Underpredicting Pain: An Experimental Investigation Into the Benefits and Risks

Kaya J. Peerdeman; Andrew L. Geers; Delia Della Porta; Dieuwke S. Veldhuijzen; Irving Kirsch


Pain. 2021;162(7):2024-2035. 

In This Article

Study 1

The primary aim of study 1 was to assess whether strong underprediction of pain could increase experienced pain intensity, that is, evoke a contrast effect. Participants were randomly assigned to one of the 2 experimental conditions that differed only in the verbal suggestions provided: (1) Strong underprediction condition, in which suggestions that 2 series of 4 presentations of a heat stimulus would all be nonpainful were given, and (2) Correct prediction condition, with suggestions that the stimulus would be moderately painful. The primary outcome was pain intensity experienced during the first postsuggestion trial given after the condition-specific suggestion to examine initial responses. The secondary outcome was expected pain intensity upon undergoing the expectation–experience (mis)match. All additional outcomes were exploratory.

Methods Study 1

Participants. In study 1, 82 healthy adults were included. This sample size was deemed sufficient to observe a small to medium effects size with the planned primary analyses, that is, an analysis of covariance (ANCOVA) analyzing the effect of the verbal suggestions on experienced pain intensity during the first postsuggestion trial, controlled for average experienced pain intensity during the 3 high reference trials. First, the required sample size for an independent t test was determined in G*power (2-sided, d = 0.35, α = 0.05, power = 0.80), which was then corrected to account for a covariate that correlates strongly with the outcome (r = 0.85,[47]) with Borm and Fransen's formula.[9] An error margin of 10% was added, resulting in a target total sample size of 82. Eligibility criteria were: age 18 to 30 years, good understanding of written and spoken English, no severe physical or psychological morbidity (eg, heart and lung diseases, or psychiatric disorders) that could adversely affect study participation, no chronic (≥6 months) past or present pain complaints, no or low current pain (≤3 on the 0–10 numerical rating scale [NRS]), no current use of any medication, no pacemaker, and no pregnancy. In addition, participants were instructed to not use any medication, alcohol, or other drugs in the 24 hours before the laboratory session.

Procedure. Data collection took place in a standard psychophysiology laboratory at the Faculty of Social and Behavioural Sciences, Leiden University, Leiden, the Netherlands from March to September 2017. The study protocol was approved by Leiden University's Psychology Research Ethics Committee (CEP17-0314_152) and preregistered at the Netherlands Trial Register (NL6228).

Participants were recruited through the university's online participant recruitment system, social media, flyers around the university, and personal communication. Potential participants were informed that the purpose of the study was to investigate pain sensitivity during heat stimuli. On registering, participants filled in a screening questionnaire and additional questionnaires (assessing demographics, optimism, general trust, pain catastrophizing, and socially desirable responding) through an online system (Qualtrics, Provo, UT; approx. 10 minutes) for which they separately provided digital informed consent.

Eligible participants were invited to make an appointment for the laboratory session. To maximize standardization and blinding, 2 female experimenters (trained bachelor and master psychology students) performed the entire procedure according to a detailed protocol. Experimenter A, who supervised the outcome assessments, remained blinded throughout the whole procedure; experimenter B was blinded until she gave the condition-specific verbal suggestion, during which experimenter A briefly left the laboratory with an excuse. Participants were not informed about the different experimental conditions until debriefing.

Once in the laboratory, after an explanation of the procedures and a brief check of the eligibility criteria, participants provided written informed consent. Subsequently, electrodes for the autonomic measurements were attached, upon which participants filled out computerized questionnaires (assessing trait and state anxiety and fatigue). Five to 10 minutes after electrode attachment, a 3-minute resting measurement of heart rate and skin conductance was taken. Skin conductance and heart rate were recorded continuously with epochs of interest marked.

Subsequently, warmth and pain perception thresholds were determined to familiarize participants with the heat stimuli. Next, a step-up heat pain calibration procedure was used to determine the temperature at which participants experienced moderately high pain. Participants received 3 presentations of a nonpainful heat stimulus before the step-up procedure and 3 presentations of a moderately painful heat stimulus afterwards, which served as reference points for the test.

