Twenty-six RCTs met the inclusion criteria of which 23 trials (2,117 patients) provided data for the meta-analyses. None of the included trials were rated as low for risk of bias across all domains. Only two RCTs described all aspects of adequate sequence generation, allocation concealment, and blinding of participants and investigators.
Sixteen RCTs (1,627 patients) reported on all-cause mortality but only five reported any deaths. There was no significant difference between metformin plus insulin versus insulin alone (RR 1.30, 95% CI 0.57 to 2.99; Ι²=0%). Trial sequential analysis indicated that 2.93% of the required information size to detect or reject a 30% reduction in relative risk was accrued for all-cause mortality.
Fifteen RCTs (1,498 patients) reported on cardiovascular mortality but only three reported any deaths. Differences between treatment groups were not statistically significant (RR 1.70, 95% CI 0.35 to 8.30; Ι²=0%). Trial sequential analysis indicated that 0.65% of the required information size to detect or reject a 30% reduction in relative risk was accrued for cardiovascular mortality.
In a fixed-effect model, but not in a random-effects model, risk of severe hypoglycaemia was significantly higher with metformin plus insulin than with insulin alone (RR 2.83, 95% CI 1.17 to 6.86; Ι²=43%).
In a random-effects model, metformin plus insulin treatment was associated with reduced glycated haemoglobin (HbA1c), reduced weight gain and reduced insulin dose compared with insulin alone. Trial sequential analysis indicated that there was sufficient evidence to support a HbA1c reduction of 0.5%, 1kg lower weight gain and 5U/day lower insulin dose.
Results of other analyses were reported.