Xuxi Zhang

intervention, 4) Dietary ‐ control intervention, and 5) Multifactorial intervention (interventions targeting more than one risk factors). We applied the Cochrane Collaboration’s tool to assess the risk of bias in our study. This tool consists of six domains: selection bias, performance bias, detection bias, attrition bias, reporting bias and other bias. For each domain, the study was graded as having a low risk, high risk, or unclear risk of bias. 22 Grades of Recommendations Assessment, Development and Evaluation (GRADE) was used to evaluate the level of evidence in the meta ‐ analysis with GRADEpro3.2. Two reviewers (Yusufu and Zhang) assessed each study independently. Disagreements between the reviewers were discussed with a third researcher (Sun) in order to reach an agreement. Data synthesis and analysis The heterogeneity between the studies was evaluated with the I 2 test. Random effects models were estimated to calculate pooled Odds Ratios (OR) of DR development, worsening and progression. For these analyses we considered a value of P < 0.05 to be significant. A sensitivity analysis was performed to test the stability of the studies by excluding one study at a time. Possible publication bias was assessed by estimating funnel plots with Begg and Egger tests, and a value of P < 0.1 was considered to be significant. 23, 24 We followed the Preferred Reporting Items for Systematic Reviews and Meta ‐ analyses (PRISMA) checklist to report our meta ‐ analysis study. 25 All statistical analyses were performed using Stata 11.0. RESULTS Study selection and Study Characteristics The 22 studies included in this meta ‐ analysis studied a total of 22,511 participants. The number of participants in each study ranged from 35 15 to 11,140 6 . In most studies, the number of males and females was similar 6 ‐ 8, 10 ‐ 13, 15 ‐ 21 , but in two studies 5, 14 , over 90% of participants were male. The follow ‐ up interval of the interventions ranged from 1 year 15 to 8 years 21 . Blood ‐ pressure ‐ control intervention was evaluated in 4 studies 8, 9, 19, 21 , glycemic ‐ control intervention was evaluated in 9 studies 5 ‐ 7, 14, 15, 18, 19, 21 . Lipid ‐ control intervention was evaluated in 2 studies 19, 21 . Dietary ‐ control intervention was evaluated in 2 studies 20 . Multifactorial intervention was evaluated in 5 studies 11 ‐ 13, 16, 17 . More details of the included studies can be found in Supplementary Table S1. Risk of bias None of the RCTs included in this review were double ‐ blinded. In all studies, no high risk of bias was found in the domains of selection bias, detection bias, attrition bias, reporting bias, and other bias. More details of the risk of bias could be found in Supplementary Table S4. Quality of the evidence for most results on new onset DR and DR worsening was moderate to high, except the results of glycemic ‐ control intervention (new onset DR), glycemic ‐ control intervention (DR Worsening), follow ‐ up <2 years (DR Worsening) and follow ‐ up >5 years (DR 7 167 Prevention and control of diabetic retinopathy