Article Text

Download PDFPDF
Population characteristics as important contextual factors in rheumatological trials: an exploratory meta-epidemiological study from an OMERACT Working Group
  1. Sabrina Mai Nielsen1,2,
  2. Helene Storgaard1,3,
  3. Torkell Ellingsen2,
  4. Beverley J Shea4,
  5. George A Wells5,
  6. Vivian Andrea Welch5,6,
  7. Daniel E Furst7,8,9,
  8. Maarten de Wit10,
  9. Marieke Voshaar11,
  10. Carsten Bogh Juhl3,12,
  11. Maarten Boers13,
  12. Reuben Escorpizo14,15,
  13. Thasia G Woodworth7,
  14. Annelies Boonen16,17,
  15. Henning Bliddal1,
  16. Lyn M March18,
  17. Peter Tugwell19,20,
  18. Robin Christensen1,2
  1. 1 Musculoskeletal Statistics Unit, The Parker Institute, Bispebjerg and Frederiksberg Hospital, University of Copenhagen, Copenhagen, Denmark
  2. 2 Research Unit of Rheumatology, Department of Clinical Research, University of Southern Denmark, Odense University Hospital, Odense, Denmark
  3. 3 Department of Sports Science and Clinical Biomechanics, University of Southern Denmark, Odense, Denmark
  4. 4 Ottawa Hospital Research Institute, and School of Epidemiology and Public Health, Faculty of Medicine, University of Ottawa, Ottawa, Ontario, Canada
  5. 5 School of Epidemiology and Public Health, Faculty of Medicine, University of Ottawa, Ottawa, Ontario, Canada
  6. 6 Bruyere Research Institute, Ottawa, Ontario, Canada
  7. 7 David Geffen School of Medicine, Division of Rheumatology, UCLA, Los Angeles, California, USA
  8. 8 University of Washington, Seattle, Washington, USA
  9. 9 University of Florence, Florence, Italy
  10. 10 OMERACT Patient Research Partner, Zaltbommel, The Netherlands
  11. 11 Department Psychology, Health and Technology, University of Twente, Twente, The Netherlands
  12. 12 Department of Physiotherapy and Occupational Therapy, Copenhagen University Hospital, Herlev & Gentofte, Denmark
  13. 13 Department of Epidemiology & Biostatistics, Amsterdam Rheumatology and Immunology Center, Amsterdam University Medical Centers, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands
  14. 14 Department of Rehabilitation and Movement Science, College of Nursing and Health Sciences, University of Vermont, Burlington, Vermont, USA
  15. 15 Swiss Paraplegic Research, Nottwil, Switzerland
  16. 16 Department of Internal Medicine, Division of Rheumatology, Maastricht University Medical Centre+, Maastricht, The Netherlands
  17. 17 Care and Public Health Research Institute (CAPHRI), 6229 ER Maastricht University, Maastricht, The Netherlands
  18. 18 Florance and Cope Professorial Department of Rheumatology, Royal North Shore Hospital and Institute of Bone and Joint Research, Faculty of Medicine and Health, University of Sydney, Sydney, New South Wales, Australia
  19. 19 Division of Rheumatology, Department of Medicine, and School of Epidemiology and Public Health, Faculty of Medicine, University of Ottawa, Ottawa, Ontario, Canada
  20. 20 Clinical Epidemiology Program, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada
  1. Correspondence to Ms Sabrina Mai Nielsen, Musculoskeletal Statistics Unit, The Parker Institute, Bispebjerg and Frederiksberg Hospital, University of Copenhagen, Copenhagen DK-2000, Denmark; sabrina.mai.nielsen{at}


Objectives To explore whether trial population characteristics modify treatment responses across various interventions, comparators and rheumatic conditions.

Methods In this meta-epidemiological study, we included trials from systematic reviews available from the Cochrane Musculoskeletal Group published up to 23 April 2019 in Cochrane Library with meta-analyses of five or more randomised controlled trials (RCTs) published from year 2000. From trial reports, we extracted data on 20 population characteristics. For characteristics with sufficient data (ie, available for ≥2/3 of the trials), we performed multilevel meta-epidemiological analyses.

