Psychometric evaluation of the Pain Vigilance and Awareness Questionnaire (PVAQ) in fibromyalgia: Confirmatory factor analysis and the development of a Swedish 8‐item version

Excessive attention to pain, or hypervigilance, is associated with negative outcomes in chronic pain conditions such as fibromyalgia. The Pain Vigilance and Awareness Questionnaire (PVAQ) is a self‐report questionnaire to measure attention to pain. This study aimed to evaluate the psychometric properties of a Swedish version of the PVAQ.


| INTRODUCTION
Excessive attention to pain, or hypervigilance, is a largely automatic process of attending to and evaluating the body for painful sensations.Attending to painful sensations and modifying behaviour accordingly is a normal and adaptive response to acute pain (Melzack & Katz, 2013).However, a greater tendency to attend to painful sensations has been suggested as a maintaining factor in chronic pain conditions (McCracken, 1997).
Fibromyalgia is a chronic pain condition characterized by widespread musculoskeletal pain, tenderness, and fatigue (Häuser et al., 2015), and is associated with a high degree of disability, reduced quality of life, and high costs on an individual and societal level (Sicras-Mainar et al., 2009).Patients with fibromyalgia have reported experiencing higher vigilance to pain (Crombez et al., 2004;McDermid et al., 1996) compared to other chronic pain patients.
While several versions of the PVAQ have been psychometrically evaluated, there has been no previous evaluation of a Swedish translation, and very few in the fibromyalgia population.This study aimed to evaluate the psychometric properties of a Swedish translation of the PVAQ in a sample of participants with fibromyalgia.Based on previous evaluations, we hypothesized that: a.The full range of item response alternatives would be used, without evidence of floor or ceiling effects.b.A two-factor model would achieve the best model fit.c.Cronbach's alpha would be approximately 0.80-0.90for the full PVAQ, and the corrected item-total fullscale correlations would be 0.40 or higher for all items except eight and 16 (McCracken, 2007).d.Positive correlations would be found with pain catastrophizing (moderate to strong), anxiety (weak to moderate), depressive symptoms (weak to moderate), functional impairment (weak to moderate), fibromyalgia severity (moderate), and pain intensity (weak to moderate).

| Design
This was a cross-sectional psychometric study based on baseline data from a randomized controlled trial of cognitive behavioural therapy for self-referred adults with fibromyalgia (N = 274).The trial was based at Karolinska Institutet, Stockholm, Sweden, was approved by the Swedish Ethical Review Authority (2021-03302) and was preregistered at Clini calTr ials.gov(NCT05058911).This research was conducted in accordance with the declaration of Helsinki, and Swedish and European Union data management and privacy legislation.

| Recruitment
The trial was advertised in social media, newspapers, and via patient organizations and health care clinics.Self-referred applicants completed a series of self-report questionnaires and provided informed consent via a secure online platform.A structured telephone interview was then held with a licensed psychologist or master-level psychology student, to assess eligibility for the clinical trial.Eligibility criteria were: (1) aged minimum 18 years, (2) to have been previously diagnosed with fibromyalgia by a physician, (3) to be living in Sweden, and ( 4 Internet.Exclusion criteria were: (1) severe depression, defined as ≥30 on the Montgomery-Åsberg Depression Rating Scale -Self-Rated (MADRS-S; Svanborg & Åsberg, 1994), (2) suicidal ideation, defined as ≥4 on the MADRS-S item nine, (3) psychosis, alcohol, or substance use disorder as a primary diagnosis or likely to interfere with treatment, (4) ongoing psychological treatment, (5) pregnancy (> 29week gestation), ( 6) another somatic condition requiring immediate treatment or deemed to be the primary condition, and (7) inability to use a web-enabled device or insufficient knowledge of the Swedish language.Applicants were asked whether they had previously been diagnosed by a physician and what year, although medical records were not reviewed.Applicants who met all eligibility criteria (N = 274) completed an online pre-treatment assessment which included the PVAQ.For participants who reported using psychotropic medication, a stable dose for at least 4 weeks was required.Applicants who were excluded were referred to routine care services when deemed necessary.

