Do pain-associated contexts increase pain sensitivity? An investigation using virtual reality

2.1 Participants

Participants were recruited though university campus advertisements, and word-of-mouth. Participants were excluded from participation if they reported any current pain, history of chronic pain, psychiatric or neurological impairment, severe or uncorrected vision, high sensitivity to motion sickness, or were outside the ages 18–65 years. Participants were informed of the nature of the study, but not of the experimental stimulus contingencies or hypotheses. The study was approved by Griffith University Human Research Ethics (GU Ref No: 2016/242). Prior to study commencement, participants completed the Depression, Anxiety and Stress Scale (DASS) and Pain Catastrophizing Scale for descriptive purposes [10, 11]. Sample size was informed by a recent study finding a medium effect of pain-associated tactile cues on pain (ηp2=0.1) [4]. Based on this effect we determined that 24 people would be required for sufficient power to detect an equivalent effect (α=0.05, 1-β=0.8). The sample size was subsequently doubled to allow for subgroup analyses by gender.

2.2 Stimulus material

Electrical stimuli were delivered using a Digitimer DS7A electrical stimulator (Digitimer Ltd, Welwyn Garden City, Hertfordshire, UK) with a 200 μs pulse duration and 300 V max. The electrodes were placed on the dorsal aspect of the left hand midway between the radiocarpal and metacarpophalangeal joints. Vibrotactile stimuli were delivered via a 10 mm, 3 V vibration motor, vibrating at approximately 200 Hz at a location adjacent to the electrocutaneous stimulus. A neoprene band was used to fixate the stimuli.

The electrical stimulus intensity was calibrated for each individual using a numerical rating scale (NRS); where 0=no sensation, 3=first instance of pain, 5=“moderate” pain, 7=“significant” pain and demanding some effort to tolerate, and 10=“worst imaginable” pain. Calibration was performed by increasing the intensity in 0.2 mA increments. Calibration was discontinued when NRS was equal to seven. If the rating of seven was eschewed, the intensity was decreased by 0.4 mA and then increased in 0.1 mA increments. The intensity corresponding to “seven” was used throughout the experiment.

The experimental flow and stimulus timing is outlined in Fig. 1. Three “neutral” scenes (bedroom, office, interior of a car) were used as contextual stimuli. We were confident that these would be inherently neutral with respect to pain, since we recently showed that even VR contexts with emotional valence do not differentially influence pain [12]. The scenes were delivered using the latest Oculus Rift CV virtual reality headset (Oculus VR, LLC, Menlo Park, CA, USA). Scenes were sourced from databases of equirectangular (360^o) photographic (rather than computer generated) images (www.flickr.com/groups/equirectangular) making them highly realistic. To account for potential order effects, the sequence of the three scenes were block-randomised. To block randomise, the conditioning phase was first divided into three blocks of six scenes, while the test phase was divided into two blocks of six, and one block of three scenes. Once divided into blocks, the order of the scenes in each block was electronically randomised. Where randomisation between blocks resulted in more than two scenes being presented sequentially, randomisation was repeated. A total of four different sequences were generated and each sequence was randomly allocated to six of the 24 participants. Within each group of six participants receiving one sequence, the painful and vibrotactile stimuli were allocated to the office or car scenes in counterbalanced fashion. This accounted for any baseline differences in scene properties. The bedroom scene acted as an additional control, and was not paired with painful or vibrotactile stimuli.

Figure 1:

Experiment flow and stimulus timing. All scenes were presented in semi-randomised order. The painful and vibrotactile stimuli were presented in randomised fashion within the 20 s scene presentation for Conditioning phase and Test Phase 2 as shown. During Test Phase 1, the painful and vibrotactile stimuli were presented in the 15 s window after PPT measures were taken.

Custom virtual reality software (MoOVi – Wearable Computer Lab) on a Windows PC (Asus G11, ASUSTek Computer Inc., Taiwan) was used to run the simulation. Audio tracks matching the scenes were also acquired and delivered via Oculus headphones.

