Is health equity considered in systematic reviews of the cochrane musculoskeletal group?




To determine whether Cochrane Musculoskeletal Group (CMSG) systematic reviews and corresponding primary studies of rheumatoid arthritis interventions report and analyze the data needed to assess the effectiveness of interventions in reducing socioeconomic differences in health and/or improving the health of the poor.


We selected all CMSG reviews on rheumatoid arthritis published since issue 1, 2003. Fourteen reviews were identified; 147 of the 156 primary studies included in these reviews were obtained and assessed. We extracted data on whether the dimensions place of residence, race/ethnicity/culture, occupation, gender, religion, education, socioeconomic status, and social capital and networks (PROGRESS) were reported or analyzed, and whether any interventions were aimed at disadvantaged or low- and middle-income country populations.


Among the dimensions of PROGRESS reported at baseline in 147 primary studies, gender (89%) was the most commonly reported, followed by education (25%) and race/ethnicity (18%). Less than 50% of the systematic reviews reported dimensions of PROGRESS even when they had been reported in the primary study. Of 147 primary studies, 6 (5%) were aimed specifically at disadvantaged populations; another 6 reported on effectiveness by at least 1 dimension of PROGRESS.


Primary studies of interventions for rheumatoid arthritis generally reported few variables necessary to answer questions about health inequalities. Most CMSG systematic reviews failed to assess those variables even when described in the primary studies. The Cochrane Health Equity Field welcomes the opportunity to provide guidance to systematic review authors on incorporating equity considerations into their reviews.


Musculoskeletal disorders are the leading cause of chronic disability worldwide. In 2001, musculoskeletal disorders were responsible for 30 million disability-adjusted life years: 9 million in developed countries and 21 million in developing countries (1). Rheumatoid arthritis, with its chronic, autoimmune, inflammatory, destructive process that results in progressive and irreversible joint damage, is one of the major contributors to the musculoskeletal burden of disease (2). It is associated with premature mortality and reduces quality of life and physical, social, and emotional functioning (3, 4). Onset occurs most commonly in the fourth and fifth decades (5) and therefore affects individuals in their peak income-earning years, resulting in high work disability (6). In Australia, 2.4% of the population report having rheumatoid arthritis (7), and a recent report estimates the prevalence of rheumatoid arthritis in American adults (≥18 years of age) at 0.6%, affecting almost 1.3 million people (8). Minimizing the impact of this devastating disease is a priority (9).

Social gradients in health are a major public health problem in both developing countries and industrialized countries. Analysis of Canadian data from 1996 found considerable disparities in disability-free life expectancy across geographic health regions. The gap between minimum and maximum disability-free life expectancy was 11.8 years (10). Socioeconomic differences in health exist worldwide, and for most diseases and conditions. Musculoskeletal conditions are no exception. For example, in the 1994 Canadian Population Health Survey, the prevalence of arthritis was 19% in persons with some secondary education compared with 11% in those with a higher education level (11). In Australia, across 5 quintiles of socioeconomic disadvantage, the percentage of people reporting arthritis in the Australian National Health Survey decreased from 20% in the most disadvantaged category to 15% in the least disadvantaged (7).

In the US, the rate of self-reported disabilities such as activity restriction, change in job status, or missed days of work increased as family income decreased in the US Health and Nutrition Examination Survey (12).

We know that the effectiveness of interventions may be different for disadvantaged and advantaged populations. For example, Hawker and colleagues (13) found that, despite a higher need, people with low socioeconomic status (SES; income and education) had lower utilization rates for total joint arthroplasty (TJA) than people with high SES. These findings may be due to implementation factors such as differential access, screening/diagnosis, practitioner factors, and consumer compliance. Indeed, Tugwell and colleagues (14) have argued that there is a substantial potential “staircase” effect of these 4 factors, which causes lower community effectiveness and an increased relative gap between poorest and least poor. For example, TJA loses almost all efficacy when one considers real-world conditions where need for TJA is underdiagnosed, access to surgery is imperfect, and willingness to undergo surgery is low. In the poorest populations, even more efficacy is lost due to even lower access, poorer diagnostic accuracy, and lower adherence, resulting in a relative gap of 1.6 in “real-world” effectiveness between the least poor and poorest populations.

