Arthritis self-management education programs: A meta-analysis of the effect on pain and disability




Some reports suggest that education programs help arthritis patients better manage their symptoms and improve function. This review of the published literature was undertaken to assess the effect of such programs on pain and disability.


Medline and HealthSTAR were searched for the period 1964–1998. The references of each article were then hand-searched for further publications. Studies were included in the meta-analysis if the intervention contained a self-management education component, a concurrent control group was included, and pain and/or disability were assessed as end points. Two authors reviewed each study. The methodologic attributes and efficacy of the interventions were assessed using a standardized abstraction tool, and the magnitude of the results was converted to a common measure, the effect size. Summary effect sizes were calculated separately for pain and disability.


The search strategy yielded 35 studies, of which 17 met inclusion criteria. The mean age of study participants was 61 years, and 69% were female. On average, 19% of patients did not complete followup (range 0–53%). The summary effect size was 0.12 for pain (95% confidence interval [95% CI] 0.00, 0.24) and 0.07 for disability (95% CI 0.00, 0.15). Funnel plots indicated no significant evidence of bias toward the publication of studies with findings that showed reductions in pain or disability.


The summary effect sizes suggest that arthritis self-management education programs result in small reductions in pain and disability.

Arthritis represents the leading cause of pain and disability in the US, accounting for 18% of all cases of disability (1) and an estimated $150 billion dollars in health care costs (2). In order to improve patients' understanding of their disease and their health outcomes, the American College of Rheumatology Subcommittee on Osteoarthritis recommended that patient self-management education become an integral part of the treatment plan for patients with osteoarthritis (1). Arthritis self-management education programs have been reported to improve pain and disability and reduce health care costs, and the Arthritis Foundation and US Centers for Disease Control and Prevention have recommended arthritis self-management programs as part of the National Arthritis Action Plan (3).

Recommendations for broad dissemination of self-management education programs are based on the perception that such interventions are effective. Indeed, results of a meta-analysis of arthritis self-help programs, conducted by Mullens and colleagues in 1987, suggested that such programs have small-to-moderate effects (4). Those authors analyzed 15 published and unpublished psychosocial trials in adult patients with osteoarthritis and/or rheumatoid arthritis. The summary effect size estimate was 0.21 for pain and 0.10 for disability; based on Cohen's classification of effect sizes, 0.21 would be considered a moderate effect and 0.10 a small effect (5). The effect size is a unitless measure of an intervention's effect and allows one to compare outcomes across many different medical conditions. For example, prednisolone compared with nonsteroidal antiinflammatory drugs in patients with rheumatoid arthritis was found to have an effect size of 0.63 for joint tenderness (6), and vitamin D with calcium compared with placebo for patients with corticosteroid-induced osteoporosis has an effect size of 0.60 (7).

Mullens et al's evaluation of self-management education programs was based on weighted averages that do not take into account study heterogeneity, nor did those authors assess the possibility of publication bias (4). We conducted an updated systematic review of published studies of arthritis self-management education programs to evaluate the effects of such programs on pain and disability, using meta-analytic techniques that take into account heterogeneity of data.


Study selection and data abstraction.

Medline and HealthSTAR were searched for English-language literature published from 1964 to October 15, 1998 with the following medical subject headings: “self-management,” “self-care,” “demand management,” “patient education,” “self-efficacy,” “arthritis,” “osteoarthritis,” and “rheumatoid arthritis.” Further relevant publications were identified by screening the reference lists of all articles. Each article obtained through the search strategy was reviewed by 2 of the authors (AW and DHS) to determine if it met the inclusion criteria. Articles were considered for review if 1) the intervention contained an arthritis self-management education component, 2) a concurrent control group was studied, and 3) pain and/or disability outcomes were evaluated. Self-management education studies that involved exclusively physical or psychosocial therapies, such as biofeedback, relaxation techniques, exercise, and group therapy, were not included in the analysis. Studies of interventions that integrated therapies into an education program were included.

