Psychosocial factors partially mediate the relationship between mechanical hyperalgesia and self-reported pain

2.1 Participants

Five hundred and sixty-five participants with knee pain were identified from a prospective population-based cohort investigating chronic pain (the epidemiology of functional disorders [EPIFUND] cohort) [17]. Participants were eligible for this study if they responded to a postal survey, had knee pain and consented to further contact. It was not possible to robustly identify whether participants in the EPIFUND cohort identified as having knee pain also had underlying OA.

2.2 Recruitment

A two-phase telephone recruitment strategy was used: eligible participants were first contacted to verify the presence of knee pain on at least 1 day in the past month and to consent to the mailing of the participant information sheet about the QST study for their consideration. The second phone call occurred at least 7 days following the first-call to ensure adequate time for delivery and consideration of the information sheet. A more detailed description of the study was provided during the second phone call and if they were interested, participants were invited to attend a 90 min appointment at a local primary care or research centre. A letter containing details of the study appointment, including the time, date and location, was mailed to each participant who agreed to take part. The present study received approval from the National Research Ethics Service Committee North West – Cheshire (12/NW/0556) in order to contact the cohort and complete the study assessments. All participants provided consent at the study visit prior to any assessments.

2.3 Study assessments

All assessments during the study visit were performed by one rater (KJM).

2.4 Self-reported pain intensity and body mass index

All participants were asked whether they had experienced pain that lasted 1 day or longer in the past 30 days and to indicate that pain on a blank-body manikin. Participants with knee pain were those who shaded one or both knee regions. Global and knee pain intensities were assessed using 0–10 (best to worst) numeric rating scales for the average pain severity experienced within the past 30 days. Body mass index (BMI) was calculated from measured weight (kilograms) and height (metres²).

2.5 Psychosocial factors

Participants were provided with a questionnaire, including items addressing anxiety, depression, illness perceptions, pain catastrophising and physical functioning, with a stamped and addressed envelope to mail back to the study team after the assessments were complete. The hospital anxiety and depression (HAD) scale [18] comprises seven anxiety and seven depression items (items scored 0, no symptoms to 3, strong indication of symptoms; anxiety and depression scales score range 0–21; 0–7 classified as normal; 8–10 as borderline cases and ≥11 as cases). The pain catastrophising scale (PCS) and brief illness perception questionnaire (IPQ-brief) measured cognitions about pain [19]. The PCS comprises 13 items scored 0 (not at all) to 4 (all of the time) forming three sub-scales: helplessness (six items, range 0–24), rumination (four items, range 0–16) and magnification (three items, range 0–12). The IPQ-brief comprises eight items scored using a 10-point numeric rating scale [20]; five items (consequences, timeline, personal control, treatment control and identity) address thoughts about the illness, two items (concern and emotion) address the emotional impact and the final item (coherence) relates to the understanding of the illness (pain in the present study). Physical functioning was addressed using the Rapid Assessment of Physical Activity (RAPA) scale. The RAPA includes nine items (range 0–10) scored “yes” or “no” with seven items for levels of physical activity (0 classified as sedentary; 1–2 as underactive, 3–4 regular underactive and ≥5 as regular active (>5), one item for strength (scored 0 or 1) and one item for flexibility (scored 0 or 2) [21].

2.6 Quantitative sensory testing

The QST assessments were performed following the protocol by the German Research Network on Neuropathic Pain [22]. A reduced QST protocol was used; there is limited evidence for the presence of abnormal mechanical or thermal detection thresholds, or paradoxical heat sensations, in knee OA patients. All other assessments were included as these are considered to be altered in those with knee pain (pressure pain; temporal summation) or the literature is conflicting (thermal and mechanical pain, and vibration detection thresholds). QST assessments, except for vibration detection, were performed at the tibial tuberosity of the most painful knee and 2 cm distal to the lateral epicondyle on the opposite forearm (if pain-free): vibration detection was performed on the nearest bony prominences to the test sites in accordance with the QST protocol (at the patella of the most painful knee and the opposite elbow). Participants who did not achieve a painful sensation during cold or heat pain thresholds were categorised as 0 °C and 50 °C, respectively.

The tender point count was included as a measure of widespread sensitivity to determine whether increased sensitivity to sensory stimuli was localised to the knee or if there was an indication of more generalised changes in pain sensitivity [23]. Participants were provided with standardised descriptions of the QST measures and ratings required; a summary of the test battery is provided in Appendix A in the Supplementary Material.

2.7 Sample size

In the EPIFUND study [24], the standard deviation of tender point count was 4.5 and of global pain intensity was 2.25, and the regression of global pain intensity on tender point count gave a regression coefficient of 0.16. Entering these values in the G*Power 3.1.2 software suggest that a sample size of 71 participants would be sufficient to give 80% power to detect this association at a significance level of 0.05.

2.8 Analysis

The characteristics of study participants, QST measures and psychosocial factors are presented as medians with the interquartile range (IQR) for continuous measures and proportions for categorical variables. Spearman’s correlation was used to determine associations between QST measures and psychosocial factors, and QST measures with both self-reported pain intensities. The QST measures (independent variables) with significant correlations (p<0.05) with one or both measures of self-reported pain intensity (dependent variables) were selected for the mediation model (path c, Fig. 1). Psychosocial factors which were significantly correlated (p<0.05) with the QST measures selected as independent variables were selected as mediating variables for the mediation model (path a, Fig. 1). Structural equation models with all psychosocial factors associated with each QST measure loading on to a latent mediator were also constructed (Fig. 2). Only participants with complete observations for all measures of interest were included in the analyses. All analyses were performed using Stata 13.1 software (Stata, USA).