The Cochrane Collaboration and the Campbell Collaboration have established themselves internationally as important, credible sources of policy- and practice-relevant systematic reviews, due to rigorous, transparent methods and a regular updating process. The Cochrane Collaboration aims to assist clinicians, patients, and policy makers in making well-informed decisions by maintaining, making accessible, and preparing systematic reviews of research on the effects of health care policies and interventions (15). The Campbell Collaboration is a sibling organization summarizing the evidence on behavioral, educational, social, and social justice intervention effects on health (16). Subsequent to 2 general reviews of the effectiveness of interventions at reducing socioeconomic health inequalities, both of which found significant gaps in available evidence (17–19), there have been calls for both the Cochrane and Campbell Collaborations to systematically synthesize evidence on what works to reduce inequalities across socioeconomic strata (20, 21). In response, the Campbell Collaboration Equity Methods Group and the Cochrane Collaboration Health Equity Field have recently been established to develop an international, systematic evidence base of the effectiveness of interventions in reducing socioeconomic inequalities in health (22, 23). The Cochrane Musculoskeletal Group (CMSG) is making this a priority, and this article describes the first phase of applying the “equity lens” to the Cochrane systematic reviews of rheumatoid arthritis.


Definition of effectiveness in reducing inequalities in health (21).

An intervention is deemed effective if it is shown to be effective in absolute terms in people who are disadvantaged (for example, school feeding programs for disadvantaged children [24], a program in Mexico that pays rather than charges poor families for clinic and school attendance, and an individual targeting technique in Colombia that provides subsidized health insurance to disadvantaged persons [25]).

Selection strategy.

We selected all 14 CMSG systematic reviews on rheumatoid arthritis published on the Cochrane Library over the last 5 years (since issue 1, 2003) (26–39).

Data extraction.

Two independent reviewers (either LM, VW, M-AN, EU, or MK) extracted data using a pretested extraction form. Any discrepancies were resolved by discussion and consultation with a third reviewer.

The extraction form was developed in consultation with the team and by pilot testing on 4 reviews. The acronym PROGRESS, developed by Evans and Brown, was used to describe dimensions of inequalities in health (40). PROGRESS represents the following 8 factors: place of residence, race/ethnicity/culture, occupation, gender, religion, education, socioeconomic status, and social capital and networks (Table 1).

Table 1. Multidimensional components affecting health equity
Potential dimensions of health inequalitiesDescription
Place of residenceRural/urban/inner-city
Race/ethnicity/cultureRacial, ethnic, cultural background
OccupationBlue collar, sedentary, etc.
GenderMale or female
ReligionReligious background
EducationYears of education, high school education, etc.
Socioeconomic statusIncome, type of housing, single-parent family
Social capital and networksMeasures neighborhood or community trust, ability to rely on neighbors for support (e.g., civic society organizations)


We defined 3 levels of analysis for both reviews and primary studies: 1) whether these Cochrane reviews and primary studies contained baseline information about the dimensions place of residence, race/ethnicity/culture, occupation, gender, religion, education, socioeconomic status, and social capital and networks; 2) whether these studies presented their results for any of these dimensions of socioeconomic disadvantage; and 3) whether any studies were conducted in low- or middle-income countries (LMICs). Data were extracted for each of these levels, where available. We cross-referenced the primary studies with the Cochrane reviews.


Systematic reviews.

Fourteen CMSG reviews (26–39) on interventions for rheumatoid arthritis were published between issue 1, 2003 and issue 2, 2007 (Table 2).

Table 2. Cochrane systematic reviews on interventions in rheumatoid arthritis, 2003–2007*
Subject of review, (ref.)Major outcomeIncluded studies, no.Sample size
  • *

    ACR = American College of Rheumatology; EULAR = European League Against Rheumatism; WHO = World Health Organization; ILAR = International League of Associations for Rheumatology; OMERACT = Outcome Measures in Rheumatology Clinical Trials; TENS = transcutaneous electrical stimulation.

  • ACR core set and response (56).

