Accepted Manuscript The effect of Cognitive Bias Modification for Interpretation (CBM-I) on avoidance of pain during an acute experimental pain task Emma Blaisdale Jones, Louise Sharpe PII: DOI: Reference:
S0304-3959(14)00222-X http://dx.doi.org/10.1016/j.pain.2014.05.003 PAIN 9205
To appear in:
PAIN
Received Date: Revised Date: Accepted Date:
19 February 2014 7 April 2014 1 May 2014
Please cite this article as: E.B. Jones, L. Sharpe, The effect of Cognitive Bias Modification for Interpretation (CBMI) on avoidance of pain during an acute experimental pain task, PAIN (2014), doi: http://dx.doi.org/10.1016/j.pain. 2014.05.003
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
The effect of Cognitive Bias Modification for Interpretation (CBM-I) on avoidance of pain during an acute experimental pain task
Emma Blaisdale Jones and Louise Sharpe*
School of Psychology, The University of Sydney, Brennan MacCallum Building A18, NSW 2006, Australia
Text pages: 35. Word Count: Introduction =485; Discussion =1499 Figures: 1 Tables: 5 Correspondence author: Louise Sharpe Room 450, Brennan MacCallum (A18) The University of Sydney NSW 2006 Australia Phone: +61 2 9351 4558 Email:
[email protected] Institutional URL: http://sydney.edu.au/science/people/louise.sharpe.php
Keywords: cognitive bias; experimental pain; interpretation; threat
1
ABSTRACT Research confirms that patients with chronic pain exhibit a tendency to interpret ambiguous stimuli as pain-related. However, whether modifying these interpretive pain biases impacts pain outcomes is unknown. This study aimed to demonstrate that interpretation biases towards pain can be modified, and that changing these biases influences pain outcomes in the cold pressor task. One hundred and six undergraduate students were randomly allocated to receive either threatening or reassuring information regarding the cold pressor. They were also randomly allocated to one of two conditions in the Ambiguous Scenarios Task, where they were trained either to have a threatening interpretation of pain (pain bias condition) or a non-threatening interpretation of pain (no pain bias condition). Therefore, the study had a 2 (threat/reassuring) x2 (pain bias/no pain bias) design. Analyses revealed that a bias was induced contingent on condition, and that the threat manipulation was effective. Participants in the pain bias condition hesitated more before doing the cold pressor than those in the no pain bias condition, as did the threat compared with the reassurance condition. The major finding was that interpretive bias mediated the relationship between bias condition and hesitance time, supporting the causal role of interpretive biases for avoidance behaviours in current chronic pain models. No differences were found on other pain outcomes regarding bias or threat condition and the efficacy of the bias modification was not impacted by different levels of threat. These results suggest that cognitive bias modification should be further explored as a potential intervention in pain.
2
1. Introduction Current models of chronic pain [3,12,44,46] suggest maladaptive interpretations of pain as harmful are important in the condition’s development. These interpretations lead to pain-related fear and anxiety, which promote avoidance of potentially pain-provoking situations. A vicious cycle is then created through increasing fear of pain and hypervigilance towards pain-provoking stimuli. Recent research on information-processing biases in chronic pain has focused mainly on attentional biases. Specifically, the dot probe task has identified attentional biases in chronic pain patients [10,16,38]. However, the effect sizes in this research are small [9], and the studies are inconsistent [37]. Despite this, Attentional Bias Modification (ABM) for pain has been successful in laboratory studies [27,40], and has shown therapeutic potential in clinical studies [36,39]. Although there is less research on interpretation biases in chronic pain, the results are more consistent. All existing studies [13,15,28,33,34] indicate that when presented with ambiguous stimuli, chronic pain patients more likely generate a painful resolution compared with controls. This raises the question of whether interpretive biases can also be modified and whether modifying them would have therapeutic value. Cognitive Bias Modification for interpretation (CBM-I) has had positive outcomes in laboratory studies of anxiety [6,26,35,42], and depression [23,24,31]. Importantly, these outcomes have translated into encouraging clinical results, [2,17,22]. However, CBM-I has not been extended to pain. This is surprising given the evidence for interpretive biases in pain patients [32]. Furthermore, other bias modification protocols such as ABM have positively impacted pain [37], despite the attentional bias literature being mixed. Before undertaking a
3
clinical study, it is necessary to show that one can change interpretation biases towards pain; and that changing these biases influences pain outcomes. The present study aimed to modify participants’ interpretations of situations as being pain related or not pain-related using CBM-I. The effects of the CBM-I were assessed on pain outcomes to an acute pain task. Although healthy participants were not expected to have a pain bias, models [3] predict a bias would be more likely under conditions of high threat. Therefore, in addition to varying the direction of the training, the relative efficacy under conditions of high and low threat was investigated. We hypothesized that: (1) CBM-I would modify participants’ biases towards pain, contingent on their assigned condition (pain bias or no pain bias); (2) that compared to the no pain bias condition, participants in the pain bias condition would have worse pain outcomes, particularly those related to avoidance and escape; and (3) that the induced bias would be associated with the pain outcomes described above, and a potential mediator between CBM-I group and outcomes. Since threat is associated with poorer pain outcomes e.g.[5], it was hypothesized that (4) those in the threat condition will report worse outcomes than the reassuring condition. However, because CBM-I has not previously been used in the pain literature, it is unclear whether one would expect CBM-I to be more or less effective under conditions of high threat.
