Efficacy of cardiorespiratory aerobic exercise in rheumatoid arthritis: Meta-analysis of randomized controlled trials




Several lines of evidence have emphasized an improvement in aerobic capacity and muscle strength after physical exercise programs in rheumatoid arthritis (RA) patients. Our objective was to evaluate the efficacy of aerobic exercises in RA on quality of life, function, and clinical and radiologic outcomes by a systematic literature review and a meta-analysis.


A systematic literature search was performed in the Medline, EMBase, and Cochrane databases up to July 2009 and in the abstracts presented at rheumatology scientific meetings during the last 5 years. Randomized controlled trials (RCTs) comparing aerobic exercises with non-aerobic interventions in RA patients were included. Outcomes studied were postintervention quality of life, function assessed by the Health Assessment Questionnaire (HAQ), a pain visual analog scale (VAS), joint count, the Disease Activity Score in 28 joints (DAS28), and radiologic damage. Efficacy was assessed by standardized mean differences (SMDs; difference between groups of mean outcome variation from baseline/SD at baseline) of aerobic exercises versus non-aerobic rehabilitation. Heterogeneity was tested. SMDs were pooled by a meta-analysis using the inverse of variance model.


Fourteen RCTs, including 1,040 patients, met the inclusion criteria. Exercise improved the postintervention quality of life (SMD 0.39, P < 0.0001), HAQ score (SMD 0.24, P = 0.0009), and pain VAS (SMD 0.31, P = 0.02). Exercise in this RA population appeared safe, since global compliance, DAS28, and joint count were similar in both groups.


Cardiorespiratory aerobic conditioning in stable RA appears to be safe and improves some of the most important outcome measures. However, the degree of the effect of aerobic exercise on the abovementioned parameters is small.


Impairment in range of motion, muscle strength, endurance, and aerobic fitness results in serious loss of function, work disability, dependency, impaired social or family function, and reduced quality of life in rheumatoid arthritis (RA) patients (1). Even if pharmacologic interventions have largely improved RA management over the past decade, physical therapy remains an important part of treatment. Previously, the exercise therapy in RA aimed only at maintaining joint mobility and muscle strength. Because exercises were thought to provoke joint damage by enhancing disease activity (2), patients with inflammatory arthritis, especially RA, were discouraged from performing weight-bearing exercises. In a previous randomized controlled trial (RCT), we showed that an exercise program had a greater impact on both disability assessed by the Health Assessment Questionnaire (HAQ) and quality of life than conventional joint rehabilitation (3), contrasting with most of the previous similar studies that failed to detect any statistical difference on the HAQ (4, 5) or quality of life (6). Several trials showed that exercises were safe in RA rehabilitation and effectively improved aerobic fitness (3, 5, 7, 8). Previous systematic literature reviews of the efficacy of physical exercises in RA patients confirmed these results (9, 10). Given the limited number of studies, reviewers decided not to pool data and were therefore unable to draw numerical conclusions concerning the efficacy of exercise programs on other important outcome measures in RA. During the past few years, several trials have been carried out to examine the effect of exercises in RA, but results with respect to pain, disease activity, functional ability, quality of life, and structural damage are still unclear. We therefore carried out this systematic literature review in order to determine whether aerobic exercises effectively improve the abovementioned parameters in RA.


Type of intervention.

A cardiorespiratory aerobic exercise is an exercise that improves VO2 and is usually performed at 50–80% of the maximal heart rate (220 − age). The American College of Sports Medicine defined a dynamic exercise program as aerobic exercises performed at between 60% and 80% of the maximal heart rate at least twice a week for 6 weeks (11). Strengthening exercises and aerobic exercises without training intensity (i.e., maximal heart rate) monitoring were not examined in this systematic review.

Non-aerobic rehabilitation was defined by the following terms: “static,” “range of motion,” “isometric,” “seated immersion,” “relaxation,” “stretching,” “no attention,” or “usual care.” As a consequence, trials with aerobic exercises in the control group were not considered.

Search strategy.