Results We identified 19 eligible systematic reviews contributing 187 RCTs (212 comparisons). Only age and sex were explicitly reported in ≥2/3 of the trials. Using information about the country of the trials led to sufficient data for five further characteristics, that is, 7 out of 20 (35%) protocolised characteristics were analysed. The meta-regressions showed effect modification by economic status, place of residence, and, nearly, from healthcare system (explaining 4.8%, 0.9% and 1.5% of the between-trial variation, respectively). No effect modification was demonstrated from age, sex, patient education/health literacy or predominant religion.

Conclusions This study demonstrates the scarce reporting of most population characteristics, hampering investigation of their impact with meta-research. Our sparse results suggest that place of residence (ie, continent of the trial), economic status (based on World Bank classifications) and healthcare system (based on WHO index for health system performance) may be important in explaining the variation in treatment response across trials. There is an urgent need for consistent reporting of important population characteristics in trials.

PROSPERO registration number CRD42019127642

  • arthritis
  • epidemiology
  • outcomes research

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Key messages

What is already known about this subject?

  • Concerns have been raised regarding the value of an overall treatment effect estimate from a clinical trial for subsequent decisions on what treatment is best for an individual patient.

  • More evidence is needed to understand how treatment effects vary between patients, and whether patient characteristics are associated with benefit or harm.

  • No one has previously investigated the impact of population characteristics within rheumatology in general.

What does this study add?

  • This meta-epidemiological study of 19 Cochrane reviews (contributing 187 RCTs, that is, 212 comparisons) demonstrates the scarce reporting of most population characteristics, hampering investigation of their impact with meta-research.

  • Our sparse results suggest that place of residence (ie, continent of the trial), economic status (based on World Bank classifications) and healthcare system (based on WHO index for health system performance) may be important in explaining the variation in treatmentresponse across rheumatological trials.

How might this impact on clinical practice or future developments?

  • There is an urgent need for consistent reporting of population characteristics in trials, which requires consensus on a list of potentially important contextual factors (characteristics) of high priority, and how to report their impact on the trial results.


Evidence-based medicine is the main paradigm for shaping clinical guidelines, and, hence, clinical practice. However, concerns have been raised over the value of an overall treatment effect estimate from a clinical trial for subsequent decisions on what treatment is best for an individual patient.1–3 More evidence is needed to understand how treatment effects vary between patients, and whether patient characteristics are associated with benefit or harm. Such characteristics may be referred to as contextual factors, including certain effect modifiers (and/or predictors), that is, in trials, treatment effects may vary according to certain baseline characteristics, and the effort to identify such factors has been termed stratified medicine4 and personalised medicine.5

In addition to improving healthcare for individuals, studying contextual factors can reveal optimal treatment options for certain social groups and thereby reduce health inequity, defined as the presence of unfair and avoidable differences in health between populations.6 Driven by global interest and the identification of suboptimal reporting,7 Welch et al developed new standards for reporting on health equity in both randomised controlled trials (RCTs)8 and systematic reviews9 based on a broad range of factors of social inequity, represented by the acronym PROGRESS-Plus. The PROGRESS factors comprise place of residence, race/ethnicity/culture/language, occupation, sex/gender, religion, education, socioeconomic status, social capital, while ‘-Plus’ refers to additional personal characteristics associated with discriminations (eg, age, disability), features of relationships (eg, smoking parents, excluded from school) and time-dependent relationships (eg, leaving the hospital, respite care, other instances where a person may be temporarily at a disadvantage).10