In the subsequent manipulation and test phase, participants received either a suggestion of no pain (Strong underprediction condition) or a suggestion of moderately high pain (Correct prediction condition) depending on the condition to which they were randomized (using a randomization sequence generated by an independent researcher with an online random number generator, stratified by sex, 1:1 ratio, blocks of 4 and 6, and concealed by using sequentially numbered, opaque, sealed envelopes). All participants then received the moderately painful heat stimulus 4 times (after the thermode had been placed on a novel spot on the arm to avoid sensitization). Immediately after, the condition-specific verbal suggestion was repeated, and 4 additional trials were given. Before each trial, participants rated expected pain intensity, certainty of the expectation, and fear of the upcoming stimulus. Directly after each trial, participants rated experienced pain intensity and pain unpleasantness.

The experimental session was concluded with a second assessment of state anxiety, and several exit questions were filled in digitally, followed by an oral debriefing. All participants were reimbursed for their participation. The total duration of the test session was around 1.15 hours. See Figure 1 for an overview of the procedure.

Figure 1.

Overview of the procedure with all measures taken in study 1. NRS, numerical rating scale; VS, verbal suggestion.

Pain Evocation and Assessment. Warmth and painful sensations were evoked by applying heat stimuli to the volar surface of the nondominant forearm with a TSA-II NeuroSensory Analyzer (Medoc, Ramat Yishai, Israel) using a standard 30 × 30-mm Advanced Thermal Stimulation thermode and accompanying Medoc Main Station software (version

Warmth and Pain Thresholds: Following a practice trial, the warmth detection threshold was determined by administering 3 heat stimuli of ascending intensity until the participant indicated by a mouse click feeling warmth for the first time (1°C/second ramp up, 8°C/second ramp down from/to a baseline of 32°C). The interstimulus interval (ISI) was set at 10 seconds. The average of these 3 measurements was taken as the threshold in line with validated procedures.[53] The same procedure was used for the heat pain threshold detection when participants indicated feeling pain for the first time.

Heat Pain Calibration Step-up Procedure and Reference Stimuli: First, 3 presentations of a warm, nonpainful stimulus were given to serve as a reference for the strong underprediction suggestion (35°C; low reference stimulus). Subsequently, we used a step-up procedure in which stimuli of increasing temperatures were given (starting at 39°C, steps of 2°C until 43°C, and then steps of 0.5°C until max 49.5°C) until the participants reported moderately high pain. Moderately high pain was defined as a pain intensity rating between 6 and 7 on an NRS ranging from 0 (no pain at all) to 10 (worst pain imaginable). Another 3 trials of this moderately high pain intensity were given to serve as a reference for the postsuggestion trials (M = 47.4°C, SD = 1.9°C, high reference stimulus). To increase focus on the low and high reference stimuli, and thereby increase chances that participants became aware of the (mis)match, participants were asked whether the trials felt approximately the same. All stimuli were set to have a duration of approximately 8 seconds (4 seconds at peak, 8°C/second ramp up and down from/to 32°C baseline) and were separated by an ISI of 10 seconds. Medoc software issues with the detection of the moment at which peak temperature was reached caused the stimulus duration to be variable and often of longer duration than intended, with stimuli being at peak temperature for 4.0 to 10.6 seconds (M = 8.2, SD = 1.1 second). These variations were independent of experimental conditions. For autonomic analyses, post hoc corrections of epoch duration were done by hand to cover the actual peak duration.

Postsuggestion Trials: Participants received 2 series of 4 trials of the same moderately high intensity (M = 47.4°C, SD = 1.9°C) and of the same duration as the high reference stimulus, with an ISI of 30 seconds.

Pain Ratings: Participants rated experienced pain intensity directly after each trial on an NRS ranging from 0 (no pain at all) to 10 (worst pain imaginable) with the option to answer with 1 decimal precision. For the postsuggestion trials, participants additionally rated for each trial: (1) expected pain intensity, (2) certainty of pain expectation, (3) fear of upcoming pain stimulus, and (4) pain unpleasantness on 0 to 10 NRSs. The distinction between pain intensity and unpleasantness was explained according to previously established and frequently used instructions.[52] Participants provided their ratings verbally for the threshold, calibration, and reference stimuli. For the postsuggestion trials, participants wrote the ratings down to reduce possible response biases.

Verbal Suggestion Regarding Pain Intensity. In the Strong underprediction condition, the first 4 postsuggestion trials were suggested to be nonpainful by the experimenter using the following suggestion "You'll now get 4 low stimuli. These are the same as the low stimuli you felt before, at the beginning of the previous test. Based on the previous test, I expect that you will experience them as nonpainful. But please rate them as you feel them now."