Acupuncture and electroacupuncture (26)Pain249
Adalimumab (27)ACR; EULAR response rate62,390
Balneotherapy (28)WHO/ILAR core set6355
Glucocorticoids (29)Radiographs181,414
Etanercept (30)WHO/ILAR core set; ACR3955
Intraarticular steroids and splints/rest (31)OMERACT outcome measures7346
Low-level laser therapy (32)OMERACT outcome measures6222
Occupational therapy (33)Pain, fatigue, functional abilities432,163
Acetaminophen (34)Pain4121
Patient education (35)OMERACT outcome measures487,369
Rofecoxib (36)ACR28,734
Short-term low-dose corticosteroids (37)Joint tenderness, pain, grip strength10320
Tai chi (38)OMERACT outcome measures3206
TENS (39)Pain378

Primary studies.

Within the 14 reviews, there were a total of 161 included studies. Five studies were included in more than 1 systematic review; after removing these duplicates, there were 156 primary studies. Of the 156 primary studies, 147 were obtained and assessed (see Supplemental Appendix A, available in the online version of this article at Three were studies in languages other than English (2 German and 1 Chinese), and these were included in the assessment with the assistance of translators. Nine studies were unavailable from the library service. There were no substantive discrepancies between the results of the data extraction among the independent reviewers.

Outcome 1.

The first outcome was whether these Cochrane reviews and primary studies contained baseline information about the following PROGRESS dimensions: place of residence, race/ethnicity/culture, occupation, gender, religion, education, social capital and networks, and socioeconomic status.

Systematic reviews.

In the inclusion criteria sections of the systematic reviews, none of the 14 reviews specified information on any PROGRESS items or the type of population. Although not specifically described in the inclusion criteria, gender distribution of participants was reported in 5 reviews and the place of residence of the sample was reported in 2 reviews.

Primary studies.

The study populations were described by ≥1 PROGRESS factor in 131 of the 147 included studies (Table 3). Gender was the most commonly reported item (131 studies [89%]), followed by level of education (36 studies [25%]), race/ethnicity (26 studies [18%]), social networks and capital (23 studies [15.6%]), and occupation (24 studies [16%]). Fifteen studies (10%) reported participants' socioeconomic status. None of the studies reported religion.

Table 3. Reviews and primary studies reporting PROGRESS factors in the description of the population*
 Reviews (out of 14) that reported PROGRESS items, no. (%)Primary studies (out of 147) that reported PROGRESS items, no. (%)
  • *

    PROGRESS = place of residence, race/ethnicity/culture, occupation, gender, religion, education, socioeconomic status, and social capital and networks.

Place of residence2 (14.3)20 (13.6)
Race/ethnicity/culture0 (0)26 (17.7)
Occupation0 (0)24 (16.3)
Gender5 (35.7)131 (89.1)
Religion0 (0)0 (0)
Education0 (0)36 (24.5)
Socioeconomic status0 (0)15 (10.2)
Social capital and networks0 (0)23 (15.6)

Importantly, 12 studies (8%) excluded patients with low education. These studies were on patient education and stipulated that patients must have attained a certain level of education (for example, a sixth-grade reading level or the ability to read and write). None of the studies reported excluding participants based on race or religion.

Outcome 2.

The second outcome was whether these studies presented their results for any of these dimensions of socioeconomic disadvantage.

Systematic reviews.

None of the 14 systematic reviews in our study reported differences in treatment effect across any of the PROGRESS items, even though some of the primary studies did so.

Primary studies.

Eleven of the primary studies provided analyses that addressed 1 or more of the PROGRESS factors (Table 4). In 6 of these (41–46), the majority of the patients were from a disadvantaged population. Three (41, 47, 48) of the 11 studies analyzed the results by education level; none found differences in the outcomes. Hall et al (48) found multicollinearity with education, income, and occupation in the initial analysis and chose education as the most representative. In many studies, education level is a carrier variable for reading level and health literacy. Two studies (47, 48) assessed outcomes by occupation; one study (47) found that people with higher occupational levels scored better on patient knowledge, although there was no correlation between the extent of improvement on the test and occupational level.

Table 4. Dimensions of disadvantage in primary studies*
Study, year (ref.)Study designInterventionPROGRESSImpact of intervention
  • *

    PROGRESS = place of residence, race/ethnicity/culture, occupation, gender, religion, education, socioeconomic status, social capital and networks; BA = before/after study (no control group); OT = occupational therapy; RCT = randomized controlled trial; ROM = range of motion; WMD = weighted mean difference; 95% CI = 95% confidence interval; CBA = controlled before/after; RA = rheumatoid arthritis; ES = effect size.