An extensive search of PubMed, EMBase, and the Cochrane Central Register of RCTs (until July 28, 2009) was made by two reviewers (AB and L-AB). The following keywords were used for database screening: (“Arthritis, Rheumatoid” [medical subject headings (MeSH)]) AND (“Exercise Therapy” [MeSH] OR “Activities of Daily Living” [MeSH] OR “Physical Education and Training” [MeSH]). The only limit of the search was “clinical trial.” A hand search of references from relevant articles, from review articles, and from abstracts presented at the Annual Scientific Meetings of the American College of Rheumatology (ACR), the Annual Congress of the European League Against Rheumatism, and the Scientific Meetings of the French Society of Rheumatology published in the past 5 years completed the search. A search on the ClinicalTrials.gov Web site was also performed to identify randomized studies that were not yet published.


Inclusion criteria were an RCT evaluating cardiorespiratory aerobic exercises performed at 50–90% of the maximal heart rate in comparison with non-aerobic rehabilitation in adult patients with RA as defined by the ACR (formerly the American Rheumatism Association) criteria (12, 13). Exclusion criteria consisted of postsurgery rehabilitation and articles written in a language other than English, French, or German.

Quality assessment.

A single author (AB) assessed the methodologic quality of each study included in the meta-analysis on both the Jadad scale (14) and the Checklist to Evaluate A Report of a Non-Pharmacological Trial (CLEAR-NPT) (15), ranging from 0–5 and 0–14, respectively, where a high score indicates high quality.

Data extraction.

One investigator (AB) selected the articles and collected the data using a predetermined form, including study design (randomization procedure, blinding, followup period, and intent-to-treat analysis), patient characteristics (number, age, sex, disease duration, functional status, treatment, rate of rheumatoid factor [RF]–positive patients, and proportion of completers), and intervention parameters (duration of an exercise session, overall duration of the intervention, exercise type, frequency, and intensity, i.e., maximal heart rate).


We used the Cochrane Musculoskeletal Group recommendation to select outcome measures (16). We reported tender joint count, swollen joint count, pain, and disability. We also reported data concerning withdrawals (i.e., the number of completers) and adverse effects in order to assess exercise safety. The following outcomes (the references of which are available in Supplementary Table 1, available in the online version of this article at http://www3.interscience.wiley.com/journal/77005015/home) were extracted from the publications: 1) quality of life, evaluated on the Nottingham Health Profile (NHP), the Rheumatoid Arthritis Quality of Life (RAQol) questionnaire, the physical component of the 36-item Short Form health survey (SF-36), the Arthritis Impact Measurement Scales Health Status Questionnaire (AIMS), and the McMaster Toronto Arthritis Patient Preference Disability Questionnaire (MACTAR); 2) function, assessed by the HAQ; 3) the Disease Activity Score in 28 joints (DAS28); the DAS4 was converted into the DAS28 as follow: DAS28 = (1.072 × DAS4) + 0.938; 4) joint count (number of swollen joints, number of tender joints, and Ritchie Articular Index); 5) pain on a visual analog scale (VAS); 6) radiologic evaluation by Larsen's method; and 7) exercise tolerance, evaluated both by exercise compliance (i.e., the number of completers) and by exercise safety. For the latter parameter, we reported adverse effects such as steroid injection due to local disease flare, cardiovascular events, and joint/muscle soreness.

Quality of life and joint count evaluations were expressed as a percentage of the maximum possible score for the method used. Standardized data abstraction concerned means or medians and measures of dispersion. Since the measure of dispersion for change was not always available, a conservative estimate was taken from baseline data and converted into an SD. When median values were given instead of mean values, the median was analyzed as a mean. Data were collected at several end points: baseline, 1 month ± 2 weeks, 3 months ± 4 weeks, 6 months ± 2 months, 1 year, and 2 years or more. When a trial was reported in several publications, the more informative publication was included in the meta-analysis. When studies reported more than 2 subgroups, we included only the first intervention group described in this study and its corresponding control group.

Statistical analysis.

