Comparator groups matter! – The impact of comparator group interventions on effect estimates of exercise trials for knee osteoarthritis: a systematic review and meta-analysis of randomised controlled trials

Purpose: In randomized controlled trials (RCTs), the effectiveness of exercise therapy for knee osteoarthritis (OA) has been determined by comparing it to a variety of comparator interventions. While several efforts have been made to investigate whether effect estimates differ between subgroups of patients and interventions, systematic reviews of exercise therapy for knee OA have largely ignored the variability in interventions provided in comparator groups and its implications for interpreting and comparing effect sizes. The aims of this systematic review and meta-analysis were to (i) assess the impact of variability in comparator group interventions on effect estimates of pain in exercise trials including people with knee OA, and (ii) investigate the association between participant and intervention characteristics and effect estimates.

Methods: MEDLINE, EMBASE, and The Cochrane Database of Systematic Reviews were searched for systematic reviews synthesizing the effect of exercise therapy for knee OA published between January 2015 and October 2020. Included RCTs were extracted from reference lists and screened for eligibility. RCTs comparing exercise therapy compliant with the American College of Sports Medicine (ACSM) criteria for muscle strength training or aerobic training to any comparator intervention measuring patient-reported pain were eligible for inclusion. Comparator interventions were grouped as no intervention, non-ACSM exercise therapy, education/self-management, and passive modalities, and compared using stratified random-effects meta-analyses. Associations between participant and intervention characteristics (age, sex, BMI, duration of intervention, difference in attention, difference in number of treatment sessions, baseline pain,) and effect estimates were assessed using univariate meta-regression analyses.

Results: Thirty-five trials including 2412 participants (68% female, mean age 62 ± 5) and 44 study comparisons were included. No intervention was applied in the comparator group in 18 comparisons, non-ACSM exercise therapy in 15, education/self-management in six, and passive modalities in five. Stratified meta-analyses showed statistically significant differences in effect estimates depending on type of intervention provided in the comparator group. Statistically significant effects in favor of ACSM exercise therapy were found when compared to passive modalities (SMD 1.76, 95% CI 0.49 to 3.04), no intervention (SMD 0.93, 95% CI 0.50 to 1.36) and education/self-management (SMD 0.27, 95% CI 0.07 to 0.47). Contrary, no statistically significant effect was found when comparing ACSM exercise therapy to non-ACSM exercise therapy (SMD 0.09, 95% CI -0.06 to 0.23) Meta-regression analyses adjusted for participant and trial characteristics showed no associations with effect estimates. Overall quality of evidence was low to moderate across subgroups. Main reasons for downgrading included risk of bias, inconsistency, and risk of publication bias.

Conclusion: The effect of exercise therapy compliant with the ACSM criteria for muscle strength training or aerobic training for knee OA pain vary considerably depending on the intervention provided in the comparator group. This variability should be considered when interpreting and comparing effect sizes of exercises therapy to other interventions.