In 2014, the international research network, ‘Outcome Measures in Rheumatology’ (OMERACT) formed the Contextual Factor Working Group (CFWG) to develop guidance on how to address contextual factors when developing core outcome sets for use in studies of rheumatological interventions. In OMERACT, a contextual factor is defined as a ‘variable that is not an outcome of the study, but needs to be recognised (and measured) to understand the study results. This includes potential confounders and effect modifiers’.11 The working group anticipates developing a generic set (ie, important across rheumatic conditions) of important contextual factors that should always be taken into account in rheumatological clinical trials. In 2018, the working group developed a provisional list of the seven most important contextual factor domains: sex/gender, comorbidities, healthcare system, psychological well-being, adherence to treatment, age and previous exposure to drugs.12

An importance of these factors must be confirmed by evidence from RCTs. However, current RCT and evidence synthesis reports lack detail in the description of important population characteristics, and rarely contain analyses stratified for such characteristics.13–16 To explore which contextual factors may be important across health conditions and interventions, ‘meta- epidemiology’ is a popular method.17 18 This method is typically characterised by the inclusion of a large collection of trials to explore the influence of trial characteristics on treatment effect estimates.17 Previous meta-epidemiological studies within rheumatology have investigated the effect of study quality (internal validity)19–22 and the effect of patient baseline characteristics on the benefit of targeted therapies for rheumatoid arthritis (RA)23 and psoriatic disease.24 To our knowledge, no one has previously investigated the impact of population characteristics within rheumatology in general. The objective of this study was to explore whether trial population characteristics are effect modifiers of treatment response across various experimental interventions and comparators for most rheumatic and musculoskeletal conditions (ie, conditions studied by the Cochrane Musculoskeletal Group (CMSG)). Our assumption was that population characteristics that act as effect modifiers on a meta-epidemiological level will likely qualify as being important contextual factor domains for individual trials.


We performed a meta-epidemiological study. Such studies assess the association between trial characteristics and treatment effect estimates based on large collections of RCTs.17 For this study, we focused on RCTs from Cochrane reviews by the CMSG; the CMSG is a review group in the Cochrane Collaboration that prepares and maintains high-quality systematic reviews for rheumatic and musculoskeletal conditions. We submitted a protocol to the International Prospective Register of Systematic Reviews (PROSPERO) before initiating the study (CRD42019127642; see the protocol in online supplementary file 1).

Eligibility criteria

Initially, we included systematic reviews examining any type of intervention versus any comparator in adults with rheumatic and musculoskeletal conditions (ie, conditions studied by the CMSG). If the systematic reviews did not explicitly state that they included adults (often described as at least 16 or 18 years), we decided to accept an age limit of 16 years or older in their inclusion criteria. We excluded network meta-analyses, overviews and systematic reviews including other study types than RCTs. The systematic reviews had to include at least five RCTs published from 2000 onwards in their first meta-analysis; the year 2000 was chosen because the ‘Consolidated Standards of Reporting Trials’ (CONSORT)25 has led to more transparent and complete reporting since the beginning of the 21st century.26 We included the individual RCTs presented in the first meta-analysis (in the first forest plot) of the systematic reviews and published in 2000 or forward.

Information sources and search strategy

We searched the Cochrane Database of Systematic Reviews (via the Cochrane Library) up to 23 April 2019. To identify all systematic reviews from the CMSG, we used the Cochrane Library’s tool ‘Advanced Search’, with the search limits ‘Cochrane Reviews’ (Content Type) and ‘Musculoskeletal’ (Cochrane Group).

Study selection and data collection process

Two reviewers (SMN and HS) independently evaluated the systematic reviews for eligibility. Disagreements were resolved by discussion. We used the EndNote X7.4 software (Thomson Reuters) to manage the records retrieved from the search. Agreement during study selection was moderate (κ=0.54, 95% CI 0.44 to 0.65) for the screening of titles and abstracts and almost perfect (κ=0.91, 95% CI 0.81 to 1.00) for the full-text assessment.27 One reviewer (HS), supported by another reviewer (SMN) extracted data using a predefined, standardised data extraction form. We pilot tested the form with one review.28

Systematic reviews: For the systematic reviews, we extracted the Cochrane review number, year of publication, first author, total number of RCTs included, number of RCTs eligible for our study, condition studied, type of intervention studied (pharmacological, physical/physiotherapeutic, surgical, psychological, other), contrast(s) used, and outcome domain and outcome measurement instrument for the effect sizes.

