Twenty-two RCTs (9,548 participants, range 20 to 2,091 participants) were included in the review. Nineteen studies had a parallel group design. Three studies had a crossover design. Seventeen studies were analysed according to intention-to-treat principles. No studies were double blinded. Overall Jadad scores ranged from 2 to 4 points (median 3). Median follow-up was 34 weeks.
Change in HbA1c levels: Compared to long-acting insulin analogues, biphasic insulin analogues were associated with a significantly greater reduction of HbA1c levels (WMD 0.19, 95% CI 0.04 to 0.34; seven RCTs; Ι²=58.7%). There was no evidence of significant differences in HbA1c levels when long-acting insulin analogues were compared with rapid-acting insulin analogues, human NPH insulin and GLP-1 analogues. Heterogeneity in the comparison of long-acting with biphasic insulin was reduced (Ι²=0%) with the exclusion of a study with shorter diabetes duration and lower baseline HbA1c levels and was no longer significant.
Change in fasting glucose: Long-acting insulin analogues were associated with a significantly greater reduction in fasting glucose than human NPH insulin (WMD -0.20 mmol/L, 95% CI -0.38 to -0.02; six RCTs; Ι²=0%) and GLP-1 analogues (WMD -1.35 mmol/L, 95% CI -1.64 to -1.06; four RCTs; Ι²=0%). There was no evidence of significant differences in fasting glucose when long-acting insulin analogues were compared with rapid-acting insulin or biphasic insulin analogues.
Change in postprandial glucose: Compared to long-acting insulin analogues, rapid-acting insulin analogues were associated with a significantly greater reduction in postprandial glucose (WMD 0.78 mmol/L, 95% CI 0.38 to 1.19; three RCTs; Ι²=5.3%). There was no evidence of significant differences when long-acting insulin analogues were compared with biphasic insulin analogues. There were insufficient data to make the other comparisons.
Change in weight: Long-acting insulin analogues were associated with a significantly greater reduction in weight when compared to rapid-acting insulin analogues (WMD -1.57kg, 95% CI -3.01 to -0.13; three RCTs; Ι²=73.3%) and biphasic insulin analogues (WMD -1.25kg, 95% CI -1.64 to -0.87; four RCTs; Ι²=5.3%). By comparison, GLP-1 analogues were more effective than long-acting insulin analogues (WMD 4.12kg, 95% CI 3.25 to 4.99; five RCTs; Ι²=74.8%). There was no evidence of significant differences when long-acting insulin analogues were compared with human NPH insulin.
Daily insulin doses by body weight: Long-acting insulin dosages by body weight were significantly lower than biphasic insulin analogues (WMD -0.07 U/kg, 95% CI -0.14 to 0.00; six RCTs; Ι²=87.2%) but significantly higher than human NPH insulin (WMD 0.03 U/kg, 95% CI 0.01 to 0.06; five RCTs; Ι²=99.3%). There was no evidence of significant differences when long-acting were compared with rapid-acting insulin analogues. There were insufficient data to make comparisons with GLP-1 analogues.
Incidence of total hypoglycaemia: Long-acting insulin analogues were associated with a significantly lower odds of total hypoglycaemic events than biphasic insulin analogues (OR 0.72, 95% CI 0.56 to 0.94; six RCTs; Ι²=61.2%) and human NPH insulin (OR 0.57, 95% CI 0.45 to 0.72; six RCTs; Ι²=30.2%). There was no evidence of significant differences when long-acting insulin analogues were compared with rapid-acting insulin analogues. There were insufficient data to make comparisons with GLP-1 analogues. Results for incidence of severe hypoglycaemia and nocturnal hypoglycaemia were reported.
Adverse events: Compared to GLP-1 analogues, long-acting insulin analogues were associated with a significantly reduced odds of any adverse events (OR 0.33, 95% CI 0.13 to 0.85; three RCTs; Ι²=84%), treatment-related adverse events (OR 0.04, 95% CI 0.03 to 0.06; three RCTs; Ι²=0%) and withdrawal due to adverse events (OR 0.19, 95% CI 0.05 to 0.66; four RCTs; Ι²=37.9%). There was no evidence of significant differences for the other comparisons. In some cases there were insufficient data to make the comparisons.
Results of sensitivity analyses and subgroup analyses were provided in the paper. It appeared that the authors did not assess publication bias due to the small number of studies.