4
2. Method 2.1 Participants Participants were 106 first year psychology students who completed the study in return for course credit. Eight participants needed to be excluded from the study, leaving a total of 98 for the analysis. Exclusion criteria included excessive consumption of caffeine or alcohol within the previous 24 hours (n=1); reported recent use of analgesics (n=2); or existing moderate pain (> 4/10 on a 1-10 numerical scale) (n=5). Participants were aged between 17 and 41 (M=20.22, SD=4.42), and 69 were female. Using www.randomizer.org, participants were randomly allocated to one of two threat conditions (threat/reassurance), and then re-randomized to one of two CBM- I conditions (pain bias/no pain bias). This resulted in a 2x2 design.
2.2 Procedure Participants were tested individually for 1 hour. The University of Sydney Human Research Ethics Committee approved the study before testing commenced. All participants read the relevant information statement and gave written consent. For the duration, those in the threat condition received all information about the cold pressor task as the ‘vasodilation task’, while the reassurance group received information referring to the ‘cold pressor task’. Both groups were lead to expect the same level of pain severity, but information about the ‘vasodilation task’ was threat inducing. In contrast, the information regarding the ‘cold pressor task’ was threat alleviating, providing reassurance about the safety of the task (see below). After consent, participants completed a series of demographics questions, followed by the Depression Anxiety Stress Scale (DASS-21), the Positive and Negative Affect Schedule (PANAS), the Fear of Pain Questionnaire (FPQ-III) and the Pain Catastrophizing Scale (PCS).
5
They were then given further information about the vasodilation/cold pressor task, and completed the manipulation check questions. Subsequently, participants were given verbal and written instructions on how to complete the bias modification task. They were presented with 2 practice questions before task commencement. After the CBM-I task was completed, participants were asked to rate how vividly they were able to imagine themselves in each scenario. They then completed the first half of the recognition task following verbal and written instructions. Since previous studies [26], have administered a filler task in the midst of the recognition task, the PANAS was readministered at the task’s halfway point. This also allowed us to investigate the impact of the manipulation on positive and negative affect. Following this, the second half of the recognition task was administered. The final stage of the study was engagement in the cold pressor task. Participants were asked to place their hand in the warm-water tank for 30 seconds, followed immediately by the cold tank. The experimenter (EBJ) timed the outcomes with a stopwatch. Timing of pain measures commenced after participants removed their arms from the warm water tank and as soon as the participants were asked to submerge their arm in the cold tank, when the experimenter issued the instruction “whenever you are ready”. Measures taken (in seconds) were hesitance (time taken to fully submerge arm following the instructions “whenever you are ready”), threshold (time taken for participant to first report feeling pain), and tolerance (length of time before participants withdrew arm from the cold tank). Additionally, participants rated their level of pain at threshold and again at tolerance. Self-reported subjective units of distress were also recorded. The task was terminated when participants withdrew their arm, or after 4 minutes [11]. Upon arm removal, participants were thoroughly debriefed.
