OBJECTIVE: To investigate how risk of bias and intervention type modify effect sizes of exercise interventions that are intended to reduce chronic low back pain intensity. DESIGN: Systematic review with meta-epidemiologic regression analysis. LITERATURE SEARCH: PubMed, CENTRAL, Embase, and CINAHL (until January 31, 2021). STUDY SELECTION CRITERIA: Systematic reviews with meta-analyses of randomized controlled exercise trials. DATA SYNTHESIS: The dependent variable was pain, calculated as standardized mean difference (SMD). Potential effect modifiers were risk of bias, exercise modes, study, and meta-analyses characteristics. Multilevel meta-regressions and inverse variance-weighted meta-regressions with random intercepts were modelled. RESULTS: Data from 26 systematic reviews (k = 349 effect sizes, n = 18,879 participants) were analysed. The overall mean effect was SMD: -0.35 (k = 349, [95% CI -0.02 to -0.7]). There was a clinically relevant effect overestimation in studies with a high risk of bias due to missing outcomes (each k = 197, Beta coefficient = -1.9 [95% CI -2.9 to -.9]) and low sample size (B = 0.01 [.001 to .01], [ie, one participant more leads to an SMD decrease of 0.01]). There was a clinically relevant underestimation of the effect when studies were at high risk of bias in allocation concealment (B = 1.3 [.5 to 2.1]) and outcome measurement (B = 1.3 [.44 to 2.0]). Motor control and stabilization training (B = -1.3 [-2.3 to -.37]) had the largest effects; stretching (B = 1.3 [-.03 to .5]) had the smallest effect. CONCLUSIONS: The effects of exercise trials at high risk of bias may be overestimated or underestimated. After accounting for risk of bias, motor control and stabilization exercises may represent the most effective exercise therapies for chronic low back pain. J Orthop Sports Phys Ther 2022;52(12):792-802. Epub: 12 August 2022. doi:10.2519/jospt.2022.11149.
Keywords: low back/lumbar spine; pain; statistical analysis/research design; systematic review/meta-analysis.