Using bivariate meta-analysis and meta-regression to investigate the effects of exercise on pain and disability in osteoarthritis

Article type
Authors
Uthman O1, van der Windt D2, Jordan J2, Dziedzic K2, Healey E2, Jordan K2, Foster N2
1University of Warwick, UK
2Keele University, UK
Abstract
Background: Previous reviews have documented that exercise interventions are more effective than no exercise control for osteoarthritis (OA) however effect estimates vary widely.

Objectives: To estimate effectiveness of exercise interventions and explore trial-level characteristics that may be associated with effect size estimates of exercise for OA.

Methods: Pain and functional limitations are both key outcomes in OA. Bivariate random effects meta-analysis was used to simultaneously synthesize effects on pain and function, taking the correlation between the two outcomes into account. We calculated 95% prediction intervals which incorporate between-study variability. A series of unadjusted bivariate meta-regression analyses was carried out to investigate the impact of trial-level characteristics on treatment effect size estimates.

Results: A total of 43 trials involving 4466 patients met the inclusion criteria. The results of the bivariate meta-analysis showed that exercise interventions significantly reduced pain ( = −1.35 cm; 95% CI −1.75 to −0.95 cm, 10 cm visual analogue scale) and improved function (1.03 units; 95% CI −1.60 to −0.80 units, WOMAC disability scale from 0 to 10). There was statistically significant strong correlation (0.740, p < 0.001) between pain relief and improvement in function. The prediction intervals suggest that exercise interventions applied at population level may not always be beneficial in all settings, about 15% future trials are likely to show exercise not to be effective for pain and function. Exercise tended to be more effective among younger adults; in hospital-based settings, and when supervised and standardized. Trials with low risk of bias showed less promising results.

Conclusions: This review provides insight into some of the sources of variability in effect estimates of exercise interventions for OA. In this bivariate meta-analysis, effect estimates for pain and function were pooled simultaneously in a single analysis in order to reduce reporting bias due to outcome measures ‘borrowing strength’ from each other.