2.3 Measurement apparatus

A SOMEDIC algometer (SOMEDIC SenseLab, Hörby, Sweden) was used to measure the pressure pain thresholds (PPTs). The device was applied perpendicular to the skin over the right tibialis anterior muscle, 2 cm inferior to the tibial tuberosity and 1 cm laterally from this point. Consistent force was applied according to the rate control indicator, and the test ceased when participants pressed the attached button indicating their pain threshold had been reached. For each subsequent measure, the point of pressure was re-positioned in a clockwise fashion, with an overlap of approximately 50%, as a means of minimising potential habituation or sensitisation occurring as a result of repeated stimulation at a single location. Prior to commencement of the study, the algometer operator was chosen from the experimenters according to a previous pilot study performed to determine the lowest intra-tester variability. The intra-tester and test-retest reliability of PPT testing with electronic algometers has previously been reported to be excellent to perfect [13].

2.4 Experimental design

A differential contextual conditioning design with context type (pain-associated vs. vibrotactile associated) as the within-subjects factor was used. The experiment consisted of two phases: the conditioning and the test phases. During the conditioning phase, one context was paired with painful electrocutaneous stimulation, another context was paired with vibrotactile stimulation and a third context was non-paired. During the ensuing test phase (Fig. 2), the effects of the three contexts on mechanical pain thresholds (Fig. 2B) and pain intensity (Fig. 2C) was then determined. An assessor blind to scene tested and recorded PPTs and pain intensity.

Figure 2:

Experimental setup. The participant was seated with electrical and vibrotactile stimuli in situ on dorsal aspect of left hand.

After providing written informed consent, the electrical and vibrotactile stimuli were placed on the dorsal aspect of the participant’s left hand (Fig. 2). The two electrodes and the vibrating motor were held in place via an elastic strap. The intensity of the electrocutaneous stimulation was then individually calibrated. Following this, the VR headset was fitted to the participant. Participants remained seated throughout the simulation, and were encouraged to explore each context by looking around as they pleased. Prior to, and after the experiment, four PPT measurements (25 s interstimulus intervals), taken from the right tibialis anterior muscle, were recorded to evaluate changes in sensitivity across the experiment.

During the conditioning phase, the three scenes were each presented for 20 s, with each scene presented six times (Fig. 1). During each presentation of the pain-associated and vibration-associated scenes, four electrocutaneous or vibrotactile stimuli were delivered randomly. Since individuals went directly from one scene to the next, the conditioning phase lasted for approximately 6 min.

In the test phase, each of the three scenes were presented five times (Fig. 2). In the first four presentations of each scene, a PPT measure was taken in the first 20 s after scene onset (Fig. 2B). Participants were informed that the PPTs would be measured at the beginning of each scene, during which time no electrocutaneous or vibrtotactile stimuli would be presented. Following PPT measures, the scenes were displayed for a further 15 s, whereby, in an effort to offset extinction learning, the scenes usual reinforcement stimuli were randomly applied three times (Fig. 2C). In the fifth and final presentation of each scene, the influence of context on pain intensity was evaluated by presenting the electrocutaneous stimuli four times in each scene and asking participants to rate the stimulus, using a 0–100 NRS, whereby 0=no pain, and 100=the worst imaginable pain (Fig. 2C). In order to equalise expectancy among contexts in the test phase, a verbal cue was given prior to commencement of each PPT measure “press the button at the first onset of pain” and prior to each electrocutaneous stimulus “three, two, one and now”. On study completion, participants were asked which stimulus belonged to which scene in order to gauge propositional knowledge of the stimulus associations for descriptive purposes.

2.5 Analysis

Data was examined for normality via scatter and box plots. Since pain threshold data violated the assumption of normality, data was log transformed prior to analysis in accordance with accepted norms.

Repeated measures ANOVA [3(Context: Pain-associated vs. Vibration-associated vs. Non-associated)×2(Sex: Male vs. Female)] was used to investigate the effect of context, and its interaction with sex for both pain thresholds and pain intensity ratings. Results were expressed in terms of statistical significance (p≤0.05), and effect size partial eta squared (ηp2). Any significant interaction with group was further examined by analysing the effect of context within each group using Repeated measures ANOVA [3(Context: Pain-associated vs. Vibration-associated vs. Non-associated)] and post hoc Bonferoni corrected pair-wise comparisons.

Repeated measure ANOVA was also used to investigate the effect of time (prior-, during- and post-intervention) or scene type (bedroom, office and car), independent of stimulus-association on PPTs.