RCTs: For eligible RCTs identified from the reviews, we extracted first author, year of publication and the country(/countries) in which the RCT was conducted. We extracted effect sizes, number of patients in each group and risk of bias assessment as reported in the reviews. For the trial population characteristics we extracted data on 20 potentially important contextual factor domains and/or factors of social inequity10 12: sex/gender, comorbidity, healthcare system, adherence to treatment, psychological well-being, age, previous exposure to drugs, patient education/health literacy, disease duration, race, smoking, pain sensitisation, (social) support at work or from family and/or friends, socioeconomic status, occupation, religion, economic status, disability, education, place of residence for the trial population, as well as the measure used in the trial report (predefined working definitions and anticipated measures in table 1 in online supplementary file 1).

If a trial report, reported more measures for the same population characteristic, we extracted the measure first mentioned. If data were not available in the trial reports, we looked for data in the systematic reviews, or, where prespecified, used country-wise estimates from the WHO,29 UNESCO,30 Pew Research Center31 and the World Bank.32

For each trial, we calculated average group means or medians for continuous characteristics (eg, age), and average group proportions for categorical characteristics (eg, sex), and, hence, assuming randomisation had ensured equal numbers and distribution of characteristics in the groups. If not reported, we assumed ‘place of residence’ to be the place (ie, continent) of affiliation for the first author. We imputed missing data with ‘unclear’ for categorical data, and grand median for continuous data. (See the list of characteristics and their operationalisation in online supplementary table 1)

Risk of bias in individual studies

We relied on risk of bias assessments reported in the Cochrane reviews. We extracted number of domains rated as low risk, total number of domains assessed, and we classified trials as ‘low risk of bias’ if all domains had been rated as low risk.

Meta-epidemiological analysis

We estimated treatment effects using ORs, coded so OR >1 indicated a positive (beneficial) effect in favour of the intervention group compared with the control group. For binary outcomes, we calculated the logORs and SE(logOR) and applied continuity correction of 0.5 in case of zero-cells.33 For continuous outcomes, we calculated standardised mean differences (SMDs) and corresponding SEs, and converted into logORs and SE(logOR) by multiplying by π/√3.34 To investigate the association between each of the population characteristics and the treatment effects (ie, logORs), we performed separate multilevel meta-epidemiological analyses.35 The primary model was a one-step meta-regression model with random effects, including a fixed factor for population characteristic and for meta-analysis. We used restricted maximum likelihood estimation.36 We converted resulting ORs and ratio of ORs into risk ratios (RRs) and ratios of RRs (RRRs) for ease of interpretation,37 based on an assumed control event rate of 30%. We quantified the variation explained by each population characteristic by %τ 2 explained=(τ 2 0-τ 2)/τ 2 0×100%, where τ2 0 is the between-trial variation for the meta-regression without the characteristic as factor in the model. Relevant characteristics (potential important contextual factors) were those with a substantial %τ 2 explained and showing statistically significant interaction.38 ‘Statistical significance’ was considered only indicative if p≤0.05, but we interpreted the final statistical outcomes with caution by taking multiple comparisons into account (eg, considering p<0.0025 (0.05/20 population characteristics) as demonstrating statistical significance per se). If data on a population characteristic were available for <2/3 of the included trials (prior to any imputations), we did not perform any analysis of the characteristic. We performed all analyses using the statistical software R V.3.2.3 (R Foundation for Statistical Computing) with the packages ‘meta’ and ‘metafor’.

Patient involvement

We involved one experienced patient research partner (PRP) (MdW) through the whole process, including reviewing the protocol drafts, assisting the interpretation of results and reviewing the manuscript drafts. A total of six PRPs, recruited from the OMERACT community, commented on the findings as presented in an online group meeting.


