Sixty-three RCTs were eligible for inclusion. Eighteen RCTs (2,720 patients) were included in the meta-analysis.
The median quality score of the 18 RCTs included in the meta-analysis was 0.71 (range: 0.53 to 0.98). The mean percentage of eligible patients that were randomised was 79%. The mean percentage of patients who were followed up was 82% (range: 36 to 100).
The funnel plot suggested the possibility of publication bias, with positive results more likely to have been published.
Main analyses (18 RCTS).
Glycaemic control.
The meta-analysis showed that interventions significantly reduced glycohaemoglobin compared with control. The effect size was 0.43 (P=0.003).
RCTs conducted in the USA showed smaller effect sizes than international RCTs (-0.26, P=0.001 versus 0.89, P non significant).
Interventions in high-quality RCTs significantly reduced glycohaemoglobin compared with control (effect size 0.51, P=0.001), but there was no significant difference between interventions for lower quality RCTs (effect size 0.38, P non significant).
RCTs with larger sample sizes showed greater reductions in glycohaemoglobin, compared with control, than smaller RCTs (-0.65, P=0.16 versus 0.31, P=0.048).
The effect size for RCTs using physicians to carry out the intervention was 0.18 (P=0.229); when using nurses and dieticians, the effect sizes were -0.71 (P=0.022) and 0.88 (P=0.043), respectively.
RCTs using interventions with groups had similar effects to interventions using individuals (-0.62, P=0.005 and 0.70, P=0.015).
RCTs of interventions that focused on medication had larger effect sizes (-0.72, P=0.032) than interventions that focused on exercise (-0.69, P=0.007), diet (-0.51, P=0.008) and blood glucose self-monitoring (-0.20, P<0.001).
Other analyses (63 RCTs).
Fasting blood glucose.
The studies showed that interventions had a non significant, moderate effect size on the mean fasting blood glucose. The difference was 12.22 mg/dL (weighted by precision), but the studies were statistically heterogeneous (P<0.05). The difference was -24.0 when weighted by sample size and 12.4 when unweighted.
Measures of glycohaemoglobin.
Only the meta-analysis of HbA1C reached statistical significance (effect size 0.52%, P=0.02). The effect sizes for glycohaemoglobin and HbA1 were not statistically significant (the results were reported).
Weight.
The studies showed a small reduction in weight with the intervention, but the reduction was not statistically significant (the results were reported).