Heterogeneity was tested using Cochran's test. I2 values >50% were defined to indicate significant heterogeneity. The efficacy of intervention versus non-aerobic rehabilitation was assessed in each study by the calculation of the standardized mean difference (SMD; difference between both groups of mean outcome variation from baseline/SD at baseline) and 95% confidence interval (95% CI). Individual SMDs were pooled using the method of the inverse of variance. Intervention safety was assessed by the odds ratio (OR) and 95% CI. The results of individual trials were pooled by meta-analysis using the Mantel-Haenszel method. In case of heterogeneity, a random-effects model was used (17). Otherwise, a fixed-effects model was applied. SMDs between 0.2 and 0.5 indicated a small effect, between 0.5 and 0.8 indicated a medium effect, and >0.8 indicated a large effect. To relate the efficacy in SMD units to a more familiar outcome, we transformed the SMD into the difference in mean outcome scores (experimental versus control) on that scale (18). Meta-analyses were done with Review Manager 5 (Cochrane), and additional statistics were developed with StatsDirect software (StatsDirect).

Sensitivity analysis and heterogeneity assessment.

A sensitivity analysis was conducted to evaluate the robustness of the meta-analysis by assessing the influence of an individual study on the overall SMD. We therefore examined the effect of removing each study individually from the meta-analysis. To explore heterogeneity, we combined studies into 2 or 3 subgroups according to the trial design (published before or after 2000, Jadad scale score <3 or ≥3), the disease characteristics (mean disease duration <5 years or ≥5 years, functional status class I–II or including class III), and the intervention parameters (supervised or home based; overall duration of the intervention <3, 3, or >3 months; exercise frequency <3 or ≥3 times/week; duration of an exercise session <30, 30–60, or >60 minutes). Heterogeneity between the subgroups was tested using a chi-square test (19).

Publication bias was assessed using funnel plot analysis, Begg's test, and Egger's test.


Trial flow.

A total of 513 abstracts were identified by database searching and 6 articles by hand searching. Of these, 140 duplicates were removed and 298 abstracts were excluded because of no physical exercise (n = 95), no RA patients (n = 171), age <18 years (n = 13), postsurgery rehabilitation (n = 9), no RCT (n = 9), and language (n = 1). Eighty-one full-text reports were analyzed and 67 articles were excluded because of previous publication (n = 7), other joint disease (n = 18), no RCT or aerobic exercise in the control group (n = 14), no aerobic exercise in the intervention group (n = 11), training intensity not provided (n = 5), and no extractable data (n = 12). Fourteen articles were finally entered in the analysis.

Study characteristics.

Study characteristics are shown in Table 1. The mean ± SD Jadad scale score was 2.4 ± 0.6. Seven trials displayed a Jadad scale score <3 and only 2 trials validated 7 CLEAR-NPT items. Among the 14 trials with cardiorespiratory aerobic exercise conditioning, the intervention fulfilled the dynamic exercise program criteria in 5 studies (5, 8, 20–22). The control group underwent range of motion exercises in 3 studies (5, 22, 23), non-aerobic exercises in 1 study (24), education programs in 2 studies (3, 25), and usual care in 8 studies (8, 20, 21, 26–30).

Table 1. Trial characteristics*
Author, year (ref.)CountryJadad scale scoreFunction statusIntervention groupControl group
Patients, no.Exercise typeMax HR, %Frequency, per weekDuration, minutesLength, weeksAttendance, %Patients, no.Comparator
  • *

    Max = maximal; HR = heart rate; nr = not reported/inapplicable; ACR = American College of Rheumatology; DEP = dynamic exercise program; ROM = range of motion.

  • The median percentage of sessions attended was 74% (interquartile range 27%), 30% of patients had a 50–75% attendance rate, and 49% of patients had a high attendance rate (>75%).

  • One hour daily (supervised) every second week plus 30 minutes daily (home based), leading to a mean ± SD training time of 339 ± 179 minutes/week.

