Nine randomised controlled trials were included (667 participants, range of sample size 14 to 193).
Risk of bias according to the criteria was low in three trials, moderate in four trials, and high in two. Overall, 60% of the risk of bias criteria were adequately fulfilled. Common shortcomings included not analysing data according to the intention-to-treat principle, not blinding outcome assessment, and incomplete follow-up of participants. Half the trials assessed adherence; 57% to 75% of participants satisfied the respective adherence criteria.
Four of nine trials achieved significant improvement in depression severity compared with control. Two of these were Chinese trials with a high risk of bias including Tai Chi and Qi Gong and these reported substantially greater effect sizes than the other trials, causing significant heterogeneity. When excluding these two trials from the analysis, there was still a small significant effect of mixed exercise on depression severity (SMD -0.34, 95% CI -0.52 to -0.17; seven trials; no heterogeneity). There was no indication of small study bias.
Results were similar for trials requiring a depression diagnosis for eligibility and for trials using a symptom checklist threshold. The effect remained significant when only considering studies with a low risk of trial (three trials), trials using an active intervention control (four trials), and in trials with extended follow-up (four trials).