The role of attention bias malleability in experiencing pain and associated disability

Participants

Participants were 55 individuals who were recruited via university platforms or social media. Inclusion criteria were being aged between 18 and 60 years, normal or corrected-to-normal vision (e.g.,  via glasses) and a good understanding of English as the experimental session took place in English. People were excluded if they reported current mental conditions, severe and/ or acute pain complaints or neurological conditions (e.g., epilepsy). Participants were also excluded if they used a drug that could affect the central nervous system or medication impacting on somatosensory sensitivity. Additionally, to ensure safety, participants were excluded if they reported heart problems, an electronic implant or were pregnant. Power calculations indicated that for a linear regression with one primary predictor and three control variables, to detect a medium effect size with 80% power, a sample size of 55 participants was needed. Of the 55 participants who were recruited, two participants had to be removed from the analysis, i.e., because the data of the experiment were not properly registered (n = 2). Experimental procedures were approved by the Ethics Review Panel of the University of Luxembourg (ERP 20-079 Malleability), and written informed consent was obtained from all participants. In addition, this study was pre-registered (https://osf.io/edr6k/).

Experimental tasks and methods

Self-report measures

Procedure

Upon arrival in the laboratory at the University of Luxembourg, participants received information concerning the experiment session and were informed that the aim of the experiment was to investigate how an emotional event influences cognitive functioning. Participants were informed that they would experience heat pain stimuli during the experiment session. Next, participants completed the questionnaires (online via Qualtrics), after which the experimenter explained the AM paradigm. Next, participants were asked to put on noise-cancelling headphones for the entire experiment. After completion of the AM paradigm participants performed the heat pain staircase procedure, which was immediately followed by the pain interference assessment for which people performed the RIR task while experiencing alternating blocks of heat pain and no pain (see section 1.2.4). Following the RIR task, participants filled out a final questionnaire probing attention for pain and pain experience during the RIR task. Contrasting the pre-registration, pain experience measures during the RIR task were not included in the analyses as they may be confounded by task performance (i.e., distraction efficacy), which may differ between participants. Afterwards participants were debriefed and thanked for their participation with a 10 EUR voucher or course credits. The duration of the complete experiment was approximately 60 min.

Data-analysis plan

Before performing analyses, incorrect responses and responses faster than 200 ms or slower than 2,000 ms were removed from the AM data. In addition, for each individual, trials with a response latency >2.5 Median Absolute Deviation (MAD) of the median were removed from the analyses of the assessment blocks, resulting in a small loss of data (i.e., data loss ranging between 7.8% and 9.1% for the assessment blocks, e.g., Todd et al., 2023). Finally, participants were excluded if their accuracy was less than 75%. An attention bias (AB) index was then calculated by subtracting the mean latency of congruent trials from the mean latency from incongruent trials for each assessment block separately (i.e., Blocks 1, 3, 5). An index of AM away from pain information was calculated by subtracting the AB index from the assessment block after training from the assessment block before training attention away from pain information (AM_Away index). Similarly, an index of AM towards pain information was calculated by subtracting the AB index from the assessment block after training from the assessment block before training attention towards pain information (AM_Towards index). Finally, an overall AM index was computed by adding the absolute value from one’s AM index towards pain information and the absolute value from one’s AM index away from pain information (AM_Overall index: —AB2 − AB1— +—AB3 − AB2—). Furthermore, and in line with previous research, data from the RIR task were cleaned before RT analyses (Verhoeven et al., 2010; Verhoeven et al., 2011). In detail, RTs faster than 100 ms (i.e., anticipations), outliers (RTs >3 SD above the individual mean) and omissions were removed from the analyses. In addition, we removed each first and last trial from each Block. In total, 9.37% of the RIR responses were excluded from final RT analyses. The pain interference variables (RTs and error rates) were calculated by subtracting the RTs/errors of the trials during the non-painful episodes from the RTs/errors of the trials during the painful episodes.