In the Correct prediction condition, participants were correctly informed that the trials would be moderately painful using the following suggestion: "You'll now get 4 high stimuli. These are the same as the high stimuli you felt before, at the end of the previous test. Based on the previous test, I expect that you will experience them as moderately painful. But please rate them as you feel them now."

The condition-specific suggestion was repeated before giving the 4 additional trials: "You will now again get 4 low stimuli, that should be nonpainful" (Strong underprediction condition), or "You will now again get 4 high stimuli, that should be moderately painful" (Correct prediction condition). To enhance the chances that participants would notice the (mis)match between the instruction and the sensation, all participants were instructed to attend closely to their sensations during the tests and to compare those sensations to the reference stimuli.

Autonomic Measures. Electrocardiography (ECG) and skin conductance were measured with an MP150 system, including an ECG100C amplifier (1000 Hz, gain 1000, 35 Hz low-pass filter, 1 Hz high-pass filter), a GSR100C amplifier (1000 Hz, gain 5 μmho/V, 1 Hz low-pass filter), and STP100C and UIM100C modules, with AcqKnowledge software (version 5.0) (BIOPAC Systems Inc, Goleta, CA). For ECG recordings, disposable electrodes (Kendall 200 Foam Electrode, Covidien, Mansfield, MA) were placed on the sternum and the left lower rib, after abrading the skin. For skin conductance recordings, disposable Ag/AgCl electrodes (EL507-10, BIOPAC Systems Inc) were placed on the medial phalanges of the index and middle finger of the nondominant hand, after the skin was cleaned with water and dried. Epochs were marked from peak temperature onset to offset by manually sending triggers through E-prime 2.0 software (Psychology Software Tools, Inc, Sharpsburg, PA). For skin conductance level and response measures, a latency window of 1 second after peak temperature onset until 4 seconds after peak temperature offset was used.[12] In addition, based on visual inspection and taking rise time into account, post hoc sensitivity analyses were run for skin conductance response with an extended latency window of 0 seconds after temperature rise onset until 4 seconds after peak offset. Inspection of the ECG and skin conductance data, calculation of the average heart rate (bpm) and skin conductance levels (μS), and detection of skin conductance response (μS) during rest, reference stimuli, and postsuggestion trials were performed with PhysioData Toolbox (version 0.5.0).[56] Skin conductance response amplitudes during these epochs were determined using an adjusted version of EDA toolbox[30] in Matlab (R2018b).

Psychological Questionnaires. In addition to basic assessments of demographic characteristics, participants completed several questionnaires to assess possible moderators of the effects: (1) optimism (Life Orientation Test–Revised [LOT-R], Cronbach's α in this study = 0.75),[54] (2) general trust (Generalized Trust Scale, α = 0.82),[63] (3) pain catastrophizing (Pain Catastrophizing Scale [PCS], α = 0.90),[58] (4) ability to modify self-presentation and sensitivity to expressive behaviors of others (subscales of Self-Monitoring Scale–Revised [SMS-R], α = 0.78 and α = 0.70, respectively),[39] (5) trait anxiety (State-Trait Anxiety Inventory, Trait version [STAI-T], α = 0.89),[57] (6) state anxiety at baseline (State-Trait Anxiety Inventory, short State version [STAI-Ss], α = 0.76),[40] and (7) fatigue at baseline (visual analogue scale from 0 no fatigue at all to 10 worst fatigue imaginable).

Exit questions reassessed state anxiety (STAI-Ss, α = 0.76) and 0 to 10 visual analogue scales (with 1 decimal precision) were used to assess participants' previous experience with heat pain tests; pain intensity during previous heat pain tests; focus on sensations during the tests; confidence in their knowledge of the study purpose; observation of a discrepancy between actual pain intensity vs the low reference, high reference, expected, and instructed pain intensity; trustworthiness of the experimenters; and possible response bias (with the following 4 items: (1) concerns about what experimenter thought, (2) change of response to help experimenter, (3) sense of owing it to experimenter to report less or more pain, and (4) providing answers experimenter wanted to hear). Additional open-response questions asked participants to report their view of the purpose and expected results of the study, as well as any further comments they might have. The exit questions are available through https://osf.io/24mu5/.