Bilberg et al, 2005 (24)Sweden3Steinbrocker I–III22Supervised cardiorespiratory aerobic conditioning70245127827Non-aerobic exercises
Van den Berg et al, 2006 (25)The Netherlands3nr82Home-based cardiorespiratory aerobic conditioning60–80510–30523478Education
De Jong et al, 2003 (20)The Netherlands3ACR I–III150Supervised DEP70–9027510474150Usual care
Westby et al, 2000 (8)Canada2ACR I–II14Home-based DEP60–75345–60527116Usual care
Van den Ende et al, 2000 (22)The Netherlands3nr34Supervised DEP60315∼4nr30ROM exercises
Melikoglu et al, 2006 (23)Turkey2ACR I–II20Supervised cardiorespiratory aerobic conditioning602202nr20ROM exercises
Hansen et al, 1993 (27)Denmark2Steinbrocker I–II60Home-based cardiorespiratory aerobic conditioning70≥3<90104>5015Usual care
Harkcom et al, 1985 (21)US2Steinbrocker II14Supervised DEP70315–3512nr6Usual care
Van den Ende et al, 1996 (5)The Netherlands3nr75Supervised and home-based DEPs70–85360127525ROM exercises
Baillet et al, 2009 (3)France3ACR I–II25Supervised cardiorespiratory aerobic conditioning60–80545410023Education
Neuberger et al, 2007 (29)US3nr173Supervised and home-based cardiorespiratory aerobic conditioning60–80360128375Usual care
Nordemar et al, 1981 (30)Sweden1ACR I–III23Home-based cardiorespiratory aerobic conditioning706052nr23Usual care
Noreau et al, 1995 (26)Canada2ACR I–II19Supervised cardiorespiratory aerobic conditioning50–70215–30128310Usual care
Lyngberg et al, 1994 (28)Denmark2Steinbrocker I–III12Supervised cardiorespiratory aerobic conditioning50–7024512nr12Usual care

Patients' characteristics.

The meta-analysis included 1,040 patients (510 patients in the intervention group and 530 in the control group). Both groups were similar in terms of age, disease duration, sex ratio, proportion of completers, rate of RF-positive patients, and pharmaceutical treatments (see Supplementary Table 2, available in the online version of this article at http://www3.interscience.wiley.com/journal/77005015/home). Mean age and disease duration in the studies ranged from 44–68 years and 1–16 years, respectively (in the 13 trials reporting these parameters). For the 11 studies for which sex was reported, 46.7–100% were women. The rate of RF-positive patients ranged from 59–93.3% in the 5 studies assessing this parameter.


Data on quality of life were available for 5 studies, including 298 patients in the intervention group and 288 in the control group. This outcome was measured by the AIMS (26), the NHP (3), the RAQol (25), the SF-36 (24), or the MACTAR (20) (see Supplementary Table 3, available in the online version of this article at http://www3.interscience.wiley.com/journal/77005015/home). A small beneficial effect of aerobic intervention on the quality of life of RA patients was shown (SMD 0.39 [95% CI 0.23, 0.56], P < 0.0001) (Figure 1). Heterogeneity was not statistically significant in this subset of studies (I2 = 45%).

Figure 1.

Efficacy of cardiorespiratory aerobic exercises on quality of life. ¥ = percentage of rate reduction compared with the control group; SMD = standardized mean difference; 95% CI = 95% confidence interval; IV = method of the inverse of variance.

Nine studies evaluated the impact of aerobic exercises on the HAQ, comprising a total of 387 patients in the intervention group and 384 patients in the control group (Figure 2). Exercises provided a small positive effect on the HAQ (SMD 0.24 [95% CI 0.10, 0.38], P = 0.0009; I2 = 29%).

Figure 2.

Efficacy of cardiorespiratory aerobic exercises on function, assessed by the Health Assessment Questionnaire (HAQ). ¥ = percentage of rate reduction compared with the control group; SMD = standardized mean difference; 95% CI = 95% confidence interval; IV = method of the inverse of variance.