3.1 Participants

Forty-eight healthy volunteers [24 males, mean age (SD)=23 (3.1)] met the inclusion criteria and were included in the study. Participants were classified as within normative ranges with respect to psychological scales including the DASS [Depression: 3.4 (3.4), Anxiety: 5.6 (7.0), Stress 11.0 (6.2)] and Pain Catastrophizing Scale [14.7 (7.6)].

3.2 Effect of conditioning on pain threshold

Overall, pain-associated scenes did not significantly modulate PPTs [F(2,92)=2.6, p=0.08, ηp2=0.053] (Fig. 3). PPTs during the contexts did, however, show a significant interaction with sex [F(2,92)=3.6, p=0.03, ηp2=0.073]. RM ANOVA for each group revealed no effect of context within the male group [F(2,46)=0.9, p=0.4, ηp2=0.04], but did show a statistically significant effect of moderate effect size within the female group [F(2,46)=3.6, p=0.047, ηp2=0.12]. Post hoc comparisons within the female group suggested that sensitivity in the pain-associated context was not significantly different relative to the vibration-associated context (p=0.28), but that sensitivity was greater (i.e. pain-threshold was lower) in the pain-associated context relative to the neutral context (p=0.04, d=0.12). However, these comparisons did not withstand subsequent Bonferroni correction (p=0.56 and p=0.08, respectively).

Figure 3:

PPTs in the pain-, vibration- and non-associated (neutral) scenes, expressed as log transformation of the raw data. Overall, Female and Male data are represented in the respective panel.

3.3 Effect of conditioning on pain intensity

There was no significant main effect of conditioning on pain intensity ratings [F(2,92)=0.78, p=0.46, ηp2=0.02] and no interaction with sex [F(2,92)=1.4, p=0.24, ηp2=0.03], although a trend paralleling the PPT data was observed (Fig. 4).

Figure 4:

Pain intensity scores (100 mm NRS) in the pain-, vibration- and non-associated (neutral) scenes. Overall, Female and Male data are represented in the respective panel.

3.4 Exploratory data

When considered independently of stimulus-association, scene type (Office vs. Bedroom vs. Car) did not significantly influence PPTs [F(2,94)=2.0, p=0.14]. Similarly, no significant effect of scene on pain intensity rating was demonstrated: [F(2,94)=0.2, p=0.8]. A comparison of overall pre-, during, and post-experiment PPTs suggested that sensitivity increased as a function of time [F(2,94)=9.8, p<0.001], with post hoc paired t-tests demonstrating that PPTs were lower post-experiment when compared with during- or pre-experiment [(p=0.001 for both, Mean (IQR): pre=511 (259), during=487 (247), post=472 (199)].

4.1 Sex differences

Sex differences in pain have been previously shown, and are of great interest in understanding the high prevalence of chronic pain among women [14, 15]. Interestingly, no context-dependent effects were apparent in males, however females showed a 6% reduction in pressure pain threshold (i.e. 6% increase in sensitivity) in the pain-associated scene relative to the neutral scene. While this is shy of clinically meaningful magnitude [13, 16], we cannot exclude the possibility that greater effects may arise in clinical scenarios, or in specific gender, or other, subgroups. Several mechanisms from biological and psychosocial domains have been proposed to explain sex differences in pain [14]. For the interpretation of the current data, differences in pain-related learning are of particular interest. Indeed, in analogous pain-related conditioning procedures, women report increasing fear of pain-associated cues, while men do not [17]. Thus, we suggest that further investigation of interactions between pain sensitivity and associative learning among women may be warranted.

4.2 Pain-associated contexts and pain sensitivity

A number of studies have provided experimental evidence of conditioned hyperalgesia, by demonstrating that associative learning procedures result in more painfully rated noxious stimuli when preceded or accompanied by a pain-associated cue [4, 5, 7, 18, 19, 20, 21]. It is not immediately apparent when reviewing the literature, why some procedures effect conditioned hyperalgesia, and some do not. However, differences in the types, strength, timing and number of stimuli are likely to play some role [5]. It is unsurprising that such variables influence the outcome of conditioning given what has been discovered about learning in recent decades. For example, well supported learning theories suggest that not all stimulus types are equally associable, because our evolutionarily history has “prepared” us to associate certain classes of events with greater efficiency than others [22, 23]. Further, conditioned responses are typically viewed as preparatory responses and need not replicate the unconditioned response. Since contexts would not typically have an immediate temporal relationship with injury or noxious stimulation, fear rather than pain may be sufficient to motivate timely defensive behaviour. Thus, pain-associated contexts, such as those used in the current study, might evoked fear rather increased somatic sensitivity.