The main hypotheses, i.e., the predictive value of pain-related AM upon experimental pain outcomes (pain threshold, pain tolerance and task interference), were tested using regression analyses which were performed separately for each AM index. In all regression analyses, we controlled for gender and baseline AB. Age was not included as a control variable due to large homogeneity in the sample. For all analyses Bonferroni corrections were applied for multiple testing. Finally, to explore the relationship between attention indices (i.e., baseline attention bias, AM indices) and individual difference variables, correlation analyses were performed. All data cleaning was performed with R (version 4.0.3; R Core Team, 2020) and statistical analysis were performed with IBM SPSS 27 (SPSS Inc., Armonk, NY, USA).

Descriptive statistics

The mean age of the final sample (n= 53) included in the analyses was 23.66 years (SD= 3.32; range 18–34 years). The majority of this sample (n= 29) was female. Participants reported no elevation in their levels of anxiety (M = 7.51, SD = 5.57), depression (M = 8.08, SD = 7.34) or stress (M = 11.62, SD = 7.25) (see also Table 1).

Attention bias malleability and attention to pain

Three hierarchical regression analyses were performed to investigate the additional value of each of the AM indices in predicting attention to pain. Assumptions for the hierarchical regression were met. For these analyses, gender and baseline AB were added in a first step (control variables) of the hierarchical regression analyses. Analyses indicated that control variables predicted a significant portion of variance in attention to pain (ΔR² = .15, F_(2,50) = 4.25, p = .02). Gender was a significant predictor of attention to pain, indicating that females showed more attention for the pain stimulus than males (b = .37, t₍₅₂₎ = 2.84, p = .01, f₂ = .17). Contrasting our hypotheses, no additional value was found for AM_Towards (ΔR² = .01, F_(1,49) = .51, p = .48, f₂ = .01) , AM_Away (ΔR₂ = .00, F_(1,49) = .00, p = .98, f₂ = .01) and AM_Overall (ΔR₂ = .03, F_(1,49) = 1.72, p = .20, f₂ = .03) in predicting attention to pain.

Attention bias malleability, pain threshold and pain tolerance

Three hierarchical regression analyses were performed to investigate the additional value of each of the AM indices in predicting pain threshold. For these analyses, gender and baseline AB were added in a first step (control variables) of the hierarchical regression analyses. Results indicated that baseline variables predicted significant portion of variance in pain threshold (ΔR² = .14, F_(2,50) = 4.02, p = .02, f₂ = .16). Baseline AB was a significant predictor of pain threshold, indicating that a higher AB at baseline was associated with a higher pain threshold (b = .28, t₍₅₂₎ = 2.09, p = .04). Contrasting our hypotheses, no additional value was found of AM_Towards index (ΔR₂ = .02, F_(1,49) = .97, p = .33, f₂ = .02), AM_Away index (ΔR₂ = .00, F_(1,49) = .02, p = .90, f₂ = .02), and the AM_Overall index (ΔR₂ = .00, F_(1,47) = .02, p = .89, f₂ = .00) in predicting pain threshold when controlling for multiple testing. In addition, three hierarchical regression analyses were performed to investigate the additional value of each of the AM indices in predicting pain tolerance. Again, gender and baseline AB were added in a first step (control variables) of the hierarchical regression analyses. Included control variables did not predict pain tolerance (ΔR₂ = .03, F_(2,50) = .81, p = .45). In addition, no significant additional value was found for AM_Towards index (ΔR₂ = .07, F_(1,49) = 4.07, p = .05, f₂ = .08), AM_Away index (ΔR₂ = .01, F_(1,49) =.34, p = .56, f₂ = .02), and the AM_Overall index (ΔR₂ = .02, F_(1,49) = .91, p = .34, f₂ = .03) in predicting pain tolerance when controlling for multiple testing.