Data on the pain VAS were available in 6 studies (138 patients in the intervention group and 123 in the control group). The overall SMD for pain measured by the VAS was 0.31 (95% CI 0.06, 0.55; P = 0.02, I2 = 30%) (Figure 3).

Figure 3.

Efficacy of cardiorespiratory aerobic exercises on a pain visual analog scale (VAS). ¥ = percentage of rate reduction compared with the control group; SMD = standardized mean difference; 95% CI = 95% confidence interval; IV = method of the inverse of variance.

Nine studies, including 228 patients in the intervention group and 209 patients in the control group, evaluated joint count. The Ritchie Articular Index was reported in 2 trials (5, 23), swollen joint count in 2 studies (21, 27), tender joint count in 2 studies (26, 28), and both swollen and tender joint counts in 3 studies (8, 22, 29) (see Supplementary Table 3, available in the online version of this article at http://www3.interscience.wiley.com/journal/77005015/home). Intervention tended to provide a positive effect on this outcome, although the difference did not reach the statistical significance (SMD 0.14 [95% CI −0.05, 0.33], P = 0.14; I2 = 0%) (Figure 4).

Figure 4.

Efficacy of cardiorespiratory aerobic exercises on joint count. ¥ = percentage of rate reduction compared with the control group; SMD = standardized mean difference; 95% CI = 95% confidence interval; IV = method of the inverse of variance.

Only 3 studies evaluating 376 RA patients for radiologic damage as assessed by radiologic findings were available (188 patients in each group). Aerobic exercise training was suspected to have a joint damage sparing effect because the SMD was 0.36 (95% CI 0.16, 0.56; P = 0.0005, I2 = 17%) (see Supplementary Figure 1, available in the online version of this article at http://www3.interscience.wiley.com/journal/77005015/home).

Disease activity, evaluated by the DAS28, was recorded in 4 studies (291 patients in the intervention group and 281 patients in the control group). The exercise group did not display a worse DAS28 score (SMD 0.08 [95% CI −0.08, 0.25], P = 0.34; I2 = 67%).

Both groups had similar numbers of completers: 460 (86.8%) of 530 patients in the intervention group versus 445 (87.3%) of 510 patients in the control group (OR 0.80 [95% CI 0.56, 1.16], P = 0.24; I2 = 55%). Forty-seven adverse events were reported in the intervention group and 34 adverse events were reported in the control group. However, the difference was not statistically significant (OR 1.67 [95% CI 0.36, 7.69], P = 0.51; I2 = 73%). Steroid injection in both groups was similar (OR 0.85 [95% CI 0.42, 1.29], P = 0.98; I2 = 0%). Three cardiovascular events were reported in the 14 trials. One myocardial infarction and one pulmonary embolism occurred in the intervention group. It was not possible to determine in which group the remaining cardiovascular event occurred. A single trial reported joint soreness and a compression fracture in the exercise group.

Heterogeneity exploration and sensitivity analysis.

No statistical heterogeneity was detected in the outcome measures, except for disease activity on the DAS28.

Aerobic cardiorespiratory conditioning ameliorated the HAQ in patients with class I–II functional status, whereas it barely had any effect in patients with more severe functional status (Table 2). Disease duration influenced quality of life and pain. In early RA (disease duration <5 years) there was a positive impact of exercises on quality of life, but no impact in established RA (disease duration ≥5 years). Pain responded in an opposite direction, with pain reduced if the duration was >5 years, but no effect if it was <5 years. Cardiorespiratory aerobic exercise conditioning had a positive impact on quality of life when performed <3 times per week, whereas it had no effect when performed ≥3 times per week. The duration of the individual session and exercise supervision also had an impact on quality of life. If the duration of the exercise session was >60 minutes there was a positive impact, whereas exercise sessions lasting ≤60 minutes had no effect. If exercise was supervised there was a positive impact, but there was no effect if the exercise program was home based and unsupervised. The entire program duration impacted pain. If the program lasted 3 months, less postintervention pain was improved, but it did not improve when the program lasted more than 3 months. Methodologic quality had no impact on any of the variables, whereas publication date impacted the pain VAS results. Studies published after 2000 showed a positive effect on the VAS, whereas studies published before 2000 did not. Sensitivity analyses showed that the SMD and 95% CI were not substantially altered by removing any of the trials (data not shown).