| Instruments
2.3.1 | Pain Vigilance and Awareness Questionnaire (PVAQ) All self-report questionnaires were administered via the study web platform.The PVAQ is a 16-item questionnaire, where each item is scored as a six-point Likert scale that ranges from zero (never) to five (always), resulting in a sum score between zero and 80 (McCracken, 1997).The Swedish translation was completed in June 2021, following the backtranslation guidelines of Brislin (1970).First, the PVAQ was translated into Swedish by two PhD-level licensed psychologists with expertise in fibromyalgia (MHL and EA).Second, the instrument was back-translated into English by a bilingual registered nurse.This translation was reviewed by the Swedish translators.Adjustments were made to the Swedish translation, which was then back-translated to English a second time, and the accuracy to the second translation was deemed to be sufficient (see Appendix S1).
2.3.2 | Self-report questionnaires used to assess construct validity We measured pain catastrophizing using the Pain Catastrophizing Scale scored 0-52 (PCS; Sullivan et al., 1995; in this study: α = 0.92), disability using the 12-item World Health Organization Disability Assessment Schedule 2.0 scored 12-60 (WD2-12, Üstün et al., 2010; α = 0.85), general anxiety using the 2-item Generalized Anxiety Disorder scored 0-6, 3 being the clinical cut-off (GAD-2; Kroenke et al., 2007; α = 0.84), depression using the 2-item Patient Health Questionnaire scored 0-6, three being the clinical cut-off (PHQ-2; Kroenke et al., 2003; α = 0.79), and overall fibromyalgia severity using the Fibromyalgia Impact Questionnaire scored 0-100 (FIQ; Burckhardt et al., 1991; α = 0.78).The FIQ is a composite measure of somatic symptoms, functional impairment, and emotional disturbance.On the FIQ, <39 is indicative of mild symptoms, 39-58 of moderate symptoms, and ≥ 59 of severe symptoms (Bennett, 2005).To measure pain intensity, we scored the FIQ pain item separately, from 0 (no pain) to 10 (very severe pain).Additionally, we employed the pain intensity subscale of the Brief Pain Inventory -Short Form pain severity subscale, scored 0 to 10 (BPI-SF; Mendoza et al., 2006; α = 0.88).In a previous RCT, for a secondary study, we included the BPI-SF as a measure of pain intensity in addition to the FIQ.Unexpectedly, we then found the BPI-SF severity subscale to be only weakly correlated with the FIQ pain intensity item.To maintain a broad perspective on pain intensity, we decided to include both measures in the present RCT as well as the present study.The BPI-SF starts with a portal question about the experience of pain other than toothache mild headache and acute injuries during the past week.In this study, if the respondent chose "no", the BPI-SF was scored as zero.If the respondent chose "yes", the subsequent four pain items were scored from 0 to 10, and the BPI-SF was scored as the mean of these items.

