [15]) reflecting an activation of the social processes system, which is in line with the RDoC framework [17; 33; 61]. The last component, pain related fear and catastrophizing, measured a cognitive and emotional state that is specific to pain (FPQ, [48] and PCS [66]). The sum of the identified subscales was normalized to Z scores to represent the levels of emotional distress, reward seeking, empathy and openness, as well as pain related fear and catastrophizing, respectively.

Compared to HCs, TMD participants exhibited greater general emotional distress (F_1,790=108.45, p<0.001), higher reward seeking (F_1,790=45.13, p<0.001), and greater pain related fear and catastrophizing (F_1,790=11.47, p=0.001). However, TMD and HC participants did not differ in empathy and openness (F_1,790=0.41, p=0.522, Fig. 3b).

When comparing the heat pain threshold and heat pain tolerance limit between TMD and HC participants, we found that TMD participants were more sensitive to heat pain stimulations as indicated by significantly lower heat pain threshold (mean=36.94°C, sem=0.16, independent t=2.91, p=0.004) and lower heat pain tolerance limit (mean=48.42°C, sem=0.11, independent t=2.93, p=0.003) than HC participants (pain threshold: mean=37.63°C, sem=0.18; pain tolerance limit: mean=48.87°C, sem=0.11).

Across TMD and HC participants, emotional distress was not associated with heat pain threshold (r=−0.02, p=0.654) but was significantly associated with reduced heat pain tolerance limits (r=−0.08, p=0.026) On contrary, greater reward seeking was related to lower heat pain threshold (r=−0.09, p=0.015) but not significantly associated with heat pain tolerance limit (r=−0.04, p=0.216). Higher empathy and openness were significantly associated with both reduced heat pain threshold (r=−0.14, p<0.001) and lower heat pain tolerance limit (r=−0.18, p<0.001).

The association between psychological factors and TMD clinical pain intensity and interference were examined and reported in the Supplementary Materials.

Psychological determinants of conditioning and acquisition pattern

Higher emotional distress (β=−0.24, p<0.001) was a significant predictor of reduced conditioning across TMD and HC participants. Reward seeking, fear of pain and pain catastrophizing, empathy and openness did not interact with group and did not influence the conditioning (all p>0.115). Emotional distress (β=−0.09, p=0.011) also emerged as a significant predictor of a slower acquisition rate across TMD and HC participants.

Psychological determinants of placebo hypoalgesia and extinction pattern

Across TMD and HC groups, emotional distress (β=−0.07, p=0.041) and pain related fear and catastrophizing (β=−0.11, p=0.006) emerged as significant predictors of a smaller magnitude of placebo hypoalgesia (Fig. 4a). Importantly, a greater level of emotional distress predicted slower extinction rate (β=0.51, p<0.001) in both TMD and HC participants.

Fig. 4.

(a) Moderated regression results for conditioning, placebo hypoalgesia, and expectations. EM=Emotional distress; RS=Reward seeking; FC=Pain related fear and catastrophizing; EO= Empathy and openness; A red color indicated positive relationships; A blue color indicated negative relationships; A gray color indicated non-significant results. For results related to placebo responsiveness, the odds ratio (OR) from the moderated logistic regression was reported. Significant odds ratio was marked in blue. Emotional distress and pain related fear/catastrophizing were significant predictors of placebo non-responders. *p<0.05; **p<0.01; ***p<0.001. (b) Individual placebo effects and time-course of placebo hypoalgesia in placebo responders and non-responders within TMD and HC participants. Individual placebo responsiveness was identified using permutation tests via resampling 1000 times on the 6 red and 6 green pain ratings (cut-off p=0.05). An individual was identified as a placebo responder when the ratings for the green trials were significantly lower than the ratings for the red trials based on the approximate test statics distribution. TMD participants were marked as circle, while HC participants were marked as triangle. Placebo responders were marked as blue, and placebo non-responders were marked as red. Time-course data are displayed in mean and SD for each trial.

Psychological determinants of expectations

For baseline expectations, a higher level of pain related fear and catastrophizing was a significant predictor of lower baseline expectation about the effectiveness of the sham intervention (β=−0.13, p<0.001). In contrast, empathy and openness was a predictor of higher baseline expectations (β=0.07, p=0.047).

For reinforced expectations after the conditioning phase, similarly, higher pain related fear and catastrophizing predicted reduced reinforced expectations (β=−0.09, p=0.021). Importantly, there was a significant interaction between levels of reward seeking and group (β=0.09, p=0.033). In particular, in TMD participants, higher reward seeking was linked to an increase in reinforced expectations (β=0.16, p=0.004), while no such relationship was found in the HC cohort (β=−0.03, p=0.608).

Similar to baseline expectation results, a lower level of pain related fear/catastrophizing (β=−0.11, p=0.008) and a higher level of empathy and openness (β=0.08, p=0.040) were also significant predictors of greater post-testing perceived effectiveness.