Table 2. Heterogeneity exploration*
 Quality of lifeHAQJoint countPain VAS
SMD (95% CI)I2, %SMD (95% CI)I2, %SMD (95% CI)I2, %SMD (95% CI)I2, %
  • *

    The efficacy of aerobic exercises versus control was evaluated in the subgroups by standardized mean differences (SMDs) and 95% confidence intervals (95% CIs). Heterogeneity between subgroups was tested using a chi-square test. HAQ = Health Assessment Questionnaire; VAS = visual analog scale; NA = not applicable.

Patient parameters        
 Functional status        
  Class I–II0.36 (−0.09, 0.82)00.60 (0.27, 0.94)280.21 (−0.14, 0.56)00.34 (0.08, 0.59)27
  Including class III0.55 (0.34, 0.77)00.16 (−0.05, 0.37)00.11 (−0.69, 0.91)0NANA
  Chi-square test, P0.47 0.03 0.48 NA 
 Disease duration        
  <5 years0.55 (0.34, 0.77)00.22 (0.01, 0.42)63NANA0.03 (−0.34, 0.41)0
  ≥5 years0.16 (−0.10, 0.42)00.26 (0.07, −0.46)150.15 (−0.05, 0.34)00.56 (0.32, 0.79)0
  Chi-square test, P0.02 0.75 NA 0.02 
Exercise parameters        
  <3 times/week0.53 (0.32, 0.73)00.16 (−0.07, 0.26)00.01 (−0.54, 0.56)00.31 (−0.25, 0.87)0
  ≥3 times/week0.12 (−0.12, 0.43)230.26 (0.11, 0.41)360.35 (0.05, 0.66)00.43 (0.21, 0.64)47
  Chi-square test, P0.03 0.28 0.29 0.71 
 Duration of session        
  <30 minutes0.07 (−0.38, 0.24)NA0.19 (0.07, 0.46)610.33 (−0.23, 0.88)00.56 (0.04, 1.08)0
  30–60 minutes0.28 (−0.18, 0.73)00.41 (−0.04, 0.87)00.01 (−0.54, 0.56)0NANA
  >60 minutes0.57 (0.36, 0.78)00.24 (−0.05, 0.42)450.36 (−0.13, 0.86)420.38 (0.17, 0.60)73
  Chi-square test, P0.03 0.71 0.56 0.54 
 Program duration        
  <3 months0.45 (−0.13, 1.02)00.52 (0.09, 0.95)620.27 (−0.35, 0.89)00.67 (0.24, 1.09)0
  3 months0.28 (−0.19, 0.74)NA0.37 (−0.04, 0.77)00.35 (−0.01, 0.70)00.50 (0.23, 0.77)0
  >3 months0.40 (0.22, 0.59)860.18 (−0.02, 0.34)340.09 (−0.46, 0.65)NA0.04 (−0.46, 0.37)11
  Chi-square test, P0.86 0.28 0.75 0.04 
  Supervised0.52 (0.32, 0.71)00.25 (0.09, 0.42)10.15 (−0.05, 0.34)00.34 (0.08, 0.59)27
  Home based0.07 (−0.24, 0.38)NA0.20 (−0.08, 0.49)73NANANANA
  Chi-square test, P0.02 0.28 NA NA 
Trial design        
 Publication date        
  Before 20000.28 (−0.19, 0.74)00.46 (0.12, 0.80)260.14 (−0.18, 0.46)00.03 (−0.34, 0.41)0
  After 20000.41 (0.23, 0.58)720.19 (0.04, 0.35)250.59 (0.09, 1.09)00.56 (0.32, 0.79)0
  Chi-square test, P0.60 0.16 0.14 0.02 
  Jadad scale score <30.28 (−0.19, 0.74)00.46 (0.08, 0.83)360.17 (−0.22, 0.56)00.46 (0.04, 0.88)0
  Jadad scale score ≥30.41 (0.23, 0.58)680.20 (0.05, 0.36)200.37 (−0.01, 0.74)420.40 (0.17, 0.62)22
  Chi-square test, P0.60 0.22 0.47 0.78 

Publication bias.