Statistical Analyses

To test the primary hypothesis, an ANCOVA was used with condition (Underprediction vs Correct prediction) as independent variable (IV) and experienced pain intensity during the first postsuggestion trial as dependent variable (DV). Average experienced pain intensity during the 3 high reference trials was included as a covariate to reduce error variance of the effect estimate and thereby enhance statistical power and provide a more precise estimate of the effect size. In addition, we explored participants' pain during the trials using a mixed-model ANCOVA with condition as between-subject IV, trial as within-subject IV, experienced pain intensity during each of the 8 trials as DV, and average experienced pain intensity during the 3 high reference trials as covariate.

To test the secondary hypothesis, an analysis of variance (ANOVA) was run with condition as IV and expected pain intensity following the first postsuggestion trial as DV. In addition, an ANOVA with condition as IV and expected pain intensity before the first postsuggestion trial as DV was run as a manipulation check. We also explored participants' expectancies regarding all trials with a mixed-model ANOVA.

Certainty of expectations, fear of upcoming pain, and pain unpleasantness during all postsuggestion trials were explored with mixed-model ANOVAs with condition and trial as IVs and the relevant ratings for all 8 trials as DVs. We similarly explored effects on autonomic responses using ANCOVAs with condition and trial as IVs, and with heart rate, skin conductance level, and skin conductance response during all 8 trials as DVs, controlled for average responses of the relevant measure during the 3 high reference trials.

Effects on psychological outcomes assessed in the exit questionnaire were explored with one-way ANOVAs with condition as IV and the respective measures as DV, with the exception of effects on state anxiety for which an ANCOVA was used with premanipulation state anxiety as covariate.

The possible moderating role of individual characteristics was explored by examining the interaction between each individual characteristic and condition in separate multiple regression analyses with experienced pain intensity during the first postsuggestion trial as DV, while controlling for the main effects of the relevant individual measure and condition, as well as average experienced pain intensity during the 3 high reference trials.

All analyses were run in Rstudio (version 1.1.463; R version 3.5.3). Test results were reported with the StatTag plugin.[61] For readability, generally only statistically significant test results (α = 0.05) are reported. For AN(C)OVAs, generalized eta squared ( ) was computed, with 0.01, 0.06, and 0.14 signifying small, medium, and large effects, respectively.[37] For repeated measures analyses, a Greenhouse–Geisser correction was used when the sphericity assumption was violated. To evaluate the robustness of the results, the primary analysis (effect of condition on experienced pain intensity during the first postsuggestion trial) was also run: (1) with the stratification variable sex as an additional covariate, (2) with the time period during which stimuli were at peak temperature (stimulus peak duration) as an additional covariate, and (3) without any covariates.

Results Study 1

Data Availability. Table Supplemental digital content 1 (available at http://links.lww.com/PAIN/B269) presents the descriptives (mean ± SD) per condition of all reported measures. The data, analysis scripts, and results files are available through https://osf.io/24mu5/ (doi: 10.17605/OSF.IO/24MU5).

Participants. Of the 82 participants included, one participant dropped out after the first high reference trial (before randomization) because they could no longer handle the pain. For analyses, data of the 81 participants who completed participation were available (50.6% female; M age = 22.9, SD = 2.7; 50.6% Dutch nationality; median baseline pain = 0.0, interquartile range = 0.2).

Effects of Verbal Suggestions on Pain Intensity. Contrary to our primary hypothesis, the strong underprediction of pain led to significantly lower pain intensity than the correct prediction for the first postsuggestion trial, F(1, 78) = 19.19, P < 0.001, = 0.20 (Figure 2A; and Supplemental digital content 1, available at http://links.lww.com/PAIN/B269). Sensitivity analyses with the stratification variable sex or stimulus peak duration as an additional covariate and analyses without covariates yielded the same conclusions.

Figure 2.

(A) Experienced pain intensity ratings (M ± SE) upon the 3 high reference trials (average) and the 8 postsuggestion trials, and (B) expected pain intensity ratings (M ± SE) before each of the 8 postsuggestion trials, per condition, in study 1. ref, average of the 3 high reference trials; VS, verbal suggestion.

An exploration of pain intensity during all postsuggestion trials indicated a main effect of condition, F(1, 77) = 13.48, P < 0.001, = 0.10, with no significant main effect of trial, nor an interaction effect, demonstrating that the between-group differences persisted over time, also upon a repetition of the verbal suggestion.

Expected Pain Intensity. The strong underprediction led participants to expect significantly less intense pain for the first postsuggestion trial than the correct prediction, F(1, 79) = 82.22, P < 0.001, = 0.51, demonstrating that the verbal suggestions successfully manipulated expectations (Figure 2B). Contrary to our secondary hypothesis, participants in the Strong underprediction condition expected the trial following a violation of their initial expectation of no pain during the first postsuggestion trial to be less painful than participants in the Correct prediction condition, F(1, 78) = 15.72, P < 0.001, = 0.17.