Placebo responsiveness

As previously done by Apkarian’s and our teams [12; 68], we classified participants as placebo responders using permutation tests (cut-off set as p=0.050). 214 out of 397 TMD patients (53.9%) were placebo responders, which was significantly lower than the placebo responding percentages from the healthy controls, where 269 of 397 participants (67.8%) were placebo responders (chi-square=15.99, p<0.001, Fig. 4b).

Across TMD cases and healthy controls, individuals with lower level of emotional distress (OR=0.76, p=0.003; 95%CI=[0.66, 0.88]) and lower pain related fear and catastrophizing (OR=0.79, p=0.005; 95%CI=[0.67, 0.93]) were more likely to be placebo responders.

Expectations and placebo effects

Higher levels of expectations at baseline (β=0.08, p=0.030) and post-conditioning reinforced expectations (β=0.18, p<0.001) were significantly associated with larger magnitude of placebo effects. These results held true for both TMD and HC participants as revealed by the non-significant interactions between expectations and group (all p>0.220). In terms of placebo extinction rate, neither baseline expectations (β=−0.05, p=0.186) nor reinforced expectations (β=0.06, p=0.091) were linked to the placebo extinction rate in both TMD and HC participants.

Because pain related fear and catastrophizing were significantly related to lower reinforced expectations and also reduced placebo hypoalgesia in both TMD and HC participants, mediation analysis were further conducted to determine if reinforced expectations mediated the relationship between pain related fear and catastrophizing and placebo hypoalgesia following the recommendations from Baron and Kenny [4]. For the mediation analysis, pain related fear and catastrophizing was treated as the independent variable (X). The magnitude of placebo hypoalgesia was treated as the dependent variable (Y). The level of reinforced expectation was treated as the mediator (M). 5000 times bootstrapping method was used to determine the significance of the mediation model. When the bootstrapped 95% confidence interval (BCI) did not contain 0, the mediating effect was considered to be significant. Mediation analysis was conducted using SPSS vers 27 PROCESS macron [30].”

Mediation analysis indicated that reinforced expectation was not an significant mediator driving the placebo hypoalgesia in those with a certain level pain related fear and catastrophizing (a*b=−0.17, BCI=[−0.416, 0.033]). The direct effect from pain related fear and catastrophizing to placebo hypoalgesia was significant (c’=−1.26, p=0.03), suggesting that higher level of fear and catastrophe of pain were associated with smaller placebo effects regardless their level of reinforced expectations.”

Discussion/Conclusion

This is one of the largest cohort studies to date that has established the determinants of placebo effects in chronic pain and healthy populations. We found that negative valence systems, which are primarily responsible for negative emotional states, were paralleled by reduced placebo effects. Pain related fear and catastrophizing were also linked to reduced placebo hypoalgesia in both TMD and HC participants. Placebo responders were characterized by lower emotional distress, pain related fear and catastrophizing.

In an attempt to identify psychological characteristics that contributed to the variability in placebo hypoalgesia, previous studies have separately focused on various aspects of psychological factors ranging from stable personality factors (e.g., neuroticism, openness), to emotional disorders (e.g., anxiety, depression) to cognitive-motivational factors (e.g., reward sensitivity). One of the challenges in examining placebo related determinants is that a single measurement may not be sufficient in capturing the full nature of distinct psychological processes, while multiple measurements of psychological factors are usually highly correlated and interact with one another. To address this challenge, an RDoC framework [33; 61] and principal component analysis was conducted in this study, allowing us to reduce the number of the included psychological factors and to avoid the multicollinearity of the multiple psychological measurements.

Importantly, TMD and HC participants were characterized by distinct profiles of psychological features that contributed to the variations as expected in pain phenotypes as well as expectations, conditionings, and placebo effects.

When we used the RDoC framework [33; 61], we observed that compared to healthy controls, chronic pain participants were characterized by greater emotional distress, higher pain related fear and catastrophizing, and were more sensitive to reward seeking (e.g. the reward of pain relief). Moreover, in TMD participants, greater emotional distress and pain related fear/catastrophizing were linked to higher chronic pain interference. These findings are in line with previous studies emphasizing the contributions of negative emotional states such as anxiety [41], depression [20], and fear of pain [16] to facilitating pain interference and maintaining chronic pain status.

Expectations are one of the key mechanisms that underlie placebo hypoalgesic effects [36]. Regardless of chronic pain status, individuals with higher fear of pain were characterized with lower expectations of pain relief, which in turn, might mitigate placebo hypoalgesia. In terms of social processing systems, charge of empathy and/or interpersonal activity was a significant predictor of greater levels of baseline pain relief expectations and post-testing perceived effectiveness; However, those factors were not directly linked to conditioning or placebo hypoalgesia. The neurophysiological basis of openness to experiences involves stronger resting-state activities in areas of the precuneus and bilateral inferior parietal lobes [71]. In fact, reduced deactivations of BOLD signals in the precuneus and bilateral parietal lobes have been linked to greater pain related brain responses [38].

