Twenty-six reports (19 studies, 7,136 participants) were included in the review. Risk of bias was considered to be low in three reports, high in 14 reports and unclear in nine reports. All except three studies were double-blind.
DPP-4 inhibitors were associated with a smaller decline in HbA1c (WMD 0.20, 95% CI 0.08 to 0.32; seven trials, Ι²=60%), and a lower proportion of patients who achieved HbA1c less than 7% (RR 1.18, 95% CI 1.07 to 1.29; seven trial, Ι²=34%), which favoured metformin monotherapy.
When combined with metformin, there was a statistically significant smaller decline in HbA1c when DPP-4 inhibitors (as a second-line treatment) were compared with other hypoglycaemic drugs (overall WMD 0.12, 95% CI 0.04 to 0.20; 10 trials, Ι²=70%). Removal of poorer quality trials did not alter these results. DPP-4 inhibitors were less effective than sulphonylurea (WMD 0.07, 95% CI 0.03 to 0.11; six trials, Ι²=0%) and GLP-1 agonists (WMD 0.49, 95% CI 0.31 to 0.67; two trials, Ι²=27%) in reducing HbA1c. There was no significant difference in the comparison with pioglitazone (three trials, Ι²=40%). Achievement of the HbA1c target of less than 7% statistically favoured pioglitazone (RR 1.33, 95% CI 1.09 to 1.63; two trials, Ι²=0%) and GLP-1 agonists (RR 1.82, 95% CI 1.50 to 2.21; two trials, Ι²=0%; figures from forest plot, error in text). There was no significant difference for sulphonylureas (five trials, Ι²=26%).
DPP-4 inhibitors as monotherapy were less effective in reducing body weight than metformin (WMD 1.50, 95% CI 0.90 to 2.11; five trials, Ι²=74%). When combined with metformin, DPP-4 inhibitors achieved significantly greater body weight reduction than sulphonylurea (WMD -1.92, 95% CI -2.34 to -1.49; four trials, Ι²=69%) and pioglitazone (WMD -2.96, 95% CI -4.13 to -1.78; two trials, Ι²=79%). There was no significant difference in the comparison with GLP-1 agonists (two trials, Ι²=0%).
Risk of adverse events was generally lower for DPP-4 inhibitors (results reported in the paper). There was no evidence of publication bias.