Study selection and study characteristics of the included trials

Overall, 212 systematic reviews from the CMSG were identified (figure 1). Nineteen systematic reviews were eligible, including 187 RCTs (212 comparisons) published in 2000 or after in their first meta-analysis in their first forest plot (included systematic reviews and trials listed in online supplementary file 2). Most common reason for exclusion were less than five RCTs presented in first meta-analysis in first forest plot of the systematic reviews (excluded systematic reviews listed in online supplementary file 3).

Figure 1

Flow diagram. *Eleven of these were excluded due to being marked as withdrawn in CDSR when the search was carried out. †Study types other than RCTs included clinical controlled trials and cohort studies, but not quasi-RCTs or pseudo-RCTs. This had to be stated in their inclusion criteria. If the systematic review had separate inclusion criteria that included non-RCTs for their assessment of harms, this was ignored. ‡If the outcome in the first forest plot was defined as adverse event, the next forest plot would be assessed instead. CDSR, Cochrane Database of Systematic Reviews; MA, meta-analysis; RCTs, randomised controlled trials.

The eligible trials included a total of 31 274 patients (18 573 in intervention groups and 12 701 in comparison groups); table 1 shows a summary of trial characteristics. The trials most frequently included patients with osteoarthritis, although RA trials were also frequent, assessed the efficacy of pharmacological interventions, placebo/sham was the most frequently used comparison and the most common outcome included in the meta-analyses was pain, although fatigue and function were also frequently assessed.

Table 1

Characteristics of the trials included in this study*

Reporting of population characteristics

Data on the various population characteristics were explicitly reported for a median of 24% (range, 0% to 92%) of the 187 trials (black bars in figure 2). For seven trials, only data from the systematic reviews and other sources were available, since the references cited in the systematic reviews were not obtainable even after contacting the review authors.

Figure 2

Completeness of data for the population characteristics. Black colour indicates data reported in the trial reports, grey colour indicates data obtained from elsewhere (ie, from the systematic reviews, WHO,29 UNESCO,30 Pew Research Center31 and the World Bank,32 and white colour indicates imputation of data prior to analysis. Population characteristics with <2/3 data prior to imputation (indicated by the horizontal dashed line) were not analysed. *Data from the trial reports were not included in the analyses since they did not explicitly describe patients according to high-income or low-income classifications (ie, country-wise data were used for all trials).

Only for sex and age, sufficient data (defined as data being available for ≥2/3 of the trials) were explicitly reported by the trial reports. With the information about the countries where the trials were conducted, data were available for nearly all trials for five additional characteristics, including healthcare system (using information on the WHO index for health system performance),29 patient education/health literacy (using literacy rate estimated by UNESCO),30 religion (using religious composition by country from Pew Research Centre),31 economic status (using the World Bank classifications based on gross national income per capita)32 and place of residence. Overall, despite data being available for 19 out of 20 population characteristics, it could only be summarised for 14 due to inconsistent reporting, and of these only 7 population characteristics had sufficient data and could be analysed.

Association between population characteristics and treatment effects

The ORs from the trials varied between 0.03 and 4698.34 with median of 1.73 (IQR 0.97 to 2.68). The maximum value was from a systematic review on chondroitin28 including a trial with SMD of 4.60 for pain reduction. The meta-regression analyses showed a statistically significant interaction of economic status (explaining 4.8% of the between-trial variation) and place of residence (explaining 0.9% of the variation) (table 2). Furthermore, healthcare system showed a nearly statistical interaction as well (explaining 1.5% of the variation; p=0.055). However, considering multiple testing, only place of residence remained a statistically significant effect modifier (ie, p<0.007 (0.05/7 tests)). Trials conducted in Asia showed higher treatment effects compared with trials in Europe, North America and ‘Other’ (Europe vs Asia, RRR 0.70; North America vs Asia 0.69; other vs Asia 0.62), that is, RRs are 43%–61% higher in Asia. Trials in lower or lower-middle income countries showed higher treatment effects compared with trials in high-income countries (RRR 1.52), that is, the RRs are 52% higher in lower or lower-middle income countries compared with high-income countries. For healthcare system, the OR decreases with a factor 0.92, that is, approximately 8% for each 0.1 increase in the WHO index for health system performance.

