The Central Aspects of Pain in the Knee (CAP-Knee) questionnaire; a mixed-methods study of a self-report instrument for assessing central mechanisms in people with knee pain

Summary Objectives Pain is the prevailing symptom of knee osteoarthritis. Central sensitisation creates discordance between pain and joint pathology. We previously reported a Central Pain Mechanisms trait derived from eight discrete characteristics: Neuropathic-like pain, Fatigue, Cognitive-impact, Catastrophising, Anxiety, Sleep disturbance, Depression, and Pain distribution. We here validate and show that an 8-item questionnaire, Central Aspects of Pain in the Knee (CAP-Knee) is associated both with sensory- and affective- components of knee pain severity. Methods Participants with knee pain were recruited from the Investigating Musculoskeletal Health and Wellbeing study in the East Midlands, UK. CAP-Knee items were refined following cognitive interviews. Psychometric properties were assessed in 250 participants using Rasch-, and factor-analysis, and Cronbach's alpha. Intra-class correlation coefficients tested repeatability. Associations between CAP-Knee and McGill Pain questionnaire pain severity scores were assessed using linear regression. Results CAP-Knee targeted the knee pain sample well. Cognitive interviews indicated that participants interpreted CAP-Knee items in diverse ways, which aligned to their intended meanings. Fit to the Rasch model was optimised by rescoring each item, producing a summated score from 0 to 16. Internal consistency was acceptable (Cronbach's alpha = 0.74) and test–retest reliability was excellent (ICC2,1 = 0.91). Each CAP-Knee item contributed uniquely to one discrete ‘Central Mechanisms trait’ factor. High CAP-Knee scores associated with worse overall knee pain intensity, and with each of sensory- and affective- McGill Pain Questionnaire scores. Conclusion CAP-Knee is a simple and valid self-report questionnaire, which measures a single ‘Central Mechanisms’ trait, and may help identify and target centrally-acting treatments aiming to reduce the burden of knee pain.


Introduction
In individuals with knee osteoarthritis (OA), joint pathology only weakly predicts the pain severity 1 , and up to 20% of people with severe OA report unsatisfactory pain relief following arthroplasty that replaces the affected joint 2 . Central pain mechanisms act through the central nervous system to augment knee pain severity 3 , and predict poor outcome following peripherally targeted treatments such as arthroplasty 4 . Central sensitisation (CS) is one such central mechanism, defined as 'increased responsiveness of nociceptive neurons in the central nervous system to their normal or subthreshold afferent input' 5 . Pain is a sensory and emotional experience, and not only a neuronal response. Therefore central factors other than nociceptive neuronal activity might augment pain, including changes in brain connectivity or appraisal. The extent to which central mechanisms augment knee pain varies between individuals. Treatments that target central pain mechanisms can reduce knee pain severity 6 , and might improve response to peripherally targeted treatments 7 .
Quantitative sensory testing (QST) or magnetic resonance imaging (MRI) approaches assess complex pathophysiological central mechanisms such as CS or other higher brain mechanisms linked to central pain mechanisms 5 . Self-report questionnaires also assess manifestations linked to central mechanisms (including psychological and somatic disturbances), and are less resource-intensive and feasible during normal clinical encounters. However, a comprehensive, validated mechanism-based self-report questionnaire is not currently available to identify people with knee pain that is importantly augmented by central pain mechanisms.
Responses to questionnaires addressing individual characteristics of anxiety, depression, catastrophizing, cognitive-impact, sleep, fatigue, pain distribution and neuropathic-like pain have each been associated with experimental markers of CS (including QST and neuroimaging biomarkers), and with pain outcomes in people with knee pain 2,8e12 , or some other chronically painful conditions 7,13 . We previously selected 8 self-report items, each representing one of these eight characteristics based on expert opinion (at least moderate agreement between experts on relevance to central pain mechanisms in knee OA), strength of association with the scales measuring the component characteristic, and with Pressure Pain detection Threshold (PPT) at a site distal to the affected knee (taken as an indicator of CS) 9 . We showed by factor analytic approaches, that these eight items contributed to a single underlying trait 9 . This 'Central Mechanisms' trait was associated with PPT at a distal site, and with knee pain severity more strongly than was any one of the eight characteristics alone 9 , and was a better predictor of future knee pain than other clinical predictors 8 . Existing questionnaires, including the short-form Central Sensitization Inventory and the Generalized Pain Questionnaire which were originally designed to assess central sensitivity syndromes such as fibromyalgia 14,15 omit important characteristics that have been associated with QST-evidence of central pain mechanisms in people with knee pain. These include neuropathic-like pain qualities and cognitiveimpact 9,12,16e19 .
Clinical decision support tools have potential to greatly enhance a clinician's ability to carefully subgroup patients, and inform mechanism-based treatment allocation in primary care settings. Employing such tools could inform choice of tailored stratified treatment pathways for individuals with knee OA pain. However, such tools are not widely available for widespread use in clinical practice.
We therefore have developed a simple questionnaire (Central Aspects of Pain in the Knee; CAP-Knee) comprising self-report items measuring each of the eight characteristics which contribute to the Central Mechanisms trait in people with knee pain. We hypothesised that this Central Mechanism trait is associated with both sensory and emotional components of knee pain. We show that CAP-Knee is interpreted appropriately, psychometrically valid and reliable, and associated with sensory and affective components of pain severity in people with knee pain.