Attention bias malleability and pain interference

Six hierarchical regression analyses were performed to investigate the additional value of each of the AM indices in predicting pain interference on RIR task performance (i.e., mean latency, % errors). For all analyses, gender and baseline AB were added in a first step (control variables) of the hierarchical regression analyses. Included control variables didn’t significantly explain variance in the mean latency (ΔR₂ = .09, F_(2,50) = 2.49, p = .09), and % of errors (ΔR₂ = .01, F_(2,50) = .15, p = .86). Next, regression analysis, controlling for multiple testing, including AM_Towards as the independent variable indicated no additional value of AM_Towards in predicting pain interference on the mean latency (ΔR² = .01, F_(1,49) = .32, p = .57, f₂ = .01) and % errors (ΔR₂ = .00, F_(1,49) = .01, p = .94, f₂ = .00) during the RIR task. Similarly, regression analysis including AM_Away as the independent variable explained no additional variance in predicting pain interference on the mean latency (ΔR₂ = .01, F_(1,49) = .26, p = .61, f₂ =.01) and % errors (ΔR₂ = .00, F_(1,49) = .06, p = .81, f₂ = .01) of the RIR task. Finally, no additional value of AM_Overall was found in predicting pain interference on the mean latency (ΔR₂ = .00, F_(1,49) = .06, p = .81, f₂ = .01) and % errors (ΔR₂ = .01, F_(1,49) = .33, p = .57, f₂ = .01) of the RIR task.

Individual differences

In an exploratory step, we investigated whether pain-related attention indices, including baseline AB and AM indices, were related to individual difference variables (see Table 1). Correlation analyses showed that the AM_Overall was negatively associated with the PCS, such that individuals with increased levels of AM showed lower levels of pain worrying.

Method

Participants

Participants were 100 individuals recruited via prolific (https://www.prolific.com/), a recruitment application to attract participants for online studies (Peer et al., 2017). Inclusion criteria were being aged over 18 years, suffer from chronic pain and experience of persistent pain for at least three months, being fluent in understanding English and have access to a computer. The study procedure was approved by the Ethics Committee of the University of Sydney, and informed consent was obtained from all participants. This study was pre-registered (https://osf.io/st8e2/). Power calculations indicated that for a linear regression with one primary predictor and three control variables, to detect a medium effect size with 80% power, a sample size of at least 55 participants was required. Due to the online nature of the study, and potentially increased drop-out rate, we did substantially overrecruit participants in our sample. The final sample included 71 participants, after excluding those who indicated that they had not been experiencing pain for the past 3 months (n = 22), those who did not complete all aspects of the study (n = 6), and those who had low accuracy rates on the AM paradigm (i.e., accuracy < 75%; n = 1).

Experimental tasks and methods

Self-report measures

Similar to Study 1, participants filled out the PCS (Sullivan, Bishop & Pivik, 1995) and the DASS (Antony et al., 1998; Colster et al., 2008). Additionally, participants completed the Graded Chronic Pain Scale (Von Korff et al., 1992) to assess daily levels of pain experience and disability. The GCPS consists of seven items. Pain intensity items assess pain intensity right now, worst pain intensity over the last six months, and average pain intensity over the last six months. These three items are averaged in a pain intensity score (range 0–100). Pain disability items assess limitation in daily activities because of pain, limitation in recreational, social, and family activities within the last six months, and limitation in the ability to work because of pain within the last six months. These three items are averaged in a pain disability score (range 0–100). Finally, the number of days disabled within the last six months is assessed. All items, except for the number of days disabled, are scored on an 11-point scale (0–10). Previous research has shown that the GCPS is reliable and valid for assessing pain intensity and pain disability in chronic pain samples (Smith et al., 1997).

Procedure

Via the prolific platform, participants received a link to an online survey programmed in Inquisit (Inquisit 5; Seattle, WA, USA: Millisecond Software) and presented via the Inquisit web application. Participants were instructed to sit in a quiet environment during the study (i.e., a room where they do not get disturbed by others), before the online survey started. Participants were presented with an online information statement and gave informed consent by clicking a button on the screen prior to participating. Following consent, participants completed the questionnaire battery. At the end of the survey, participants performed the AM paradigm. The entire duration of the experiment was approximately 45 min, and participants were reimbursed £3.75 for their time.