6
2.3 Measures 2.3.1 Depression Anxiety and Stress Scales-21 (DASS-21)[25] The DASS-21 is a 21-item measure with three 7-item subscales assessing depression, anxiety and stress. Its psychometric properties are good, with reliabilities ranging between .82 and .93 [25]. It has also been shown to possess adequate construct validity[7]. The internal consistency for the present study was high, at α = .93
2.3.2 Positive and Negative Affect Schedule (PANAS)[47] The PANAS is a 20-item self-report measure of positive and negative affect. The internal consistencies of the PANAS scales, are high (α = .89 and .85 for positive and negative scales respectively) [8]. Negative and positive affects were measured at 2 time point to determine the impact of CBM-I on affect. This ensured that any impact of CBM-I could not be attributed to changes in affect alone. In the current study, the internal consistency was α = .95 for positive affect, and α = .92 for negative affect.
2.3.3 Fear of Pain Questionnaire (FPQ)-III [29] The FPQ consists of 30 self-report items assessing pain-fearfulness. The test-retest reliability is good (r=. 74) and internal consistency is excellent (α=. 92) [29]. For the current study, the internal consistency was α = .92.
7
2.3.4 Pain Catastrophizing Scale (PCS)[43] The PCS is made up of 13 items assessing catastrophizing in response to pain. There is excellent consistency in both community (α=. 95) and chronic pain samples (α=. 92) [30]. The internal consistency for this study was α = .90.
2.4 Materials 2.4.1 Threat Manipulation [5] Participants in the ‘threat’ group received information relating to the ‘vasodilation task’, which highlighted some physical mechanisms of pain through use of biomedical terminology, whilst also providing an extreme example of exposure to the cold (i.e. frostbite), as the following excerpt from the participant information statement indicates: “It [the vasodilation task] is designed to stimulate the sympathetic nervous system, and increases both blood pressure and the radial arterial pulse. The arteriolar vasoconstriction also results in decreased blood flow to the skin. This particular biological reaction is most commonly seen in patients suffering from frostbite. (Frostbite is a condition which occurs upon exposure to extreme cold and refers to the freezing and destruction of bodily tissues as a result of poor circulation. As an effect of increasing numbness, there are often no warning symptoms at the onset of the condition, and when severe, it may result in gangrene or amputation of the affected limb.)”
Contrastingly, those in the reassurance group received reassuring information relating to the ‘cold pressor task’, which used every day terminology to explain the processes involved, and likened the task to pulling a drink out of an esky, as the following excerpt indicates:
8
“Your body may experience a number of biological reactions in response to the cold, such as numbness or slightly increased heart-rate. Many of these processes also occur as a result of ‘everyday’ situations (e.g. experiencing numbness/’pins and needles’ from sitting in one position for too long). While the task might seem quite unusual, the cold pressor experience is very similar to searching for a can of drink in an ice-filled container at a party.”
2.4.2 Cognitive Bias Modification for Interpretation (CBM-I) As CBM-I has never been administered in pain research before, we needed to choose one of the many paradigms that have been used to modify cognitive biases. The Ambiguous Scenarios method [26] was selected for use in the current study because it has been the most commonly used and has resulted in robust results in the anxiety literature. Participants were presented with a description between one and two lines in length that ended with a word fragment. They were instructed to use their understanding of the paragraph to guide solution of the word fragment. Descriptions remained ambiguous in terms of their emotional interpretation until solution of the final word, which alone determined the assignment of the pain bias or no pain bias condition. Participants were instructed to imagine they were the character being described. The 30 descriptions were run once through in a random order, and then displayed to the participant a second time, again in a random order. Therefore, 60 descriptions were presented in total. An example of a description according to CBM-I condition is shown in Table 1. Once the word fragment was solved, a comprehension question followed, which could be correctly answered only by having the appropriate interpretation of the previous paragraph. For example, after the previous word fragment was solved, the following comprehension question was:
9
“Did you cut yourself?” A (Yes/No) prompt appeared underneath each question, and feedback was given for answers depending on the assigned condition. Solving the word fragment and correctly answering the comprehension question therefore confirmed the valence of the interpretation. The 60 descriptions and associated comprehension questions were presented as one block.