We did not identify any publication bias for quality of life, pain, joint count, radiologic damage, and DAS28 assessments. Concerning the HAQ meta-analysis, both the Begg's test result (P = 0.045) and the Egger's test result (P = 0.017) were significant, indicating a potential bias. Inspection of funnel plots for the HAQ assessment revealed that studies with an important standard error, i.e., with a few patients, and a negative SMD, i.e., showing better HAQ assessment in the control group than the intervention group, were lacking (see Supplementary Figure 2, available in the online version of this article at http://www3.interscience.wiley.com/journal/77005015/home).

Clinical relevance.

Quality of life, HAQ, and pain VAS postintervention changes were considered clinically meaningful when they were greater than 0.2 (31). It was noteworthy that several clinical trials achieved such clinical significance on the HAQ (3, 8). However, the SMDs were smaller than 0.5, indicating a small effect of aerobic exercise on each outcome measure (Figures 1–4).


This systematic review and meta-analysis showed that cardiorespiratory aerobic exercises improve some of the most important RA patient outcomes: function, quality of life, and pain. Moreover, it appears that aerobic exercise decreases radiologic damage and pain. DAS28, joint count, compliance, and adverse events were similar in both groups, indicating that exercises were safe in stable RA. Only RCTs focusing on the efficacy of aerobic exercise in RA patients were considered in this review, creating two groups with similar demographic disease-related parameters and comparable treatments at baseline and providing a strong internal validity.

Although our review suggests that exercise, on its own, improves most disease outcomes, a few limitations should be emphasized. First, data were extracted from the literature by a single observer and although this extraction was supervised every two months by two other reviewers, a potential bias may exist. Second, there was some methodologic bias in most studies. Lack of blinding of outcome assessors, for instance, may cause bias. A recent study (32) emphasized differences in the physical function level of RA patients in different European countries with higher levels of physically active patients in northern European countries than in the South. We did not examine the influence of the baseline physical activity on either outcomes or heterogeneity, as this parameter was seldom reported. Since we focused on RCTs, this potential confounder was probably equally distributed in control and intervention groups, but we could not determine whether exercise programs were as effective with physically active patients as with patients without regular physical activity or without exercise program experience.

This meta-analysis suggests that aerobic exercise therapy is safe in stable RA. Appraisal of adverse effects during nonpharmacologic interventions is less stringent than during pharmacologic interventions, a fact explaining, at least partially, a lower detection of adverse events (33). Therefore, adverse events occurring during exercise programs may be underestimated. Similarly, data on compliance are more often described in pharmacologic treatment reports than in nonpharmacologic treatment reports. The rate of completers is probably a less accurate measurement of compliance than the percentage of the maximum number of sessions that could be attended. Unfortunately, only a few studies reported this outcome measure, mostly in long-term intervention trials. Fatigability with exertion and cardiovascular adverse events are important proxies for exercise tolerance, which have not yet been accurately evaluated. According to Metsios et al, these two outcome measures deserve more attention in future trials (34).

The efficacy of exercise on pain assessed by a VAS, quality of life, disability assessed by the HAQ, and joint count was statistically significant, but the magnitude of the effect was small (SMD <0.5). However, most of these outcome measures were designed for monitoring patients in pharmacologic trials and may not be appropriate for the evaluation of physical interventions. Therefore, a small improvement of the HAQ or quality of life may reflect an inability of these indices to detect the effects of an exercise program rather than a failure of aerobic exercises to improve patients' health status (9, 35, 36). The efficacy of aerobic exercises on disability, quality of life, pain, and joint count should be more rigorously compared with conventional pharmacologic treatment in further investigations in order to determine their place in RA management.