Articles that met the inclusion criteria were independently reviewed by 2 of 3 authors (AW, DHS, PSW), using a standardized abstraction form. Each study was examined to determine patient recruitment procedures, whether and how subjects were randomized, patient demographics, noncompletion (dropout) rates, education methods, and pain and disability outcomes. We assessed sample sizes before and after dropout, patient age, and patient sex. When trials involved multiple treatment arms, we combined those that included self-management education. The dropout rate was calculated as

equation image

Characteristics specific to the education program reviewed included the duration of education, the number of education sessions or “contacts,” the training of educators, the setting of the education program, group or individual encounters, educational format (e.g., written, audio, video, telephone, or face-to-face), and use of a formal syllabus. Each study was also reviewed to ascertain whether a behavioral science model was used in designing the education program. Two commonly employed behavioral frameworks included cognitive behavioral therapy (8) and social cognitive theory, in which self-efficacy is an important construct (9).

While many studies included a variety of end points, we focused on pain and disability as the primary outcomes of interest. We recorded the values for these end points noted in each study. In our primary analysis, data from the last followup for each treatment group were examined. As a sensitivity analysis, we examined data from interim end points obtained prior to the final followup but after the interventions were completed. When data in a report were presented in graphic format, we visually estimated the values. If no standard deviations were presented, we calculated them from the standard error. If the standard error of an outcome was not presented, we used the median standard deviation from other studies included in the meta-analysis that used the same outcome measure.

Data analysis.

Effect sizes were calculated for each pain and disability outcome as the value for the end point in the control group minus the end point in the experimental group, divided by the standard deviation of the outcome in the control group at trial end (10). We used the standard deviation of the outcome in the control group at trial completion because of a concern that the education program may alter the variance in the intervention group. Effect sizes are a unitless measure used to examine the effect of a treatment on a study outcome. If a treatment effect is equal to that of placebo, the effect size will be 0. Effect sizes <0.2 are usually considered small, between 0.2 and 0.5 moderate, and >0.5 large.

We noted substantial heterogeneity among studies in their design and thus decided to calculate a random-effects model as the primary analysis. We also formally assessed for heterogeneity using the Q-statistic (10). Because of the heterogeneity in study design, we represent the random-effects model results as the primary outcome (11). Further, because of heterogeneity in intervention types and an interest in determining the efficacy of the Arthritis Self-Help Course taught through many chapters of the Arthritis Foundation, we conducted subgroup analyses on interventions that closely resembled the program described in the handbook for that course (12).

Evaluation for publication bias.

We investigated the possibility of publication bias among our sample of studies by generating funnel plots. These plots graph the effect size of a study on the horizontal axis and the sample size of the study on the vertical axis. If no publication bias exists, studies with larger sample sizes would have smaller variations in effects, and the effects of smaller studies would range equally above and below this value; therefore, the plot would take on the shape of an inverted funnel. However, if there is bias against the publication of results indicating a null or adverse effect, the funnel plot would be asymmetric, with fewer values toward the left side. Because very large studies might be published even if the findings are nonsignificant, this effect would be especially pronounced for small studies, leading to an absence of studies on the lower left of the funnel plot.


Our search identified 35 potentially eligible trials, of which 18 were subsequently excluded: 2 due to lack of a control group (13, 14), 10 because neither pain nor disability was included as an outcome (15–24), 1 because original data were not described (25), and 5 because the self-management intervention did not have an education component (26–30). The analysis included the remaining 17 trials summarized in Table 1. The mean patient age was 61 years, and 69% of the patients included were female. Of the 4,114 participants in the trials included in this meta-analysis, 19% dropped out of their respective intervention program prior to the final followup (range 0–53%). Most trials recruited patients from primary care or arthritis practices.

