Data analysis
We carried out meta-analysis using STATA version 14.2 (StataCorp, College Station, TX, USA) and based on the recommendation for the meta-analysis of observational studies. We conducted meta-analysis only among those studies that reported type 2 diabetes. Heterogeneity among studies was assessed with the Cochran chi-square (χ2) and quantified with the I2 and tau-square (τ2) and the values have been mentioned in the text. Substantial heterogeneity was anticipated when I 2 values were greater than 75%. The pooled proportion was estimated using the inverse variance method. Confidence Interval (CI) was estimated at 95% for each proportion. Random effects model was used for the overall pooled estimate meta-analysis. We did not conduct meta-analyses for factors affecting non-adherence, as ORs were reported in less than five dimensions of adherence defined by the WHO, including conditions-related factors, health care system-related factors, social and economic factors, therapy-related factors and patient-related factors14.