Data-analysis plan

Before the analyses were performed on the data of the AM paradigm, incorrect responses, and responses faster than 200ms or slower than 2000ms were removed from the data. In addition, for each individual, trials with a response latency >2.5 MAD of the median were removed from the analyses of the assessment blocks, resulting in a small loss of data (i.e., data loss ranging between 7.9% and 10.8% for the assessment blocks). Similar to Study 1, a baseline AB index and AM indices (AM_Away index, AM_Towards index, AM_Overall index) were calculated. To investigate the relationship between pain-related AM and chronic pain outcomes, separate regression analyses were performed for each AM index for both pain intensity and pain disability scores of the GCPS. In all regression analyses, we controlled for gender, age and baseline AB. Data cleaning was performed with R (version 4.0.3; R Core Team, 2020) and statistical analyses were performed with IBM SPSS 27. For all analyses Bonferroni corrections were applied for multiple testing. Finally, to explore the relationship between attention indices (i.e., baseline AB, AM indices) and individual difference variables, correlation analyses were performed.

Descriptive statistics

The final sample comprised 71 people reporting chronic pain that had persisted for at least 3 months. Participants reported experiencing pain for an average of 10.18 years (SD = 9.93; range 3 months–39 years). The majority of the sample (i.e., 43 participants) was female, with a wide age range (M_age = 41.55 years; SD = 14.83; range 20–74 years). Participants reported moderate levels of pain intensity (M = 51.36; SD = 17.51) and had a pain disability score of 41.92 (SD = 25.38) on the GCPS. Furthermore, participants reported mild to moderate levels of anxiety (M = 10.14, SD = 8.93), depression (M = 15.75, SD = 11.62) and stress (M = 17.13, SD = 10.64) (see also Table 2).

Attention bias malleability, pain intensity and pain disability

Three hierarchical regression analyses were performed to investigate the additional value of each of the pain-related AM indices in predicting levels of pain intensity. For all analyses, gender, age and baseline AB were added in a first step (control variables) of the hierarchical regression analyses. Included control variables did not significantly explain variance in pain intensity (ΔR2 = .06, F_(3,67) = 1.39, p = .26). Furthermore, the results indicated no additional value of AM_Towards (ΔR2 = .00, F_(1,66) = .00, p = .97, f₂ = .01), AM_Away (ΔR2 = .01, F_(1,66) = 0.74, p = .39, f₂ = .01), and AM_Overall (ΔR2 = .05, F_(1,66) = 4.03, p =.05, f₂ = .06) indices in explaining the variance in participants’ pain intensity score when controlling for multiple testing.

Next, three hierarchical regression analyses (controlling for multiple testing) were performed to investigate the additional value of each of the pain-related AM indices in predicting pain disability. Again, for all analyses, gender, age and baseline AB were added in a first step (control variables) of the hierarchical regression analyses. Included control variables did not significantly predict pain disability (ΔR2 = .06, F_(3,67) = 1.39, p = .25). Furthermore, no additional explanatory value was found for AM_Towards (ΔR2 = .01, F_(1,66) = .46, p = .50, f₂ = .01) and AM_Away (ΔR2 = .01, F_(1,66) = .95, p = .33, f₂ = .01) indices. Yet, the results indicated that the AM_Overall index significantly added value in explaining the variance of participants’ level of pain disability (b = .27, t₍₇₀₎ = 2.59, p = .01; ΔR2 = .10, F_(1,66) = 7.56, p = .01, f₂ =.09), such that increased levels of pain related AM_Overall was associated with higher pain disability.