Table 1. Arthritis self-management education program trials included in the meta-analysis*
Author (ref.)Arthritis typeStudy designTotal n% dropoutMean age% femaleRecruitment siteFormal syllabusNo. of sessionsProgram duration, weeksEducation modeProgram facilitatorBehavioral model
  • *

    OA = osteoarthritis; RA = rheumatoid arthritis; RCT = randomized controlled trial; W = written materials; V = video programming; NA = not applicable to the study, or adequate information not available; SCT = social cognitive theory (of which Bandura's self-efficacy theory [9] is a part); F = face-to-face contact; HE = health educators; T = telephone contacts; RN = registered nurses; FM = fibromyalgia; LE = lay educators; MH = mental health workers; CBT = cognitive behavioral theory; OT = occupational therapists; PT = physical therapists; RD = registered dieticians; MD = physicians; A = audiocassettes.

  • “Patient,” “location,” and “blocked group” refer to the type of randomization.

  • Primary mode is listed first.

Fries et al (31)OA, RARCT, patient1,099266472ClinicYes2–3 mailings26W, VNASCT
Lindroth et al (32, 44)OA, RANonrandom196536171ClinicYes66FHESCT
Mazzuca et al (33)Knee OANonrandom211226285ClinicYes34F, TRNNone
Simeoni et al (34)OA, RA, FMNonrandom175296681ClinicNo66F, WHESCT
Weinberger et al (35)OARCT, patient439116288ClinicYesMonthly telephone calls and clinic visits up to 44 weeks44F, TLENone
Applebaum et al (38)RARCT, blocked group18446211ClinicYes108FMHCBT
Barlow and Wright (39)RARCT, patient95125881NANo1 leaflet1WNASCT
Bradley et al (40)RARCT, patient68225181ClinicYes1515F, WMHSCT
Cohen et al (41)OA, RARCT, patient96106678Clinic, adsYes66F, WOT, PT, RD, MDSCT
Keefe et al (42, 43)OARCT, patient9966472ClinicYes1010F, W, T, ARN, MHCBT
Lorig et al (45,52)OA, RARCT, location854186484Print adsYes66F, WLESCT
Lorig et al (46)OA, RARCT, patient100156573Print adsYes66F, WLE, PT, MDSCT
Maggs et al (47)PolyarthritisRCT, blocked group16275769ClinicNo11F, WOTNone
Parker et al (48)RARCT, patient830614ClinicYes1-week hospital visit, group visit every 1–3 months52F, W, VNACBT
Radojevic et al (49)RARCT, patient6595476ClinicYes66F, VMHCBT
Riemsma et al (50)RARCT, blocked group249135866ClinicYesEducation packet: video, audio, book, and “passport”24W, V, ARN, PT, MDSCT
Shearn and Fireman (51)RARCT, patient105235676ClinicNo1010FMHNone

For 13 studies (76%), there was a description of a formal syllabus for the educational intervention. For 3 of the trials, the Arthritis Self-Help Course was the syllabus described (12). The programs included 1–15 weekly sessions and ranged in duration from 1 week to 52 weeks. Nine were taught using a group format, 6 an individual format, and 1 used both group and individual teaching; the format of 1 program could not be determined. The educational modalities of programs varied greatly, from face-to-face classes to instructional videotapes or audiocassettes. Fourteen (82%) used face-to-face education as the primary instructional method, and all but 4 (76%) used multiple methods of education. Many different types of personnel participated as program facilitators in the trials, including lay educators (3 trials), trained health educators (2 trials), physicians (3 trials), mental health workers of various training (5 trials), physical or occupational therapists (4 trials), nurses (3 trials), and registered dieticians (1 trial). We examined whether interventions were based on named behavioral theories and found that 13 (76%) were based on a referenced behavioral theory (4 [24%] on cognitive behavioral theory and 9 [53%] on social cognitive theory).

The pain and disability effect sizes for each study and the pooled effect are presented in Figure 1. The effect sizes for studies that assessed a patient's pain (n = 16) ranged from −0.32 to 0.55, and for those that assessed disability (n = 12), effect sizes ranged from −0.27 to 0.39. Only 4 studies that assessed pain and 1 that assessed disability had a 95% confidence interval (95% CI) that was above 0.

Figure 1.

