Fifty-six trials were included in the meta-analysis. Trial quality varied. All studies reported equality of care and most had similar baseline characteristics. Adequacy of randomisation and reporting of withdrawals and drop-outs was present in more than half of the included trials. Blinding was only present in five trials. Allocation concealment was recorded in 18 trials. Intention-to-treat analysis was used in around half of the trials. Follow-up ranged from two to 48 months.
Compared with oral antihyperglycaemic agents, a statistically significant effect in favour of insulin was reported for reduced HbA1c (MD -0.62, 95% CI -0.97 to -0.26, Ι²=74.2%; 13 trials). Weight gain was significantly higher with insulin (MD 2.60, 95% CI 1.31 to 3.89, Ι²=78.2%; eight trials). There was no significant difference between groups for risk of hypoglycaemia.
Premixed and basal-bolus regimens were statistically more effective in reducing HbA1c (MD 0.30, 95% CI 0.03 to 0.57; 13 trials, Ι²=93.1%) than basal regimens (MD 0.33, 95% CI 0.03 to 0.63; eight trials, Ι²=52%). Basal regimens were more effective in reducing weight gain (basal versus premixed MD -1.03, 95% CI -1.94 to -0.13; 12 trials, Ι²=65.5%) and basal versus basal-bolus MD -1.41, 95% CI -2.05 to -0.77; eight trials, Ι²=44.9%). Risk of hypoglycaemia was statistically significant only for premixed versus basal regimens (OR 1.31, 95% CI 1.15 to 1.49; eight trials, Ι²=0%; from appendix 5). There were no statistically significant differences between groups for HbA1c between intermediate and long-acting or between morning and evening basal insulin (where there were sufficient data). Intermediate insulin was favoured for reduced weight gain (MD -0.20, 95% CI -0.22 to -0.18; four trials, Ι²=0%), but the risk of hypoglycaemia was higher (OR 1.70, 95% CI 1.19 to 2.41; five trials, Ι²=45.1%) than long-acting insulin.
Other pooled analyses of insulin versus insulin plus oral antihyperglycaemic agents (and results for vascular outcomes, lipid profile and quality of life) were reported in the paper.