| Statistical analysis
We conducted the statistical analyses in Jamovi (version 2.2.5) and the analysis proceeded in three phases: First, we evaluated item properties and interrelations to ensure that the data were suitable for factor analysis.Second, we assessed the factor structure of the scale.Third, informed by the factor analysis, we proceeded to assess the internal consistency of the scale, as well as the convergent and discriminant validity versus other constructs.A significance level of 5% was used, except for the calculation of conventional 90% confidence intervals for the Root Mean Square Error of Approximation (RMSEA) (see below).
In phase one, descriptive statistics were inspected to assess potential floor and ceiling effects both on item and scale level.Floor or ceiling effects were considered present if >15% of the participants obtained the lowest (0) or highest (80) possible score on the total PVAQ or the lowest (0) or highest (5) possible score for each item (Terwee et al., 2006).Item score distributions were further inspected visually for deviations from normality.
In phase two, confirmatory factor analysis (CFA) was used to systematically evaluate factor structures I -V found in earlier studies (see Table 1 for an overview).Models were fit using the maximum likelihood estimator and the following fit indices were used: the Comparative Fit Index (CFI), the Tucker Lewis index (TLI), the Root Mean Square Error of Approximation (RMSEA), and the Standardized Root Mean Square Residual (SRMR).Benchmark criteria for an acceptable model fit were: CFI ≥0.95; TLI ≥0.95; SRMR <0.08 and RMSEA <0.08 (fair), or ideally <0.06.For the CFI and TLI, values closer to one are indicative of a better fit, and for the SRMR and RMSEA, values closer to zero are indicative of a better fit (Hu & Bentler, 1999).For each model, if the model fit was not satisfactory, modification indices (MI) were computed to provide more information about the model fit.If theoretically justifiable, error covariance was added over item pairs with high MI values to improve model fit (MacCallum et al., 1992).
In phase three, internal consistency was assessed by calculating Cronbach's alpha.Coefficients were calculated for the subscales of the PVAQ as well as for the total scale.Cronbach's alpha ≥0.9 is usually regarded as excellent, ≥ 0.8 as good, and ≥0.7 as acceptable.These coefficients were complemented by adjusted item-total correlations (ITC), i.e., correlations between items and the rest of items of a sum score.A benchmark for ITC of >0.50 was used.

| Sample characteristics
Sample characteristics are presented in Table 2, and the recruitment flow is illustrated in Figure S1.The typical participant was a 51-year-old (SD = 12) woman (98%) who had lived with a diagnosis of fibromyalgia for 11 (SD = 9) years.Sixty-nine percent of participants reported one or more somatic comorbidities, including other chronic pain conditions.

| Item properties
Item score distributions of the 16-item version (PVAQ-16) are presented in Table 3.There was no evidence of

| Internal consistency
The PVAQ-9 and PVAQ-8 whole scales showed evidence of adequate internal consistency, fully on par with the PVAQ-16 (all three versions αs = 0.90).The Passive Awareness and Active Vigilance subscales were also comparable as both subscales had an α of 0.86 in all three versions.Adjusted item-total correlations (ITCs) for the PVAQ-9, PVAQ-8, and PVAQ-16 are presented in Table 4.All ITCs for the PVAQ-9 and the PVAQ-8 were above the benchmark.This was not the case for the PVAQ-16, which had several items with weak ITCs (1, 4, 8, and 16).Correlations with other constructs are illustrated in Table 5.Overall, the correlations with other constructs were very similar between the Swedish versions of the PVAQ-8, the PVAQ-9, and the PVAQ-16 in this study.Correlations with pain intensity were weak and not significant for the BPI-SF.The BPI-SF and the FIQ pain item correlated moderately (r = 0.40).For the BPI-SF, 17 out of 274 respondents answered "no" to the portal question, resulting in a score of 0. Some of these participants simultaneously scored high on the FIQ pain item which raised questions about potential measurement error.Excluding these participants, the correlation between the BPI-SF and the PVAQ-8 was significant, although still weak (r = 0.19), and there was a stronger correlation between the BPI-SF and the FIQ pain item (r = 0.65).when conducting factor analyses, and the evaluation of previously published factor structures in a confirmatory framework.The participants also completed a relatively large number of questionnaires which could be used for the assessment of construct validity.

| Expected item characteristics
In line with previous studies, item scores were approximately normally distributed with no evidence of floor or ceiling effects.Item 10 deviated slightly from the normal distribution and items one and four generally showed weak inter-item correlations.The reverse-keyed items, eight and 16, exhibited weak inter-item correlations, and item 16 did not correlate significantly with any of the other items.This was an expected finding, considering that most previous studies have excluded these reversekeyed items due to weak item correlations (e.g., Roelofs et al., 2003;McCracken, 2007).Suárez-Álvarez et al. (2018) suggest that a combination of positive and reverse items can be a threat to the reliability of the scale, while Terwee et al. (2018) state that items and response options have to be understood by the intended population to ensure validity.A clinical observation in our study was that some participants reported difficulties answering the reversed items, since they are negated statements with response options such as "Never" and "Always".