Barry et al, 1994 (50)BA1-hour instruction/treatment session by OT   X    No association between patient performance and gender (P = 0.78).
Goeppinger et al, 1989 (41)RCTIntervention: booklets and audiotape to home study group Control: 2-hour lessons at a community site to a small groupXX X X XLevel of education did not affect the amount of change in depression, pain, or knowledge after the interventions. Changes in social support were not related to improvement in pain.
Hall et al, 1996 (48)RCTIntervention: hydrotherapy Control: seated immersion, land exercise, or progressive relaxation  XX    Women receiving hydrotherapy had significantly increased knee ROM between pre- and posttest measures (P = 0.049). Compared with land exercise, hydrotherapy was statistically significant in women (WMD 9.90; 95% CI 2.75, 17.05) but not in men (WMD −1.10; 95% CI −11.4, 9.2).
Hass et al, 1997 (51)CBAIntervention: instruction on selection of OT devices through a workshop, group meeting, home visit, and telephone followup Control: traditional selection process   X    No significant difference within groups in the change score for pain in either men >64 years (intervention: P = 0.331; control: P = 0.134) or women >64 years (intervention: P = 0.424; control: P = 0.166).
Hill et al, 2001 (43)RCTIntervention: patient education program on RA treatments and various coping Control: drug information leaflet   X X  Primary outcome of adherence to treatment as measured by pharmacologic marker. End of study compliance rates: 85% in treatment group; 55% control group.
Helliwell et al, 1999 (44)RCTIntervention: 4-week, 2-hour patient education program Control: usual care     X XOnly within the education group, there was significant improvement in social functioning subscale. No significant differences between groups found for Larsen radiologic score.
Kaplan et al, 1981 (47)RCTIntervention: education session + group counseling Control: education session alone XXX XX No relationship found between the change in improvement on the education test and education (r = 0.16, P < 0.1) and occupational levels (r = 0.18, P < 0.1) on initial and followup patient education test results.
Parker et al, 1988 (46)RCTIntervention: cognitive–behavioral pain management Control: attention-placebo group and control group   X  X Cognitive–behavioral group significantly more likely able to feel confident in ability to decrease pain than control group at 12 months (3.3 versus 2.9, 0–6 scale; P < 0.05). However, no significant group differences in pain or disease status measures at 12 months.
Parker et al, 1984 (45)RCTIntervention: inpatient rheumatology care and a 7-hour patient education program Control: inpatient care only   X  X Patient education group had significantly higher scores on knowledge score (P < 0.05), but also had significantly higher pain scores (P < 0.05) and more impairment of activity levels than control group.
Savelkoul et al, 2001 (42)RCTCoping intervention group: action-directed coping sessions led by therapist Mutual support control group: led by 2 trained patients Waiting list control group: usual care       XCoping intervention increased participants' action-directed coping compared with mutual support and functional health status compared with waiting list control postintervention. These effects did not persist at 6 months' followup. Action-directed coping: ES = 0.18 (P < 0.05); functional health status: ES = 0.08 (P < 0.05).
Stern et al, 1996 (49)Cross-over RCTIntervention: 3 commercial wrist extensor orthoses Control: no orthosis   X    Changes in grip strength were not statistically significant in men or women.

The patient population in 2 patient education studies comprised participants with an education level lower than high school (43, 44). In both of these studies, there was no significant difference in effectiveness of the intervention between the treatment and control groups in terms of pain or radiologic progression.

Five of the 11 primary studies analyzed the results by gender (41, 48–51); only 1 study reported a difference (48). In this study, Hall et al reported the results of balneotherapy on knee range of motion separately for men and women (48). In men, there was a trend for hydrotherapy to improve knee range of motion compared with land exercise (weighted mean difference −10.1; 95% confidence interval [95% CI] −21.39, 1.19), whereas in women, the trend was in favor of the land exercise group (weighted mean difference 10.8; 95% CI −0.82, 22.4); however, neither of these results was statistically significant.

In a study targeting a rural setting, the sample largely comprised white women with a high school education. Self-care lessons were administered to a home study group and a community-based small group. Both interventions improved arthritis knowledge and pain at 12 months (41).