Study design
A mixed methods, cross-sectional observational study combining thematic analysis of interviews with people with knee pain, with quantitative analysis of questionnaire psychometric properties and external validity.

Participants and recruitment
Eligible participants were aged !40 years, reported knee pain on most days for the past month, provided informed consent, and could communicate in English. Participants were excluded if they reported having an inflammatory arthritis such as Rheumatoid Arthritis. Details of participant recruitment are provided in Fig. 1.
Small samples (5e15 participants) may fail to uncover common problems in questionnaire evaluation studies, thus a default sample size of 30 participants is endorsed for cognitive interviews 20e22 . Recruitment ended when data saturation had been reached, and the research team confirmed when no new themes emerged.
All participants contributing to Rasch or repeatability analysis reported that their knee was their most bothersome joint pain, and that their knee pain had lasted !4 weeks. Participants were recruited from those consenting to further contact in the Investigating Musculoskeletal Health and Wellbeing (IMH&W) study (clinicaltrials.gov NCT03696134) 23 . For Rasch analysis, data from 250 to 500 participants may provide accurate and stable person and item estimates, as well as a good balance for statistical interpretation of the fit statistics 24 . Additionally, it is recommended that a minimum sample size for repeatability studies should be 50 25 . Thus, for the Rasch study and the nested repeatability study, complete data for 250 participants were included.
IMH&W participants were recruited from GP surgeries, or a database of participants in previous studies undertaken by the University of Nottingham who had consented to be contacted with information about future research. Ethical approval for interviews was obtained from the Nottingham Research Ethics Committee 2 (17/EM/0480), and for nested recruitment and questionnaire analysis from the London Central Research Ethics Committee (18/ LO/0870).

Self-report questionnaire assessment
Data for age, sex, weight and height, physician-diagnosed conditions, CAP-Knee questionnaire, and question groups 1e20 from the McGill Pain Questionnaire 26 were collected by self-report from all participants. Weight and height were used to calculate body mass index (BMI) 27 . Participants were asked to state the joint with the most bothersome pain, and to rate its severity on an 11-point Likert scale ('over the past 4 weeks, how intense was your average pain or the average aching feeling in your most bothersome joint, where 0 is 'no pain' and 10 is 'pain as bad as could be'?). In the current study, items originally selected from a larger item pool used in the Knee Pain in the Community study 28 were reworded and response options modified to provide a consistent structure and response format across the newly developed questionnaire. The final version of the CAP-Knee following item refinement is shown in Fig. 2.
The CAP-Knee questionnaire comprises eight items which assess each of the eight characteristics linked to Central Mechanisms of knee pain. Each of the items originated from established questionnaires measuring Neuropathic-like pain29, Fatigue 30 , Cognitive impact 31 , Catastrophizing 32 , Anxiety and Depression 33 , Sleep disturbance 34 , and Pain distribution 35 .