An exploration of expected pain intensity during all trials indicated main effects of condition, F(1, 77) = 24.09, P < 0.001, = 0.17, and trial, F(4.09, 315.24) = 31.50, P < 0.001, = 0.12, and an interaction effect, F(4.09, 315.24) = 19.95, P < 0.001, = 0.08. In the Strong underprediction condition, participants' expectations of pain increased substantially upon experiencing the first postsuggestion trial, decreased upon repetition of the verbal suggestion, and then increased again upon experiencing the subsequent trial, whereas for participants in the Correct prediction condition their expectations remained stable across trials.

Other Pain Outcomes

Certainty of Pain Expectation. An exploration of certainty of pain expectations for all postsuggestion trials indicated main effects of condition, F(1, 76) = 4.08, P = 0.047, = 0.03, and trial, F(4.62, 351.02) = 3.95, P = 0.002, = 0.02, and an interaction effect, F(4.62, 351.02) = 8.06, P < 0.001, = 0.04 (Figure Supplemental digital content 2, available at http://links.lww.com/PAIN/B269). Participants in the Strong underprediction condition were less certain than participants in the Correct prediction condition following the first postsuggestion trial and repetition of the suggestion.

Fear of Upcoming Pain. An exploration of fear of upcoming pain for all postsuggestion trials indicated main effects of condition, F(1, 77) = 5.44, P = 0.022, = 0.06, and trial, F(4.32, 332.52) = 9.41, P < 0.001, = 0.02, reflecting continuously lower fear of pain in the Strong underprediction condition than in the Correct prediction condition (Figure Supplemental digital content 3, available at http://links.lww.com/PAIN/B269).

Pain Unpleasantness. An exploration of pain unpleasantness for all postsuggestion trials indicated a main effect of trial, F(3.97, 309.86) = 12.00, P < 0.001, = 0.02, reflecting reductions in pain unpleasantness after the first trial in both conditions (Figure Supplemental digital content 4, available at http://links.lww.com/PAIN/B269).

Autonomic Responses

There were no significantly differential effects of the strong underprediction vs correct prediction on heart rate, skin conductance level, or amplitude of the first skin conductance response (with regular or extended latency window) during the postsuggestion trials (Figure Supplemental digital contents 5 and 8, available at http://links.lww.com/PAIN/B269). There was a significant interaction effect for skin conductance level, F(3.28, 249.33) = 2.88, P = 0.032, = 0.01, but no clearly distinct patterns were observed. A main effect of trial was observed for the skin conductance response with an extended latency window, F(3.48, 264.72) = 5.16, P < 0.001, = 0.04, reflecting a decrease over time upon an initially strong response in both conditions.

Psychological Outcomes

Descriptives (Table Supplemental digital content 1, available at http://links.lww.com/PAIN/B269) and between-group differences in awareness of the (mis)match of actual vs expected stimulus intensity, F(1, 78) = 4.42, P = 0.039, = 0.05, and of actual vs suggested stimulus intensity, F(1, 78) = 31.34, P < 0.001, = 0.29, indicated that participants were aware of the expectation/instruction–experience (mis)match. Notably, participants rated the trustworthiness of the experimenter who provided the verbal suggestion to be significantly lower in the Strong underprediction condition than in the Correct prediction condition, F(1, 78) = 4.96, P = 0.029, = 0.06. No significantly differential effect of the suggestions on state anxiety was observed, nor were there significant between-group differences in the other measures (eg, observation of (mis)match of the reference and postsuggestion trials, trustworthiness of the other experimenter, and response bias measures).

Moderation by Psychological Characteristics

The effect of the suggestions on experienced pain intensity during the first postsuggestion trial was significantly moderated by state anxiety, b = −0.08, SE = 0.03, t = −2.60, P = 0.011, reflecting that participants who reported lower state anxiety showed larger pain reductions upon the strong underprediction than those who reported higher state anxiety (pain intensity Strong underprediction condition at lower vs upper quartile STAI-Ss = 5.5 vs 6.4). None of the other psychological characteristics (ie, optimism, general trust, pain catastrophizing, ability to modify behavior, sensitivity to others, trait anxiety, or baseline fatigue) significantly moderated the effects.