General discussion

The aim of the present series of studies was to investigate to what extent individual differences in the readiness to acquire an attentional bias for pain are associated with pain and pain disability or pain-related task interference. This was investigated in a controlled experimental condition with healthy volunteers and using self-reported pain and pain disability measures in chronic pain patients. In Study 1, including healthy volunteers, findings did not show an association between pain-related AM indices, heat pain experience and pain-related task interference. These null findings are in contrast with our hypotheses. However, for Study 2, including chronic pain patients, results indicated that increased levels of AM_Overall did relate to increased levels of pain disability. As such, findings of current research partially supported the hypotheses that malleability in pain-related attention bias (irrespective of its direction) is associated with pain outcomes.

The finding that higher levels of pain-related AM are associated with higher levels of disability, above and beyond participants’ level of pain-related AB, aligns with previous research in the context of anxiety. In particular, findings of Clarke, Chen & Guastella (2012) suggested that the level of AM is a good predictor of self-reported anxiety, whereby increased readiness to acquire an attentional bias (irrespective of its direction) predicted elevations in trait anxiety in response to extended stress (Clarke, MacLeod & Shirazee, 2008). Based upon these findings, they suggested that variability in the ease with which an attention bias can be acquired (as assessed with the AM paradigm), would predict who will naturalistically develop an attentional preference for threat when exposed to extended stressors. Similarly, Lester et al. (2019) stated that their findings indicate that a stronger preparedness to acquire a threat bias in response to an experimental contingency supporting selective threat processing might forecast the development of an attention bias toward threat when facing an extended stressor, predicting experience of increased levels of anxiety. In corroboration with these suggestions, we propose that current study results may be explained by the fact that high levels of AM for pain-related information facilitate the naturalistic acquisition of an attentional preference for pain-relevant information when exposed to prolonged pain, whereby this attentional preference for pain-relevant information may then fuel the maintenance and exacerbation of chronic pain-related disability. Notably, although not surviving multiple testing correction, findings in chronic pain patients concerning the relationship between pain-related AM and pain experience mirrored the findings between pain-related AM and pain-related disability.

Available research in the anxiety domain does suggests that this is only part of the story, indicating that the (mal-)adaptive value of AM is dependent upon contextual features. Indeed, besides being related to elevated levels of trait anxiety in the context of extended stressors (Clarke, MacLeod & Shirazee, 2008), increased levels of AM were also found to be associated with improvement in psychological therapy outcomes when aiming to positively modify attention biases in therapy (Clarke, Chen & Guastella, 2012) . The impact of context may also explain null findings in our healthy population (Study 1) where, in contrast to Study 2 (chronic pain population), a context of prolonged pain is absent. Together, this suggest that the influence of pain-related AM depends upon an individual’s current environmental contingencies whereby uncontrollable extended stressors, such as chronic pain, render an individual more vulnerable, whereas the introduction of psychological treatment may result in a more favorable outcome due to greater malleability of cognitive processes surrounding pain. Following this reasoning, it could be expected that also in people suffering pain that is relatively manageable, increased levels of AM may result in more favorable outcomes (Todd et al., 2023). This contextual view on AM is in line with contemporary views on the existence and potential impact of attention bias upon pain outcomes (e.g., Van Ryckeghem et al., 2019), which suggest that attention bias is dynamic in nature, with the alignment between cognitive processing of pain-related information with context and goal pursuit, rather than cognitive biases themselves being key for maladaptive outcomes (see also Verhoeven et al., 2010). Hereby it may be essential that valuable goals are identified and activated, so the pursuit of goals become salient (Schrooten & Vlayen, 2010; Schrooten, Vlayen & Morley, 2012; Crombez et al., 2016; Van Damme et al., 2010). Moreover, it has been proposed that pursuing an “important” goal for the individual may increase their attention and inhibit less important goals (Schrooten & Vlayen, 2010; Van Ryckeghem & Crombez, 2018). In clinical practice it may accordingly be of importance to identify salient non-pain related goals (e.g., enjoy hobbies; see also Van Ryckeghem & Crombez, 2018; Van Ryckeghem et al., 2019).