Cognitive interviews
KAA conducted all cognitive interviews at the University of Nottingham. Eligibility was confirmed by asking participants "Have you had knee pain on most days of the last month?", and knee OA clinical classification used American College of Rheumatology (ACR) criteria 36 . Participants completed the CAP-Knee questionnaire immediately before the interview.
Interviews used probing techniques that were standardised according to an interview guide (Supplement 1). To elicit difficulties or comments related to interpretation of questionnaire items 37 , probes were developed for each of the four-stage question response model, including stages of comprehension, retrieval, judgement and response formulation 38 . Interviews were audio-recorded, transcribed verbatim using a transcription service, and transcription accuracy was checked against recordings.
The full text of each transcript was coded and analysed with the aid of the NVivo 12 qualitative software programme 39 . Transcripts were iteratively analysed in groups of five or six until data saturation was achieved. Throughout qualitative analyses, disagreements or questions were discussed and interpretations were validated within the research team (KAA, DAW, RdN, EF), to minimise the influence of researcher subjectivity and preconceptions on identifying potential themes 40 . Following assessment of the original version of the CAP-Knee ( Supplementary Fig. 1), a further wave of five cognitive interviews after revision of CAP-Knee items was undertaken using the above procedures.  Recruitment flowchart for the quantitative and qualitative assessment sub-studies.þ39 out of 50 individuals responding to study between December 2017 and June 2018 consented to further contact and were therefore screened for study eligibility. *489 people who had provided baseline questionnaire data to the IMHW study between May 2018 and December 2018 were screened for inclusion, of whom 250, each registered with one of 32 different general practices within the East Midlands region, met the eligibility criteria for this study. # Data from the first 250 participants who completed the CAP-Knee within the IMW&H study and met the inclusion criteria for psychometric assessment of the CAP-Knee were assessed. To test repeatability, 193 of the 250 participants were mailed within 7-days of receipt of their questionnaires with an invitation to complete the CAP-Knee questionnaire a second time. 105 of the 193 participants completed and returned the CAP-Knee questionnaire a second time.

Fig. 2
Recruitment flowchart for the quantitative and qualitative assessment sub-studies.+39 out of 50 individuals responding to study between December 2017 and June 2018 consented to further contact and were therefore screened for study eligibility. *489 people who had provided baseline questionnaire data to the IMHW study between May 2018 and December 2018 were screened for inclusion, of whom 250, each registered with one of 32 different general practices within the East Midlands region, met the eligibility criteria for this study. # Data from the first 250 participants who completed the CAP-Knee within the IMW&H study and met the inclusion criteria for psychometric assessment of the CAP-Knee were assessed. To test repeatability, 193 of the 250 participants were mailed within 7-days of receipt of their questionnaires with an invitation to complete the CAP-Knee questionnaire a second time. 105 of the 193 participants completed and returned the CAP-Knee questionnaire a second time.

Osteoarthritis andCartilage
Content analysis adopted a summative approach seeking to assign codes and sub-codes across participant responses within each transcript, following Tourangeau's question response model 37 . In order to guide item revision, problems were identified based on sub-codes of each main code of comprehension (completely-, partially or not aligned to the researchers' intended meaning of each item), retrieval (no-, partial-and completeretrieval difficulty), judgement (certain initial or uncertain initial response) and response formulation (consistent or inconsistent)supplement 2. One researcher (KAA) categorised each participant based on quotes contributing to sub-codes across each item 41 . An item was revised if (i) emerging themes were not aligned with researcher's interpretation, and (ii) the item was not well understood by a majority of participants (including complete nonalignment, complete retrieval difficulty, uncertain initial response and no response consistency) 42 .
Thematic analyses also informed item revision by assessing the range of interpretations provided by participants for each CAP-Knee item. An inductive approach to thematic analysis was used to identify themes that were exhaustive and mutually exclusive. Analysis of the transcripts followed a sequential process 43 allowing the researcher to search for themes and patterns across the entire dataset, rather than within participant interviews. Themes were extracted according to DeSantis and Ugarriza 44 . Identified themes were reduced or expanded to account for various interpretations, and meaning was attributed to the interpretation and experiences expressed by participants about each item 45 .

Statistical analysis
Based on our previous study 9 and prior to Rasch analysis, CAP-Knee items one to seven were scored 0 to three according to responses 'never', 'sometimes', 'often', or 'always'. Item eight was allocated a binary score of '0 0 or '3' based on pain reported at one knee, plus any other additional region below the waist, by shading on the pain distribution manikin9. All subsequent analyses used Rasch transformed scores according to findings made in the current study.