Table 2

Association between population characteristics and the treatment effects


This meta-epidemiological study of RCTs in rheumatology showed considerable variation in treatment response across trials and suggests that place of residence, economic status and healthcare system may be important in explaining this variation. Sufficient data were explicitly reported for only two characteristics (sex and age), but information about the countries of the trials from predefined external data sources allowed analysis for a total of 7 out of 20 (35%) characteristics (ie, sex, age, healthcare system performance, patient education/health literacy, predominant religion, economic status and place of residence).

Interestingly, only one of the three potentially important characteristics, that is, healthcare system, was on the provisional list of the seven most important contextual factors developed by the CFWG.12 When the findings were presented to PRPs, they found it surprising that age did not modify the treatment effects. They had expected elderly to experience less treatment effect. However, other meta-epidemiological studies confirm the lack of association for sex and age in patients with RA and broader populations investigated in Cochrane Reviews.23 39 40 A study by Panagiotou et al supports our findings on economic status by showing that trials from more developed countries found less favourable treatment effects compared with trials from less developed countries.41 The PRPs suggested that this may be explained by other underlying factors, for example, higher expectations and access to better healthcare (resulting in lower disease severity at baseline of the trials) leaves less room for improvement for trials conducted in wealthy countries. A cross-sectional study showed that patients with RA with higher education or from countries with high gross domestic product had lower disease activity.42 Similar points are discussed by Panagiotou et al, as were potential biases in reporting and/or study design; they suggest that this could be important to consider when generalising evidence across different settings. However, interpretation of variations in treatment effects according to geographical region should be done with care.43

Inconsistent and scarce reporting of population characteristics makes it impossible to investigate their impact, as demonstrated in this study. Attwood and colleagues13 found a similar pattern of available data for PROGRESS-Plus factors. In 2017, CONSORT-Equity was published aiming to improve the reporting of RCTs where health equity is relevant.8 CONSORT-Equity states that baseline characteristics across relevant PROGRESS-Plus factors should be presented, as well as results of analytic approaches related to equity objectives (eg, subgroup analyses). Some may argue that similar requirements for important contextual factors (ie, modifiers of treatment effects) should apply to any clinical trial to facilitate research investigating for whom (and in which settings) treatments have an effect. Such analyses should be prespecified and procedures for multiple testing must be applied (or analyses should be presented without statistical tests).

Strengths of this study include the study selection done by two independent reviewers, the extensive data extraction, use of prespecified external data sources, relying on outcome data reported in the Cochrane reviews, which in general are considered of high quality,44 45 as well as using one-step meta-epidemiological analyses. We included only RCTs to avoid inherent problems when investigating effect modifiers in non-RCTs.46 Nevertheless, several limitations should be considered. First, insufficient reporting of the included trials made it impossible to investigate most predefined population characteristics. Second, other characteristics may be relevant, such as diet. Third, our restrictive selection criteria may have made our sample less representative, leaving out less commonly investigated treatment–disease combinations. Fourth, since we did not do double data extraction by two independent reviewers, the amount of missing data may be overestimated. Fifth, data assumptions to enable analysis may not be valid in all cases, including using country data for the trial populations and simplifying categorical variables (eg, using continent for place of residence). PRPs were concerned that country data may not sufficiently reflect the trial populations. Adherence was only measured for the intervention period, even though baseline questionnaires exist.47 For the effect sizes, most were reported as SMDs or mean differences so converting to logORs makes the interpretation of the results indirect. Sixth, meta-regressions, despite including RCTs, are observational and their results may be confounded by other characteristics.48 Some may relate to the condition studied, for example, typical eligibility criteria and consequent representativeness of the trial population, country of trial conduction, available funding etc. For example, in our study, RCTs of psychological interventions were only conducted in Europe and in patients with fibromyalgia. However, like most meta-epidemiological studies, we did not adjust for possible confounders.49 Finally, associations between treatment effects and aggregated data for population characteristics (eg, proportion of females or average age) may be misleading, referred to as ecological fallacy, that is, incorrect inferences about individual characteristics are made based on aggregate statistics.50 The PRPs believed there may be patterns within specific rheumatic conditions that are not seen across rheumatology.