Fig. 1:

Mediation model. The direct effect is represented by path c between the independent and dependent variables. The indirect path is represented by path a between the independent and mediator variables, and path b between the mediator and dependent variables.

Fig. 2:

Mediation model including latent psychosocial mediating variable. Path c represents the direct effect between the independent and dependent variables. Paths a and b represent the indirect path between the independent and mediator variables (path a), and between the mediator and dependent variables (path b). ε = error term; QST = quantitative sensory testing.

3.1 Participants

Five hundred and sixty-five of 1530 responders (37%; median age 60 years; 62% female) of the EPIFUND cohort reported knee pain; 213 (38%) of those identified with knee pain were contacted during the first telephone recruitment phase (Fig. 3). One hundred and thirty-three people (24%) received a study information sheet and 80 people (14%) either could not be contacted, reported no knee pain within the past month, or declined participation. Ninety-two people (16%) agreed to participate in the study. However, 11 people (2%) did not attend study appointments and eight people (1%) were withdrawn during the study visit. Seventy-two participants (13%) completed the study. Sixty-one participants (11%; median age 64 years; 59% women; Table 1) had complete data for QST measures, psychosocial factors and self-reported pain intensities, and were included in the analyses. The proportion of female EPIFUND responders with knee pain was comparable to the proportion in this study (62% and 59%, respectively), although participants in this study were older than the EPIFUND responders (median age 64 years and 60 years, respectively).

Fig. 3:

Recruitment flowchart.

3.2 Participant characteristics

The median BMI of the 61 participants was 27.7 (IQR 26–30.7; Table 1). The medians for self-reported global pain intensity and pain intensity at the tested knee were 5 (IQR 3–7 for both pain intensities). The median for tender point count was 0 (IQR 0–2) with no participants meeting the 11 tender point threshold outlined in the 1990 ACR Fibromyalgia criteria [23]. The median ratios for temporal summation at the knee and forearm exceeded 1 (2 and 1.9, respectively) indicating greater pain was reported in response to the train of mechanical stimuli compared with the single stimulus. Apart from the median scores for dynamic mechanical allodynia (DMA) (which were identical at both test sites), the median thresholds for cold pain, mechanical pain sensitivity (MPS) and vibration detection were higher at the forearm than the knee. The median thresholds for heat and pressure pain were lower at the forearm than the knee.

The median scores for the HAD anxiety (3; IQR 1–6) and depression (5; IQR 2–8) sub-scales were below the score for borderline cases (0–7 normal; 8–10 borderline case; 11–21 case). In the present study, 12 borderline and seven cases of anxiety, and eight borderline and three cases of depression were identified using the HAD sub-scale cut-offs (Table 2) [18]. Median scores of 3 (IQR 1–2), 1 (IQR 1–3) and 3 (IQR 2–5) were reported for the rumination, magnification and helplessness sub-scales of the PCS, respectively. The median scores for items of the IPQ-brief were: consequences (3; IQR 2–5); timeline (8; IQR 4–10; inversely scored); personal control (5; IQR 3–7; inversely scored); treatment control (5; IQR 2–7; inversely scored); identity (2; IQR 1–4); concern (4; IQR 2–7); coherence (2; IQR 1–5); emotion (3; IQR 1–4). A median score of 4 (regular underactive; IQR 4–5) was observed for the RAPA.

3.3 Correlations between QST measures and self-reported pain intensity

There were significant positive correlations (p<0.05; Table 3) between global pain intensity and number of tender points, knee and forearm MPS and DMA at the knee. Knee and forearm MPS, and DMA at the knee were also significantly positively correlated with higher pain intensity at the tested knee.

3.4 Correlations between QST measures and psychosocial factors

In total, 10 psychosocial factors were significantly correlated (p<0.05; Table 4) with tender point count; seven factors were significantly correlated with knee MPS; nine factors were significantly correlated with DMA at the knee; and five factors were significantly correlated with MPS at the forearm. Impact of illness on life, increased illness duration and higher levels of concern (illness perceptions) and magnification (pain catastrophising) were all significantly positively correlated with tender point count, MPS at the knee and forearm and DMA at the knee.

3.5 Mediation analysis

Nine significant partial mediators of the associations between QST measures and the self-reported global and knee pain intensity measures were identified (Table 5). The total effect of tender point count on global pain intensity was 0.467 (β-coefficient; 95% CI 0.184, 0.749; Table 5). Within the mediation model, the direct effect (path c; Fig. 1) between tender point count and global pain intensity was non-significant (β 0.467; 95% CI −0.086, 0.462). The indirect effect (path a×path b; Fig. 1; β 0.279; 95% CI 0.093, 0.465) between tender point count and concern item of the IPQ-brief (path a), and between concern and global pain intensity (path b) was significant. The proportion of the total mediated effect was determined by dividing the β-coefficient for the indirect effect by the β-coefficient for the total effect (0.279/0.467=60%).

The association between increased number of tender points and increased global pain intensity was also significantly partially mediated by the consequences item of the IPQ-brief, and the helplessness and rumination sub-scales of the PCS explaining 57%, 56% and 34% of the total effect, respectively.

Increased knee MPS and increased global pain intensity was partially mediated by concern, consequences and helplessness explaining 45%, 40% and 40% of the total effect, respectively, and increased knee MPS and increased pain intensity at the tested knee were partially mediated by consequences and concern, explaining 30% and 29% of the total effect, respectively.

The inclusion of psychosocial factors loaded on to a latent mediator rather than individual items (Fig. 2) accounted for 75%, 52%, 63% and 35% of the total effect of tender point, knee MPS and knee DMA on global pain intensity, and knee MPS on knee pain intensity, respectively (Table 6). However, the latent psychosocial mediator was not a partial mediator of the association between knee DMA on knee pain intensity (30% total effect mediated; Table 6).