Rasch analysis
Rasch analysis was performed as described in a previous study 46 , using the Test Analysis Modules (TAM) package within the R programmer software 47 . Differential item functioning (DIF) was explored for sex (males and females) and age (<64 years, 64e71 years and over 71 years), using the ManteleHaenszel test. Residual Mean Square (MNSQ) values between 0.7 and 1.3 were deemed acceptable fit residuals for each item response category 48 . Where disordered response thresholds were observed, items were re-scored by collapsing appropriate adjacent response options 49,50 . The person separation index (PSI) was calculated to estimate measurement reliability of CAP-Knee, with PSI¼ >0.70 set as the cut-off for reliability 51,52 .
Confirmatory factor analysis (CFA) and floor/ceiling effects To determine whether CAP-Knee data fit a one-factor model, CFA within a structural Equation Modelling (SEM) framework used MPlus version 7.4 53 as previously described 9 . Floor and ceiling effects were considered present if >15% of respondents achieved the lowest/highest possible scores (50).  1. Neuropathic-like pain ('Cold or heat touching my knee was painful')* 10 (59%) Thermal allodynia (n ¼ 12) e Weather induced pain (n ¼ 9)y Thermotherapy (n ¼ 5)y Item one revised: Neuropathic-like pain ('Cold or heat (e.g., bath water) on my knee was painful ')*

Reliability and associations with knee pain
Reliability was assessed using Stata 14.2 54 . Item-redundancy and internal-consistency were investigated by calculating Cronbach's alpha (a), with values < 0.70 indicating poor internal consistency, and values >0.90 indicating item redundancy 55 . Testeretest reliability was assessed by Intra-class Correlation Coefficient (ICC 2, 1 ) 56 .
MPQ responses were summed to give a single score for knee pain intensity, and summated ranks of descriptive terms within groups 1e10 and 11e15 were used to derive sensory and affective pain subscales respectively 26 . Associations between CAP-Knee scores and MPQ total scores for knee pain severity were calculated using Pearson correlation coefficients, and between CAP-Knee and McGill Pain Questionnaire scores using Spearmans rho. Multivariable linear regressions were used for associations between CAP-Knee and pain scales, and were adjusted for age, BMI and sex.

CAP-knee items interpretation
Twenty-two cognitive interviews were conducted (participant characteristics in Table I), comprising four transcript groups, at which point data saturation was achieved. Mean interview duration was 29 min (range 16e57 min). Twenty-one participants (95%) fulfilled ACR clinical classification criteria for knee OA, of whom 11 (52%) had bilateral knee OA. Participant interpretation of all items in the final CAP-Knee questionnaire aligned well with their intended meanings (Table II), although participants interpreted each item according to multiple constructs. Table III provides example quotes for each theme and sub-theme.
Fatigue and sleep items were interpreted as discrete constructs by most participants, although five attributed fatigue to sleep disturbance. Most (n ¼ 13) participants described the fatigue item in terms of physical exertion as a source of fatigue. Seven participants attributed fatigue to other causes (older age, medical problems including thyroid problems, diabetes, and fibromyalgia, or medication). Fatigue was also described in terms of a need to rest in order to obtain relief (n ¼ 8). Sleep disturbance was related to knee pain or discomfort (n ¼ 16) while trying to sleep, but also to the impact of other painful sites (n ¼ 7). Participants often described their sleep disturbance by reference to aids that they used to help them sleep (n ¼ 8), including pharmacological analgesics or sleeping pills, or non-pharmacological aids such as cushions placed between their knees. The cognitive item was described in terms of task distraction (n ¼ 10) or hypervigilance (n ¼ 12). The catastrophizing item was interpreted in terms of causes and consequences of knee pain (n ¼ 11), or of avoidance behaviours (n ¼ 9).
The anxiety item was related to three subthemes, fear for the integrity of their knee (n ¼ 7), fear of falling over (n ¼ 6), and fear of future disability (n ¼ 3). The depression item was related to aspects of both social (n ¼ 11) and physical (n ¼ 14) functioning that participants no longer enjoyed.
The pain distribution manikin was variously interpreted by participants, who indicated different qualities or diagnoses for pain at different sites and at different times (n ¼ 14), or indicated sites of pain for which they had sought professional help (n ¼ 9), or which had different impacts on their life (n ¼ 5).
The neuropathic-like pain item in the developmental questionnaire was not well understood by 59% of participants (Table II). Participants referred to weather-induced pain or thermotherapy which were not aligned with the intended meaning of thermal allodynia. Single participant quotes are provide for each theme/subtheme.