Study designs overcoming many limitations are meta-analysis of within-trial subgroup estimates or individual patient data (IPD) meta-analysis.48 50 Both approaches are not yet feasible within rheumatology, since it is uncommon to report subgroup estimates in trial reports and not feasible to obtain IPD for >100 trials. Therefore, efforts are needed to improve reporting in trials—preferably based on a consensus-based list of potentially important contextual factors of high priority.

In conclusion, our sparse results suggest that place of residence (ie, continent of the trial), and possibly economic status (based on World Bank classifications) and healthcare system (based on WHO index for health system performance) may be important contextual factors explaining some of the variation in treatment response across trials. However, considering the study limitations, most importantly, the scarce and inconsistent reporting of most population characteristics, these results should be interpreted with great caution. There is an urgent need for consistent reporting of (important) population characteristics in trials. This requires consensus on the list of characteristics, and how to report their impact on the trial results.


The authors thank the PRPs, Esen Cam (Turkey), Pam Richards (UK), Niti Goel (North America), Michael Gill (Australia), Marieke Voshaar (The Netherlands) and Maarten de Wit (The Netherlands) for contributing with valuable points in the discussion of the results of this study. Furthermore, the authors also thank Wolfgang Viechtbauer, PhD, associate professor of methodology and statistics in the Department of Psychiatry and Neuropsychology (Faculty of Health, Medicine, and Life Sciences) and the School for Mental Health and Neuroscience (Division 2: Mental Health), Maastricht University, The Netherlands, for providing support on statistical programming in R using the metafor package. Finally, they thank associate professor Dorcas Beaton from the Institute for Work and Health, University of Toronto, Canada, for valuable inputs for the manuscript. The OMERACT Contextual Factors Working Group is a large international working group including members from a broad range of disciplines within rheumatology, and this work has benefited directly (or indirectly) from all inputs provided during discussions at working group meetings, teleconferences or mail correspondences.


View Abstract


  • SMN and HS are joint first authors.

  • Handling editor Gerd R Burmester

  • SMN and HS contributed equally.

  • Contributors RC, SMN and HS conceived the study and developed the protocol. HS collected the data. SMN and HS did the analysis and interpreted the analysis in collaboration with RC. SMN, HS and RC drafted the manuscript. All authors critically revised the manuscript for important intellectual content and approved the final version of the manuscript. RC, TE and SMN obtained funding. SMN, HS and RC are the guarantors.

  • Funding The Parker Institute is grateful for the financial support received from public and private foundations, companies and private individuals over the years. The Parker Institute, Bispebjerg and Frederiksberg Hospital is supported by a core grant from the Oak Foundation (OCAY-18-774-OFIL). The Oak Foundation is a group of philanthropic organisations that, since its establishment in 1983, has given grants to not-for-profit organisations around the world. SMN has received PhD scholarships from the Faculty of Health Sciences, University of Southern Denmark and Odense University Hospital, and an introductory scholarship from the BFH Research Foundation. The funders had no role in the study design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all data of the study and had final responsibility for the decision to submit for publication.

  • Competing interests MdW reports personal fees from Abbvie, through Stichting Tools; personal fees from BMS; personal fees from Celgene; personal fees from Lilly; personal fees from Novartis; personal fees from Pfizer; personal fees from Roche; from outside the submitted work. AB holds a research grant from Abbvie and received honoraria for participation in advisory boards from Lilly and Galapagos. All compensations were paid to the department.

  • Patient and public involvement Patients and/or the public were involved in the design, conduct, reporting or dissemination plans of this research. Refer to the Methods section for further details.

  • Patient consent for publication Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement Data are available on reasonable request. Dataset available from the corresponding author after a postpublication period of 1 year allowing time for follow-up projects.