| Benefits of the PVAQ-9 and PVAQ-8
After evaluating previously published factor solutions by way of CFA, a two-factorial 9-item version of the PVAQ achieved the best fit.This factor solution was introduced by Esteve et al. (2013) who studied a Spanish sample with chronic back pain and subsequently replicated the same model in a Spanish sample with fibromyalgia (Pilar Martínez et al., 2014).The goodness-of-fit indices were similar in this study, but stricter benchmark criteria were used.To achieve an acceptable model fit for the PVAQ-9, Model V was modified by the addition of residual variance for items 9 and 11 (Model VI).Considering that these items are similarly worded, it is unsurprising that they share local error variance which may justify the addition of the covariance parameter (MacCallum et al., 1992).An 8-item model that excluded item 11 (the item with the lowest factor loading of the two) exhibited an acceptable model fit without the addition of error covariance.In all other respects, the psychometric properties of the PVAQ-8 were seemingly similar to those of the PVAQ-9.The internal consistency for the PVAQ-8 was excellent for the total scale and good for the subscales.Adjusted inter-item correlations were satisfactory, and overall, these results are similar to previous evaluations of the PVAQ-9 (Esteve et al., 2013;Pilar Martínez et al., 2014).An advantage of the 8-item version is that it contains an equal number of items per factor, while simultaneously achieving a strong factor structure without undue influence of local item dependencies on the sum score.In other words, each item loads on a factor, and no item is redundant or repeated in a problematic manner.Thus, tentatively, out of two versions with similar psychometric properties, the shorter PVAQ-8 could be considered the first choice when assessing pain vigilance and awareness in chronic pain patients in research and clinical care.
The abbreviation of psychometric instruments can be a challenge.With fewer items, there is a risk that certain aspects or dimensions of a construct are not accurately represented.Ideally, a psychometrically valid full scale is used as the basis for a shortened version, and an effort is made to maintain solid psychometric properties in the brief version.Conceivably, this can be based on a combination of empirical data and feedback from experts and the target population (Boateng et al., 2018;Terwee et al., 2018).However, as is evident from Table 4 and the previous literature, the full 16-item scale of the PVAQ has typically not been found to be a valid scale in the psychometric sense.This means that, in the study described here, development of the PVAQ-8 could not rely as much on parameter targets derived from studies of the PVAQ-16.To shed further light on the existing literature, improve the likelihood that findings would replicate, and ensure that selection of items was anchored in the wider expertise of the field, we based confirmatory factor analysis on previously published solutions for different versions of the PVAQ (Tables 1 and 4).The PVAQ-8 appeared to be a minor improvement of the PVAQ-9, and appeared to capture the intended construct with considerably more promising psychometric properties than the PVAQ-16.
Correlations between the PVAQ-8 and measures of other constructs were largely in line with previous evaluations of various versions of the scale, supporting the construct validity of the Swedish PVAQ-8.Correlations with pain catastrophizing, anxiety, depression, disability, and overall fibromyalgia severity were largely consistent with the hypotheses.It should be noted, however, that comparisons between studies are complicated by differences in samples, language and nationality, number of items retained, and choice of questionnaires.The correlation between the PVAQ-8 and pain catastrophizing (PCS) was moderate and slightly weaker than in earlier studies.Since cognitive processing of pain involves being attentive to painful stimuli, the association between pain catastrophizing and attention to pain appears non-controversial and theoretically reasonable.Associations with disability (WD2-12) were weaker than some previous findings (e.g., McCracken, 1997), but similar to others (e.g., Monticone et al., 2016), and the PVAQ-8 correlated moderately with fibromyalgia severity (FIQ) as seen in Pilar Martínez et al. (2014).Weak correlations were found between the PVAQ-8 and anxiety (GAD-2) and depression (PHQ-9), similar to Pilar Martínez et al. (2014).Although depression and anxiety are common comorbid conditions in fibromyalgia, the construct of attention to pain should theoretically be distinct from depression and anxiety, which is supported by the relatively weak correlations found in our study.It is also possible that correlations with depression, in our study, were weaker due to the exclusion of participants with severe depression (defined as ≥30 on MADRS-S).