Table III
Participant quotes supporting themes and subthemes identified for each item included within the CAP-Knee Scale Osteoarthritis andCartilage ( Fig. 1) references an example of tangible physical stimulus: 'Cold or heat (e.g., bath water) on my knee was painful', and was understood well by all participants.

Rasch modelling, scale and item responses
Baseline characteristics for study participants across the interview and Rasch/repeatability studies are shown in Table I. Likelihood ratio test showed significant differences (p< 0.001) between the partial credit formulation and the rating scale model. The partial credit formulation therefore was used, thereby avoiding assumption of uniform threshold distance across all items.
The summary item-fit statistics from analyses of CAP-Knee scores with four response levels per item indicated misfit to the Rasch model, with significant itemetrait interaction [X 2 (df) ¼ 63 (28); p< 0.001] ( Table IV). The cognitive-impact item showed misfit (Table V). CAP-Knee was well targeted, but the sleep disturbance item exhibited disordering of the step difficulty (i.e., the difficulty of a higher step was lower than that of its adjacent lower step), displaying a disordered response threshold (Supplementary Fig. 2). Principal Components Analysis of the residuals found that all eight items loaded on the first component. Eleven (4% [Binomial CI: 2e8%]) of 246 t-tests were significant (Table IV). Four items (neuropathic-like pain, fatigue, anxiety and depression) showed misfit for outfit values in one or more response options (Table V). CAP-Knee item residuals demonstrated no correlations (r < 0.3) between items. None of the items exhibited non-uniform DIF for age or sex. None of the items showed uniform-DIF for age, however, the pain distribution item showed uniform-DIF (P ¼ 0.03) for sex.
CAP-Knee analysis was modified by collapsing response categories until the thresholds demonstrated sequential levels of severity. Merging the highest categories 3 ('Always') and 2 ('Often') for the first seven items produced a scale that performed better than using four response levels (Tables IV and V, Fig. 1). The final scale, in which each item was allocated a score from 0 to two using collapsed respones (possible total score range; 0 to 16) was unidimensional, with no local item dependency observed, although the pain distribution item still showed uniform-DIF for sex. The Rasch-transformed CAP-Knee scoring regime is shown in Supplement 3.

Confirmatory factor analysis and reliability
Factor analysis confirmed the one-factor model (comparative fit index ¼ 0.98; Tucker Lewis Index ¼ 0.97; model X 2 (df) ¼ 38 (20); root mean square error of approximation ¼ 0.08). All eight items loaded significantly on to the single latent factor, which we named 'Central Mechanisms' (Table VI). Cronbach's alpha was 0.74. Of the respondents, one (0.4%) had the minimum CAP-Knee score of 0, and two (0.8%) had the maximum score of 16. Pairs of complete questionnaires were obtained from each of 76 of 105 participants who completed the CAP-Knee the second time for the reliability study (Table I). The median interval between the first and second assessments was 20 days (IQR 17e24 days). The reliability intra-class correlation coefficient (ICC 2, 1 ) was 0.91 (95% CI 0.86 to 0.94).

Discussion
We show that a Central Mechanisms trait, measured by a psychometrically validated CAP-Knee questionnaire was associated both with sensory and affective components of knee pain 8,9 . CAP-Knee items were understood well by people with knee pain, aligned to our mechanistic model of central pain augmentation. CAP-Knee scores were strongly associated with worse knee pain, consistent with contributions from central mechanisms to both sensory and affective pain components. CAP-Knee is suitable for use in research measuring the Central Mechanisms trait to develop, target and evaluate treatments that aim to reduce the burden of knee pain by reducing central pain augmentation.
The current study confirms our previous findings in another participant cohort that items addressing each of eight characteristics of neuropathic-like pain, fatigue, cognitive-impact, catastrophizing, anxiety, sleep disturbance, depression, and pain distribution contribute to a single 'Central Mechanisms' trait 9 . We show here that the Central Mechanisms trait as measured by CAP-Knee is associated with each of the sensory and affective subscales of the MPQ. This supports an integrative model of knee pain in which central mechanisms contribute both to its severity and to its psychological impact. We have also identified heterogeneous participant interpretations of CAP-Knee items, although interpretations aligned well to our mechanistic model of central pain augmentation. Each of these participant-centred interpretations might be indicative of central pain mechanisms, and each also reflects an area of concern to people with knee pain. These findings highlight the diversity and impact of knee pain within the study population.
Rasch analyses demonstrated that respondents might have difficulty distinguishing between response categories 'often' and 'always'. Scoring transformation that collapsed these responses