Several lines of evidence suggested that any exercise is better than no exercise at all (32), but the exercise parameters (intensity, duration, frequency, and type) that result in better effects are not clearly defined. The ideal way of looking at the effect of exercise intensity, duration, frequency, and type would be to have an RCT comparing these parameters (head-to-head comparisons). Unfortunately, such data are sparse. In our meta-analysis, we performed subgroup analysis–based heterogeneity exploration as an alternative approach to evaluating the influence of patient characteristics or intervention parameters on the sizes of the treatment effect. Aerobic cardiorespiratory conditioning improved the HAQ in patients with class I–II functional status but did not ameliorate patients with more severe functional status. Both program duration and disease duration impacted pain, with better results in case of established RA and short-term programs, whereas data concerning quality of life suggested that exercise benefits more early RA patients than established RA patients. Supervised exercise and 60-minute exercise sessions performed biweekly or less had a greater effect on quality of life than home-based exercises and 30-minute exercise sessions performed more than 3 times per week, respectively. Similarly, Neuberger et al showed that patients taking part in a 12-week class exercise program experienced significant amelioration in disease-related parameters, whereas home-based exercise did not have that result, probably because patients in this group exercised at a lower intensity (29). Indeed, most of the studies reporting positive effects of exercises on quality of life were supervised intervention trials (3, 20, 24), whereas most of the studies in which this outcome was not modified were home-based exercises (25). It is likely that supervised exercises result in higher adherence to the exercise program and positive group enthusiasm, explaining the trend of higher SMDs in subsets of trials with supervised intervention. However, the impact of supervised dynamic exercise programs on work and on consumption of medical and of paramedical resources is unclear. Van den Hout et al suggested that supervised class exercises in the Rheumatoid Arthritis Patients In Training cohort (37), which are more expensive than home-based interventions (3), provided insufficient improvement to justify the additional costs as a medicoeconomic issue.

An exercise program could be of particular interest to the elderly by reducing the risk of falls and fractures (38–40) and by improving the cardiovascular disease risk profile (41). Since a single RCT evaluated the efficacy of aerobic exercise in RA patients age >65 years, we were unable to look for heterogeneity of exercise efficacy or safety according to this parameter. Therefore, our results showing the benefit of aerobic exercises in middle-aged RA patients should not be generalized to the older RA population. However, Lyngberg et al (28) showed that an individually adapted training session performed at 50–70% of the maximal heart rate can be performed in elderly RA patients. Physiologic function capacity decreases linearly with aging (42, 43). As a consequence, a baseline evaluation of physiologic function capacity and cardiovascular risk factors is mandatory in order to adapt exercise intensity to older RA patients. Moreover, stringent monitoring of adverse events as well as a progressive increase of exercise intensity appear appropriate in this population. A possible way to increase the compliance of older RA patients would be to perform “natural” aerobic exercise such as walking or cycling.

RA patients are dramatically physically inactive (32). This systematic review supports a more frequent recommendation of exercise to RA patients. Besides the positive effect of the intervention on patients' psychological well-being, aerobic exercise should be considered as a safe therapy, the efficacy of which has been underestimated. The clinically meaningful and economic impact of such treatment must be investigated in further trials in order to clearly define the place of aerobic exercises in RA management.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Mr. Baillet 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 conception and design. Baillet, Zeboulon, Gossec, Juvin, Dougados, Gaudin.

Acquisition of data. Baillet, Bodin.

Analysis and interpretation of data. Baillet, Zeboulon, Gossec, Combescure, Juvin, Dougados.


Abbott France organized a meta-analysis methods workshop but played no further role in the project. Abbott France did not participate in drafting the article, did not see the manuscript before submission; and in particular, Abbott France did not choose the subject and did not participate in the study design, the data collection, or the data analysis. This article was not submitted for approval to Abbott France.


The authors thank Drs. Michel Guinot, Catherine Bioteau, and Gaëtan Gavazzi for their assistance in the manuscript redaction.