| Unexpectedly weak correlation with pain intensity
An unexpected result was that, in this study, no version of the PVAQ correlated significantly with pain intensity as measured using the BPI-SF.In contrast, the correlation with pain intensity as measured using the FIQ pain item was weak but significant, and more in line with previous findings (e.g., Kunz et al., 2017;Monticone et al., 2016;Pilar Martínez et al., 2014).It is not clear why this discrepancy arose.We note that, surprisingly, the BPI-SF and the FIQ pain item correlated only moderately (r = 0.40).One possibility is that either the BPI-SF, which has not been validated in a Swedish population with fibromyalgia, or the FIQ pain, which has not been validated as a separate scale and could be influenced by surrounding items, was not a valid measure of pain intensity in this setting.Unlike scores on the FIQ pain item, scores on the BPI-SF followed a bimodal distribution.Most probably, this was because a score of zero could be assigned either because the respondent scored zero on all pain intensity items, or because they responded "No" to the portal question (in which case the severity items were never administered).Surprisingly, some participants who received a score of zero on the BPI-SF scored high on the FIQ pain item.We can only speculate on potential reasons, including questionnaire fatigue or misunderstanding of the portal question.Correlations with pain intensity as measured by the BPI-SF (r = 0.19), as well as correlations between the BPI-SF and the FIQ pain item (r = 0.65), were still unexpectedly weak when participants with this unexpected response pattern were excluded from the analysis.In addition to the possible role of measurement error, conceivably, pain intensity could be unrelated to attention to pain as measured by the PVAQ in some circumstances.Most probably, vigilance to pain is only partly determined by pain intensity (and vice versa), and this association may vary from person to person, and situation to situation.This, in turn, could lead to variable findings between studies.Regardless, the discrepancies in correlations of the PVAQ with measures of pain intensity warrant further research.