Within the present series of studies, we did not only investigate the role of pain-related AM, but also provided a methodology to assess the readiness to acquire an attention bias for pain-related information. This methodology is informed by available research on anxiety-related disorders (Clarke, Chen & Guastella, 2012), but was also significantly improved upon, as the link between the AM indices and pain outcomes cannot be due to the attention bias modification training itself. In doing so, this study provides a tool that can be used in future research and, in later stages clinical practice, to provide an index of peoples’ readiness to acquire an attentional bias towards or away from pain-related information. Taking this AM index into account may then allow to (a) further interrogate the value of AM in predicting the development and/or maintenance of chronic pain problems and associated disability levels, and (b) increase treatment efficacy by tailoring interventions based upon peoples’ readiness to acquire an AB for pain-related information. At current, only a few studies, predominantly those where attention bias training successfully reduced pain-related attention, reported better pain outcomes following attention bias training (Todd et al., 2015; Sharpe et al., 2012, but see Carleton et al., 2020). Therefore, researchers have aimed to improve traditional attention bias modification paradigms by increasing the relevance of pain-related information (Notebaert et al., 2015; Van Ryckeghem, Van Damme & Vervoort, 2018; Vermeir et al., 2023) or including game features to increase motivation and engagement (Vermeir et al., 2022). Yet, current research suggests that differences in peoples’ readiness to acquire an attentional bias for pain-related information may also, at least partly, explain inconsistent findings in existing literature concerning the effect of attention bias modification to reduce pain and/or disability. Depending upon the proportion of individuals having high levels of AM included in the sample, attention bias modification may be less or more successful in modifying attention biases for pain information. This may explain why some studies found changes in individuals’ level of attention bias for pain-related information, whereas others did not and consequently did not find an impact upon pain outcomes (Van Ryckeghem, Van Damme & Vervoort, 2018; Van Ryckeghem & Crombez, 2018).

Some aspects of the current study require further consideration. First, to assess pain interference in healthy volunteers, we made use of the Medoc to induce heat pain stimuli of 20 s combined with the RIR task performance. The use of multiple stimuli of 20 s differs from previous studies aiming to assess pain interference using the RIR task and pain experience or disability in daily life contexts. It may be that participants are able to elevate their effort for a short duration during which pain is experienced, shielding task performance from interruption, which would not be possible for the longer pain stimuli (e.g., induced via the cold pressor task or daily life) (Verhoeven et al., 2010; Verhoeven et al., 2011). Second, participants in Study 1 were mainly pain-free undergraduate students experiencing experimental pain. The homogeneity of the study sample may have limited the variability in individual difference variables and AM, reducing the possibility to find associations with these variables. Therefore, future research may want to replicate the current study in more heterogeneous populations (e.g., non-student populations, older population) to address the impact of individual difference variables upon AM. Third, English proficiency of participants in Study 1 was not explicitly tested but based upon self-evaluation and the observation of the test leader. To ensure that participants have the ability to process semantic and emotional content in an automatic manner future studies using this design may only want to include native English speaking participants. Fourth, within current study, we assessed pain intensity and pain disability in chronic pain patients using self-report measures reflecting over the past six months. Although, we used well-validated measures, they do not allow to assess daily fluctuations in pain intensity and pain disability. More fine-grained assessment methods, such as ecological momentary assessment methodology, do allow to also investigate the role of pain-related AM in daily fluctuations of pain and disability (Van Ryckeghem et al., 2013; Todd et al., 2023). Sixth, it should be noted that attention bias was assessed using the traditional operationalization, i.e., using symbolic information presented in a reaction time paradigm. The paradigm developed to assess AM has similar limitations as the traditional dot-probe paradigm (i.e., the use of static stimuli, low reliability; Dear et al., 2011a; Dear et al., 2011b). Future research should aim to overcome these limitations by using eye-tracking methodology (Franklin, Holmes & Fowler, 2019; Jackson, Yang & Su, 2019), or paradigms that make use of more ecological valid pain stimuli, such as conditioned pain cues (Van Ryckeghem & Crombez, 2018) or realistic pain information presented via virtual reality methodology.