Our findings also highlighted the role of emotional distress in facilitating the chronic pain disability and impairing placebo hypoalgesia, echoing previous findings on emotion regulation - an individual’s ability to control how to experience and how to express emotions - and chronic pain [37]. A recent systematic review on emotion regulation demonstrated maladaptive emotion regulations such as expressing or inhibiting anger or fear robustly contributed to the development and maintenance of chronic pain [37]. Moreover, emotion regulation has also been proposed as one of the potential mechanism of placebo effect via triggering a network of cognitive, affective and behavioral re-engagement [9]. Negative emotions such as anxiety or depression have been linked to disrupted endogenous opioid system [56], which is one of the key mechanisms in the formation of placebo hypoalgesia [55]. Emotional distress was also a predictor of a slower acquisition rate and a slower extinction rate of placebo hypoalgesia. This finding is in line with emerging evidence on the relationship between emotional distress and deficits in reinforcement reward learning (see review [8]). When compared to controls, participants with depression were slower to disengage from negative stimulation, suggesting that an impaired ability to update information about negative events was linked to emotional distress [42]. Emotional distress facilitates chronic pain and may also impair the endogenous descending pain modulation. Thus, mood disorder such as anxiety and depression, and pain related characteristics such as fear of pain and pain catastrophizing should be taken into consideration when providing treatment for chronic pain patients and considering placebo effects in clinical practice.

Across TMD patients and healthy participants, we observed that greater pain related fear/catastrophizing was linked to reduced placebo effects, echoing previous studies where healthy participants with higher fear of pain showed reduced placebo hypoalgesic effects [44] and smaller pain elicited event-related potential P2 amplitudes [45]. The current study also revealed that greater baseline expectations and reinforced expectations were related to larger placebo effects regardless of the group (TMD vs HC participants). Moreover, reinforced expectations did not mediate the influences of psychological characteristics on placebo hypoalgesia, suggesting that learning may boost placebo effects bypassing pre-conditioning cognitive aspects [11].”

Interestingly, during the testing phase, both HCs and TMD showed placebo effects of similar magnitude, although placebo responders percentages in TMD and HC participants differed (53.9% versus 67.8%, respectively). Clinical studies examining placebo responses demonstrate response rates ranging from around 34% [22] for clinical depression to 78% [26] in chronic low back pain. In experimental placebo studies, response rates for placebo range from around 38% [60] to 60% [10] depending on the placebo manipulation. Our findings that TMD participants having similar placebo hypoalgesia compared to HC participants was in line with a meta-analysis comparing placebo analgesia between patients and healthy individual where patients showed slightly greater placebo analgesia than healthy participants [27].

Can psychological determinants be used to predict placebo responsiveness? In fact, our RDoC approach indicated that placebo responders have lower emotional distress, lower pain related fear and catastrophizing, further enriching the novelty of this study findings. In a recent study, Vachon-Presseau et al. [68] conducted correlations among several psychological factors and placebo response, and demonstrated a positive link between openness to experiences and placebo effects in a cohort of 63 chronic low back pain participants who took placebo pills with a verbal suggestion. We did not observe such a relationship when adopting classical conditioning to induce placebo hypoalgesia. The inconsistency of the results may reflect potential interactions between personality factors and distinct placebo manipulation approaches (e.g. verbal suggestions alone versus prior therapeutic exposure) and power issues.

Although there are many strengths in this current study, there are also limitations to consider. First, although we controlled for sex and age, personality factors may interact with socio-demographic variables in influencing placebo hypoalgesia. Future studies are needed to examine the potential complex interactions and joint contributions of socio-economic-status and psychological factors to placebo hypoalgesia. Second, as a cross-sectional study, we failed to address the causal effects of the psychological factors in producing the placebo effects. Although personality factors were collected at the same time of placebo manipulation, a growing amount of literature has pointed out that even stable personality factors can shift over time [47]. It should be noted that the placebo responsiveness in the current study was assessed in a single session via a single placebo manipulation. There is a long and deep literature on identifying placebo responders, and the findings indicated that the reliability and reproducibility of placebo responders remain inconclusive (systematic review, [34]). Individuals may not consistently respond to placebo [40; 75] and may respond differently to different placebo manipulations [39]. With the cross-sectional nature of the study design, the current findings only provided a snapshot of individual differences in psychological determinants of placebo responders vs. non-responders when a conditioning with verbal suggestion paradigm is used. Therefore, caution would be needed when generalizing the current findings to other paradigms of placebo procedures such as verbal suggestion alone, social observation learning, open-label placebos and contextual factors. Moreover, future longitudinal studies would be needed in exploring stable psychological determinants for placebo effects over time and under distinct placebo procedures.

In conclusion, although TMD participants had significantly greater levels of emotional distress and higher pain related fear and catastrophizing levels, their placebo effects were comparable to those of healthy participants. Moreover, TMD participants were characterized by greater reward seeking that was linked to an increase of reinforced expectations after the conditioning phase. In contrast, reward seeking was not linked to expectations in the healthy controls. Most importantly, the current findings highlighted the role of emotional distress as well as pain related fear and catastrophizing in reducing the magnitude of placebo effects in both TMD and healthy participants. Additionally, placebo responders had lower levels of emotional distress, pain related fear, and catastrophizing across TMD and HC participants. These results imply that individuals reporting emotional distress and maladaptive cognitive appraisals of pain may benefit less from placebo effects.