The majority of participants in 2 studies, one on patient education and the other on cognitive–behavioral therapy, were men in a low socioeconomic class (45, 46). The participants treated with cognitive–behavioral therapy were more likely than controls to feel confident in their ability to reduce pain. Interestingly, the patient education group had higher levels of pain after treatment compared with the control group.

Outcome 3.

The third outcome was whether any studies were conducted in LMICs.

Systematic reviews.

None of the systematic reviews in this sample reported any studies conducted in LMICs, nor did they report on low- or middle-income relevance.

Primary studies.

None of the primary studies were conducted in LMICs.


The evidence base on the effectiveness of interventions to address inequalities must be augmented. Gwatkin writes that while there have been gains by researchers in understanding and conceptualizing health inequalities, “[research] has not yet reached the heart of the matter: the identification of measures that can effectively deal with the inequalities that have been uncovered” (52).

To reduce health disparities, we need to understand what works and what does not. The Cochrane Collaboration now contains more than 3,000 systematic reviews of treatments across most health issues. Musculoskeletal disease represents an important proportion of chronic conditions, and rheumatoid arthritis is one of the most disabling and is the prototype for many new intervention modalities. This condition, therefore, provides a good model for developing an approach to applying the equity lens to systematic reviews for Campbell, Cochrane, and other groups carrying out evaluations of interventions for clinicians, patients, and policy makers.

As demonstrated in the “equity effectiveness loop” (15), a number of different factors are known to impact the effectiveness of interventions. These may often involve more than health care. It is likely that socioeconomic factors may have a larger impact in nonclinical interventions such as education, building codes, road safety, and work modification. This is supported by our finding from 4 studies that patient education was more effective for people with higher education and from another 3 studies showing that patient education did not work for people with low education/low income; these results are consistent with findings from Gepkens and Gunning-Schepers (18). This suggests that patient education may not be the intervention of choice for people from low SES backgrounds, and indeed may contribute to increasing inequalities. Contextual factors need to be taken into account, particularly for individuals with low income. It does no good to educate someone on how to exercise when they do not have the time, money, facilities, or transportation to do so. Thus, different types of social interventions, beyond those that are focused on health outcomes only and that attempt to change context, may be more appropriate. Although not evaluated in these studies, the connection between reading level, health literacy, and education level needs to be explored.

The included systematic reviews on interventions to improve outcomes in rheumatoid arthritis did not provide an analysis of the effects across socioeconomic factors, nor did they point out gaps in available knowledge regarding socioeconomic differences in intervention effectiveness. However, there were a few primary studies in the rheumatoid arthritis reviews that did. A few others focused on interventions aimed specifically at people who were disadvantaged. Although in the long term we need to encourage an equity orientation in all primary studies, we need to immediately encourage authors of systematic reviews to highlight and extract data from those studies that have analyzed subgroups relevant to equity. We recognize that while some PROGRESS factors are easier to measure than others (e.g., religion), it is still important to encourage authors to report all of these factors.

Our results justify the need to encourage greater consideration of equity in Cochrane and Campbell reviews and provide a clear basis for the promotion of greater consideration of equity implications and social impact in systematic reviews in general. These results also provide an impetus for researchers who conduct primary studies to collect and then make these data available to systematic review authors so that equity considerations can be explored.

Most of the analyses in primary studies were designed to control for PROGRESS factors or verify that PROGRESS factors were not confounding the primary comparison. The studies that merely analyzed SES as a confounder and that found null results give us very little information. It is more important to know what the effect sizes were for different SES strata. Even if such studies do not reach significance due to low power, there still may be meaningful differences in effect size by SES. Thus, separate subgroup analyses by PROGRESS factors are needed. We also found that some studies explicitly exclude disadvantaged patients because they may not fully comply with the intervention; this is quite an important finding for people interested in how inequalities may be created or maintained through health care interventions. If trials tend to exclude disadvantaged groups (which they do), then the opportunity to learn about the means of reducing inequalities may be more limited than it could be, and may bias our understanding of the effectiveness of such interventions (53).