4.3 Perceptual processes

Most studies investigating the effect of pain-predicting cues on pain have assumed a dominant role of conscious expectancy in driving the effects, such as that observed in the nocebo and placebo response [18, 19]. In the current study, we investigated implicit effects, by controlling for conscious expectation in the test phase. This was achieved by informing participants that no electrocutaneous stimuli would be delivered during the PPT measures, and further, participants were verbally warned prior to each test stimulus. Therefore, it is important to note that our null effects relate specifically to the ability of pain-associated contexts to influence pain over and above the influence of explicit expectation. Other studies have also attempted to isolate the non-conscious effects of pain-associated cues. For example, Jensen et al., showed that “painful” faces presented for an imperceptable duration (12 ms), resulted in the subsequent noxious stimulus evoking significantly more pain. Two further studies have used tactile stimuli presented either simultaneously or at imperceptible durations prior to noxious stimuli, and have also shown pain augmenting effects [4, 7]. Overall, the current study does not support the idea that context-related conditioned hyperalgesia can occur independent of explicit expectation.

4.4 Clinical relevance

One must be cautious in generalising laboratory studies to clinical scenarios. Nonetheless, the study was designed to mimic the clinical scenarios where patients may report that their pain relates more so to certain contexts – such as work, sport, or driving environments – than to others. It is implied by clinical reasoning models that this points to possible mechanical nociceptive contributions, resulting from ergonomic or biomechanical factors inherent in those environments [24]. The current study tested the possibility that past experience might render certain environments capable of contributing to pain through non-nociceptive means, perhaps through implicit or explicit expectation of pain or threat. The results of this study provide no general support for this idea in a healthy pain-free population, although it highlights the possibility of a small effect in some subgroups. We note that (apparently) pain-associated non-nociceptive sensory cues have been shown to modulate [6] and in some cases evoke [9] pain in clinical scenarios, giving some support to the possible clinical relevance of specific pain-associated sensory cues in some pain states, where ongoing investigation might be more fruitful. Should pain-associated cues prove relevant in clinical scenarios, procedures aiming to extinguish pain-related associations may have pain-reducing utility.

4.5 Limitations

A key novel part of this study was its attempt to observe non-conscious effects of pain-associated cues, by equalising the effects of explicit expectation across conditions. Perhaps a flaw of this approach is that the methods used to equalise expectancy, might also invoke cognitive processes that annul implicit effects. Studies that use short inter-stimulus timings, rather than verbal cues, to rule-out the effects of explicit expectation may present a more robust methodology [4, 7, 20, 21], however, such timings would not allow for the investigations of context or the use of mechanical pain threshold outcomes that were used in this study.

Limited effects have frequently been seen in laboratory studies of conditioned hyperalgesia. While this may reflect limitations in the effect itself, it may reflect limitations in laboratory attempts to replicate physiologically relevant scenarios. For example, real-world scenarios would frequently involve greater noxious inputs, longer durations of conditioning, and relationships between stimuli that reflect those from our evolutionary history. We suggest that improved methodologies may assist further in determining the limits of conditioned hyperalgesia. This might be achieved by employing more lengthy and aggressive conditioning procedures; perhaps using wearable devices that deliver pairings of cues and painful stimuli over more significant periods – although ethical and technological obstacles remain barriers.

We also note that we did not account for individual differences in psychological variables that may interact with the conditioning process and also with pain reports. Given the lack of overall effect and the homogeneity of the sample, however, subgroup analysis by these variables was not considered relevant. We further note that the observed effect cannot be generalised beyond healthy, young individuals. Indeed, some individuals, such as those who acquire chronic pain, may have a different propensity towards conditioned hyperalgesia that could be elucidated by future investigation.