Estimated effect size (solid circles) on arthritis pain and disability and 95% confidence intervals (horizontal bars), for studies included in the meta-analysis. Vertical line indicates a zero effect size, or equality between control and intervention groups. Values on the left of the vertical line represent trial effects in which the control arm had better outcomes, and values and on the right represent trials in which the self-management education program (SMEP) results were better. Summary effect sizes, calculated using random-effects models, are displayed at the bottom.

The summary effect size revealed a small effect of self-management education interventions on pain (effect size 0.12 [95% CI 0.00, 0.24]) and an even smaller effect on disability (effect size 0.07 [95% CI 0.00, 0.15]). For studies that assessed pain, the Q-statistic P value was 0.004, indicating significant heterogeneity. In contrast, the P value for the Q-statistic of the studies examining disability was 0.4, indicating no evidence of heterogeneity. In sensitivity analyses including data from interim, as opposed to final, end points, the pooled effect size for pain was increased (effect size 0.19 [95% CI 0.11, 0.27]); the results for disability were unchanged.

Subgroup analysis of the 3 studies that most closely resembled the Arthritis Self-Help Course (12) revealed a small, nonsignificant effect on pain (effect size 0.04 [95% CI −0.19, 0.27]) and on disability (effect size 0.10 [95% CI −0.05, 0.24]).

Finally, we examined the funnel plots to determine whether there was evidence of bias toward publication of trials with positive results (Figures 2A and B). Both plots showed relative symmetry, reflecting a lack of evidence of publication bias.

Figure 2.

Funnel plots from studies evaluating the effect of arthritis self-management education programs on pain (A) and disability (B). Sample size was plotted against effect size for each study. In the absence of any publication bias, one would expect a symmetric funnel shape, from large studies representing the true effect to smaller studies with more scattered results. If bias occurs and smaller studies with null results are not published, fewer points would be expected on the left side of the funnel. Neither the pain nor the disability plot indicates any firm evidence of publication bias.


Findings of this meta-analysis suggest that arthritis self-management education programs result in small reductions in pain and disability. There was no clear evidence of publication bias against reporting of small studies with null or negative results. Overall, there was a substantial dropout rate before the final followup point, which may have introduced some bias. A dropout rate of 20% in a community-based program is to be expected, but in the setting of clinical trials a rate of 19%, as we found in this meta-analysis, raises concern regarding validity. Interestingly, studies with the largest effect sizes also had the largest dropout rates. The 5 studies that had statistically significant positive effect sizes (31–35) were examined qualitatively to determine whether specific intervention characteristics might predict better outcomes. We could not find any specific feature that distinguished these studies from the rest.

One possible limitation of this analysis, and of any meta-analysis in this area, is the heterogeneity between trials. The interventions we included varied substantially, but by using narrow inclusion criteria and prespecified end points we attempted to improve our ability to interpret the pooled results. However, inclusion of interventions of different intensities may have diluted the apparent effects of self-management education programs that provided more direct interaction between the educator and the patient. The use of the effect size makes it possible to combine end point data measured with different scales (such as the Arthritis Impact Measurement Scales [36] or Health Assessment Questionnaire [37]). Another limitation of this study comes from weaknesses and unreported data in some of the underlying literature. While many trials were carefully conducted and described, several of the reports did not indicate use of validated outcome measures, did not explain how scales were interpreted, or did not include a description of the educational intervention. It is clear from the results of this review that methodologic standards for designing, conducting, analyzing, and reporting self-management education trials need to be developed.

In summary, this meta-analysis suggests that arthritis self-management education programs lead to small but significant reductions in pain and disability. It is generally assumed that these programs are a highly effective and relatively inexpensive way of providing patients with tools to better manage their arthritis. It is important that we understand the efficacy of such interventions in multiple settings before devoting substantial resources to their wide dissemination. High dropout rates in the trial setting and unclear methodologic standards make it difficult to draw any firm conclusions. There is a pressing need for additional high-quality, independent, controlled trials to further assess the benefit and cost-effectiveness of arthritis self-management education programs.