The results of this review further demonstrate the need for primary studies to provide quantitative details of SES results so that trends can be identified from pooled data. The concept of “equity proofing” of policies and programs is discussed in the recent World Health Organization (WHO) report on social determinants of health (54). This concept requires that, in the development of policies and programs, there is an opportunity to “identify, assess and address its potential health equity impacts,” and that this should apply even to those programs without a stated equity focus. It is clear that when we control for a potential confounder of an intervention, such as social status, we lose the ability to equity proof. The WHO report makes the point that the shape of health gradients may look different in different countries; therefore, interventions (or, rather, how they are implemented) may need to differ in different countries. Again, this requires knowledge about how these gradients differentially affect different groups in society (for example, younger versus older, men versus women, or urban versus rural).

We found that none of these Cochrane reviews included studies conducted in LMICs. We propose 2 strategies to ensure LMIC relevance of Cochrane reviews: 1) analysis of results across country settings, if possible, and 2) identification of gaps in existing knowledge. For example, D'Souza and D'Souza (55) identified a complete lack of evidence from LMICs regarding the effectiveness of vitamin A for treating measles in children, and concluded that the results may not be generalizable to LMICs. The Cochrane Collaboration has now developed an advisory group on the relevance and inclusion of LMICs. Ongoing efforts from this group and others to recruit Cochrane reviewers from LMICs should help to promote greater relevance of Cochrane reviews to LMICs. The time lag in adoption of new interventions and therapy in LMICs after they have been introduced in high-income countries does not appear to be correcting itself. Even in industrialized nations, disadvantaged populations do not adopt interventions that have been adopted by the majority in higher SES groups (such as smoking cessation, physical activity, or cholesterol-lowering drugs).

The major limitation of this article is selection bias. We analyzed Cochrane reviews, which mostly focus on controlled trials. Studies of other designs may well have valuable information, particularly on the effects of policy-level interventions, but for the purposes of this project we focused on controlled trials. Controlled trials (at least in the past) were unlikely to examine differences in effect across socioeconomic strata. Even if controlled trials did include distributive data on the included populations and analyzed the impact on treatment outcome, these data would be considered observational unless treatment groups were stratified by the PROGRESS factor of interest. The types of distributive analyses we are seeking may be found more frequently in different study designs (such as controlled before/after studies, cohort studies, and interrupted time series). It is also important to note that where these differential effects have been reported, care in using them to directly inform policy is indicated; in some cases, they may not have been the prior focus of the study in question, but emerged during further analyses, and will require confirmation in subsequent trials. However, they do suggest where interventions may be effective (or harmful) in addressing inequalities and where future research efforts may be directed.

Few Cochrane reviews present evidence about dimensions of health inequalities. Furthermore, the majority of reviews fail to report outcomes stratified by socioeconomic factors that are described in the original studies. Most primary studies do not assess the impact of interventions across socioeconomic strata (only 10%). The Cochrane Collaboration Health Equity Field and the Campbell Collaboration Equity Methods Group are developing methods for standardizing assessment of health equity factors in Cochrane and Campbell reviews. These groups will advise Cochrane reviewers about when and how to incorporate data on effectiveness for different socioeconomic groups into systematic reviews, refine methods for searching the literature on studies including equity indicators, develop methods for systematically reviewing studies on interventions to reduce health inequities, refine methods for meta-analysis of equity-oriented studies, and advise on software development.

We urge attention to the important challenge of establishing a global evidence repository to ensure that interventions being advocated in the name of the poor are not going to be a waste of money, or worse, contribute inadvertently to widening inequalities.


Dr. Tugwell had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Tugwell, Maxwell, Welch, Kristjansson, Ueffing, Tsikata.

Acquisition of data. Tugwell, Maxwell, Welch, Buchbinder, Suarez- Almazor, Nowlan, Ueffing, Khan, Tsikata.

Analysis and interpretation of data. Tugwell, Maxwell, Welch, Kristjansson, Wells, Buchbinder, Suarez-Almazor, Ueffing, Tsikata.

Manuscript preparation. Tugwell, Maxwell, Welch, Kristjansson, Petticrew, Wells, Suarez-Almazor, Nowlan, Ueffing, Khan, Shea.

Statistical analysis. Tugwell, Maxwell, Wells, Nowlan, Khan.


We thank Wu Taixiang and Jane Zochling for their assistance in translating the nonEnglish primary studies.