2.4.3 Word Stimuli Word stimuli were adapted from a lexical decision task in a previous pain study [45], which had been piloted to ensure accessibility of completion words. Participants in the pain bias condition solved word fragment relating only to painful/health-threatening outcomes, whereas those in the no pain bias condition solved only for non-threatening outcomes that did not relate in a painful outcome.
2.4.4 Cold Pressor Apparatus Two water tanks manufactured by Thermoline Scientific Australia were used to conduct the cold pressor task. Each tank was comprised of a 20L circulating water bath chamber (model TLWB-30), immersion cooler (TIC-400) and a heat circulator (TU-3). In order to ensure participants initial arm temperature was controlled for, they were firstly asked to submerge their arm up to the elbow for 30 seconds in the warm tank, with a maintained temperature of 37°C (± 0.5). Following this, they submerged the same arm up to the elbow into the tank of cold water, maintained at 5°C(± 0.5). Research in the past indicates the cold pressor task produces the “Lewis Effect”, whereby pain is produced by vasodilation of the blood vessels due to warm water, quickly followed by vasoconstriction due to cold water [1].
10
2.5 Manipulation Checks 2.5.1 Threat Manipulation Check [5] The first manipulation check consisted of four questions assessing the degree to which the given information about the cold pressor task successfully induced or alleviated threat. Questions related to perceived harm and worry about the task, predicted level of pain, and perceived self-efficacy to cope with the task. Each was rated on an 11-point numerical scale (010). If the manipulation was effective, differences would be seen for worry, harm and coping, but not for pain, as both groups were instructed that they would have similar levels of pain.
2.5.2 Bias Manipulation Check: ‘Recognition Task’ The procedure for the bias manipulation check was adapted from the ‘Recognition Task’ [26]. It assessed whether participants made valenced interpretations (pain-related or not painrelated) of new ambiguous descriptions matching their assigned bias condition. Firstly, participants were presented with ten ambiguous pain/health-related paragraphs where completion of the final word fragment was required, much the same as in the ambiguous scenarios task. However, in contrast to the ambiguous scenarios, the final word was designed to preserve the ambiguity of the situation. Similarly, the comprehension question that followed each scenario required no emotional interpretation, because the question was designed only to ensure the context of the scenario was understood. Each of the scenarios was uniquely identified by a title. A sample test paragraph is shown in Table 2. The second part of the recognition task was designed to assess the interpretations made of the original scenarios. In a randomized order, the identifying title of each scenario, followed by four versions of the final sentence was presented to participants. Of these sentences, there were
11
two possible valences, (pain interpretation or no pain interpretation) relating to the bias conditions. Sentences were also categorised into either targets (most relevant to outcomes of the bias modification stimuli) or foils (more generally valanced outcomes, less relevant to bias modification stimuli). The sentences in Table 3 are examples of these. Participants were told none of the sentences were worded identically to the initial scenario, but that any number of them could be related to the previous situation in meaning. They then rated each sentence, independently of all others, for its similarity in meaning to the original. For each set of similarity questions, the 4 sentences were presented together with boxes to rate the similarity with the previous scenario, ranging from 1 (very different in meaning) to 4 (very similar in meaning). All of the word stimuli for this task were novel.
2.6 Imagery Check Visual imagery of descriptions enhances the degree to which CBM-I can induce emotional outcomes [18]. Therefore, participants were asked to rate on a 5-point numerical scale how easily they were able to imagine themselves as the protagonist in the ambiguous scenarios. This ranged from 1 (not at all) to 5 (extremely).
2.7 Pain Measures Five pain measures were taken during the experiment. The dependent variables were hesitance, pain threshold, pain tolerance, pain ratings and distress. Because theories [e.g.47] suggest that interpretation of pain as harmful is important in prompting avoidance of painprovoking situations, we expected that CBM-I would be most likely to impact on the behavioural indices: hesitance (avoidance) and tolerance (escape). Hesitance is the time taken in seconds for
12
participant to submerge their arm and therefore is the operationalization of avoidance. Tolerance was the maximum time in seconds the participant kept their arm in the cold pressor before withdrawing it and hence, relates to escape from a pain-provoking task. We also included the usual outcomes that are measured in cold pressor studies: threshold, pain and pain-related distress. Threshold is the time taken in seconds for participant to first report pain after placing their arm in the cold pressor, most similar to hypervigilance (i.e. how quickly pain is reported). The pain ratings required participants to report a level of current pain on an 11-point numeric scale (0-10), with zero being no pain at all, and 10 being the most extreme pain. These were recorded at threshold and again at tolerance. Finally, participants rated how distressing they found the cold pressor on an 11-point numeric scale (0-10), with zero being no distress, and 10 being high distress.