Table VI
Item loading for Raschtransformed CAP-Knee scores Osteoarthritis andCartilage (each scored 2) resolved disordered thresholds, improved the Rasch properties of the CAP-Knee, and met the Rasch assumption for unidimensionality of the latent trait of Central Mechanisms measured by the CAP-Knee, without otherwise impairing CAP-Knee's psychometric properties. We therefore suggest use of these Rasch transformed scores in future studies to aid longitudinal tracking of the Central Mechanisms trait, or as an outcome measure using parametric analyses. The good measurement properties of CAP-Knee demonstrated here are prerequisite for its use as a mechanistic outcome tool measuring the ability of an intervention to reduce central pain mechanisms.
Our current study is subject to some limitations. Characteristics additional to those eight measured by CAP-Knee, some as yet unidentified, might additionally contribute to central pain mechanisms. Our study might be affected by sources of bias. Study participants were recruited through the IMH&W survey study, which itself recruited from a range of sources in order to optimize the generalizability of our findings. However, different participant samples might have given different results. CAP-Knee was completed following postal receipt by participants, a key advantage if to be used in epidemiological research or surveys.
While previous work demonstrates that the Central Mechanisms trait performs fairly across age, sex and BMI groups, the current work identified DIF by sex for the pain distribution item. This finding reflects previous work which identified women are more likely to report greater widespread pain 57,58 . The pain distribution item contributed least to the Central Mechanisms trait identified in this study. In both our current and previous study 9 , the distribution item loaded less strongly to the Central Mechanisms trait factor than did other items, suggesting that pain distribution might less reliably indicate central pain mechanisms than do other characteristics. However, based on our earlier data and an extensive literature indicating that widespread pain, as seen in fibromyalgia, is associated with central sensitisation, we recommend retention of the 'Pain Distribution' item within CAP-Knee.
Ours is a cross-sectional investigation, and future longitudinal research should determine whether CAP-Knee equals or improves upon outcome prediction by other measures of central pain mechanisms, such as QST 59 . CAP-Knee might also help identify areas of concern for people with knee pain, but does not include items addressing characteristics of additional concern to patients (e.g., swelling or stiffness) that were not linked to central pain mechanisms. CAP-Knee might be useful as an outcome measure to evaluate treatments aiming to reduce central pain mechanisms, but was not designed as a patient-centred outcome measure. Application of CAP-Knee as a stratification tool would require future work to determine clinically relevant cut-off scores, for example, to determine subgroups with knee pain who will have poor pain prognosis, or respond to a drug or psychological intervention designed to modify central pain mechanisms. CAP-Knee would require further validation before use in conditions other than knee pain. Combination of CAP-Knee with other predictors of poor pain prognosis might best identify people most likely to benefit from interventions which target central pain mechanisms 8 .
In conclusion, our findings indicate the heterogeneous nature and impact of knee pain. We describe and validate the CAP-Knee questionnaire, suitable for measuring the Central Mechanisms trait in people with knee pain. We have demonstrated construct validity of the CAP-Knee as a measure of the Central Mechanisms trait, and shown convergent validity of the CAP-Knee with existing sensory and affective measures. Eight self-report items are related to eight discrete characteristics of Neuropathic-like pain, Fatigue, Cognitive-impact, Catastrophising, Anxiety, Sleep disturbance, Depression, and Pain distribution. CAP-Knee is short, easy and inexpensive to administer, either by post or face:face, and was easily understood by people with knee pain. CAP-Knee is suitable for further development as a stratification tool to identify people who might most benefit from new or established treatments designed to reduce augmentation of knee pain by the central nervous system.

Funding
This work was supported by the University of Nottingham acting as Sponsor and host institution. This research was co-funded by Pain Centre Versus Arthritis (Centre initiative grant number 20777) and by the NIHR Nottingham Biomedical Research Centre. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.
Role of the funding source Versus Arthritis UK (Centre initiative grant number ¼ 20777), and University of Nottingham as sponsor and host institution. Study sponsors had no role in the design and conduct of the study; collection, management, analysis and interpretation of data; and preparation, review or approval of manuscript.