| Limitations
The findings of this study should be considered in light of its limitations.Because the data were derived from a clinical trial, participants were self-recruited based on their interest in receiving a 10-week online cognitive behaviour therapy within a research project.Certain applicants were excluded, such as those with severe depression or other severe comorbidity, those inexperienced with web-enabled devices, and those with insufficient language skills.In addition, the proportion of participants with tertiary education was high (64%), which might limit generalizability.Generalizability could have been improved by recruiting a clinical fibromyalgia sample without the restrictions of eligibility criteria, as well as a clinical sample with various chronic pain conditions.However, the sample in the trial resembles previous trials with consecutive clinic patients regarding clinical characteristics and fibromyalgia symptoms (Bernardy et al., 2013).We were also able to replicate the findings of Esteve et al. (2013) who studied a Spanish sample with chronic back pain and Pilar Martínez et al. (2014) who studied a Spanish sample with fibromyalgia, both supporting a 9-item version of the PVAQ.None of these studies had samples recruited based on RCT eligibility criteria.We find this promising in terms of generalizability because of the similarities in results, despite the Swedish and Spanish languages belonging to different language families.This study involved no evaluation of sensitivity to change.Finally, this study computed associations with the PVAQ and instruments that were available through the clinical trial.This includes two measures of pain intensity that have not been validated in Swedish.Future evaluations of the Swedish PVAQ should investigate associations with different pain instruments.It would be relevant to investigate if the strong factor structure of the 8-item version can be replicated in a clinical fibromyalgia sample, without the aforementioned exclusion criteria.It would also be interesting to investigate whether the Swedish PVAQ-8 would display stronger associations with other instruments to measure pain intensity.Assessments of sensitivity to change and cross-validation in different samples are needed to further evaluate the validity and reliability of the Swedish PVAQ, as well as research to better understand the construct of attention to pain and its potential role in chronic pain conditions.
Abbreviations: AC, awareness of change; ACP, attention to changes in pain; AP, attention to pain; AV, active vigilance; CFA, confirmatory factor analysis; I, Intrusion; M, monitoring; PA, passive awareness; PCA, principal component analysis.
Modification indices (MI) for Model I -V were suggestive of shared covariance for several of the item pairs.For instance, Model I had 40 item pairs with modification indices >3, of which seven pairs ≥15.Model V revealed fewer high modification indices and only two item pairs with ≥15.Item pair 9 and 11 stood out, appearing as the top suggested modification in four of the models, with indices ranging from 26 to 33.Item 9 "I know immediately when pain starts or increases" and item 11 "I know immediately when pain decreases" are similarly worded, and it was seen as theoretically justifiable to modify Model V by adding residual covariance over these items.The resulting Model VI (PVAQ-9) had an acceptable fit: χ 2 (25) = 44; CFI = 0.99; TLI = 0.98; SRMR = 0.04 and RMSEA = 0.05 (90% CI: 0.02-0.08).Last, we evaluated an additional model (Model VII) which excluded item 11, the item in the pair (9 and 11) with the lowest factor loading.The resulting 8-item PVAQ (PVAQ-8) also had an acceptable fit: χ 2 (19) = 32; CFI = 0.99; TLI = 0.98; SRMR = 0.04 and RMSEA = 0.05 with (90% CI: 0.02-0.08).In conclusion, two acceptable factor solutions for the PVAQ were found: one two-factor solution for the 9-item version equivalent to that presented by Pilar Martínez et al. (2014) (model VI) and one two-factor solution for a novel 8-item version (model VII).These factor solutions are presented in Figure 1.
This study aimed to evaluate the psychometric properties of a Swedish translation of the PVAQ in a sample of individuals with fibromyalgia.To our knowledge, this is the first psychometric evaluation of a Swedish version of the PVAQ.Strengths of this study include the large sample size (N = 274), which is strongly preferred F I G U R E 1 Factor solutions for the Pain Vigilance and Awareness Questionnaire 9-item version (PVAQ-9) and 8-item version (PVAQ-8).Abbreviations: PA, passive awareness; AV, active vigilance.
(Esteve et al., 2013;Pilar Martínez et al., 2014)n or unclear occupational status.bIncludesallergy,thyroiddisorders,arthritis,asthma,migraine,hypertension,irritablebowelsyndrome,heartconditions, chronic fatigue syndrome, diabetes, and other somatic conditions.cIncludesdepression,exhaustiondisorder,anxietydisorders,panicdisorder,GeneralizedAnxietyDisorder,post-traumaticstress disorder, attentiondeficit/hyperactivity disorder, and other psychiatric disorders.floororceilingeffectsandalthoughitem10exhibitedaslightdeviation in skewness (1.16) and kurtosis (1.22), item distributions were approximately normal in all other respects.As is presented in TableS1, items 16, 8, 4, and 1 generally showed weak inter-item correlations (≤0.25) and for item 16, no correlations were significant.The Kaiser-Meyer-Olkin (KMO) statistic was 0.91.Results from CFA are presented in Table4.None of the models I -IV achieved acceptable model fit in the current data set.Model V(Esteve et al., 2013;Pilar Martínez et al., 2014)was borderline acceptable with the RMSEA being 0.08 (<0.08 commonly regarded as "fair").T A B L E 3 Item score distributions.Abbreviation: PVAQ, Pain Vigilance and Awareness Questionnaire.aItem 8 and 16 are reverse-keyed when computing the total sum scale.For the mean and median results presented in this table, scores have not been reversed.Scores were reversed for computing adjusted item-total correlations.T A B L E 4Goodness-of-fit indices derived from confirmatory factor analysis of potential factor structures for the Pain Vigilance and Awareness Questionnaire.Note: All χ 2 values were significant, p < 0.05.