2.8 Analyses Based on the effect size achieved on state anxiety indices using the Ambiguous Scenarios Method (d=.63) [26], power analyses [14] revealed a sample size of 20 per group would be sufficient to detect this in the current 2 x 2 design. In order to determine whether the baseline variables had been evenly distributed across groups, a series of 2 (threat) x 2 (CBM-I) ANOVAs were performed. Any variables differing significantly between groups were used as covariates in the main analyses to ensure the difference did not account for or conceal any of the major findings. To ensure that the experimental manipulations had successfully altered the variables of interest, manipulation checks were carried out. Firstly, a series of one-way ANOVAs were run on the four manipulation check questions to determine whether the threat condition had
13
successfully induced threat. Successful induction of the relevant bias is demonstrated by a higher similarity rating to disambiguated versions (targets) matching participants’ assigned valence (pain or no pain) of the bias modification conditions. Previously, the analysis method for bias induction from the recognition task has been comparison of both valence and target scores in relation to bias condition via a mixed models ANOVA [26]. However, this method does not produce a single value to account for a general ‘bias score’. It was of interest to determine not only whether there were training effects for CBM-I, but also whether the interpretive bias was associated with pain outcome. Therefore, an interpretive bias score was required for each participant in order to obtain a single outcome variable for use in the regression analyses. In other CBM paradigms, a bias score is developed by subtracting the response of participants to trials where the probe follows the neutral stimulus from those where an experimental stimulus is followed by the probe [20,21]. Hence, a similar bias score was developed in the current study by subtracting the average no pain target score from the average pain target score.
Bias score = (average pain target score) – (average no pain target score).
The score thereby represents the degree to which participants favoured the painful interpretation (a positive score) in comparison to the no pain interpretation (a negative score). In order to determine the impact of CBM-I on positive and negative affect a (2) (time) x 2 (threat) x 2 (bias) mixed model ANCOVA was conducted to determine change scores in both positive and negative affect scores from baseline to post bias modification. Previous research [18,19] suggests that the ability to imagine the scenarios is associated with the success of CBM-I. So, an independent samples t-test was run across CBM-I group for imagery, to determine whether the ability to imagine was independent of bias group.
14
The main analyses were interested in the impact of threat and CBM-I on pain outcomes. Hence, a series of 2 (threat) x 2 (CBM-I) between subjects ANCOVAs were conducted on hesitance, threshold, tolerance, pain rating and distress to determine any group differences. Finally, a series of multiple hierarchical regression analysis were planned to establish predictors for those pain outcomes where an effect of CBM-I (or an interaction) were observed, using the steps outlined by Baron & Kenny [4]. Those baseline or demographic variables that correlated with the relevant pain outcome were added to the first step of the model. Controlling for covariates, we determined whether CBM-I was a predictor of pain outcome; whether CBM-I was a predictor of interpretive bias; and whether interpretive bias was predictive of pain outcome. The full model was then conducted to determine whether the relationship between CBM-I and outcome was still significant when interpretive bias was added to the model. Potential mediation was supported, and so a Sobel test was conducted to confirm mediation.
3. Results 3.1 Participant Characteristics A summary of participant characteristics is shown in Table 4. Significant group differences were found only for FPQ-III scores on CBM-I condition (F(1,94)=4.29,p=.041) and DASS stress scores on the threat x bias interaction (F(1,94)=5.18,p=.025). FPQ-III scores were controlled for in all analyses where CBM-I condition was an independent variable. DASS stress scores were controlled for in all analyses where a threat x CBM-I interaction was examined. In order to test the robustness of the results, we re-ran the analyses without the covariates and the pattern or results remained unchanged.
15
3.2 Manipulation Checks 3.2.1 Threat Manipulation Check On average, participants in the threat group were significantly more worried about the cold pressor task (F(1,96)=9.75, p=.002), thought the task was significantly more likely to cause them harm (F(1,96)=27.47,p
