Incident diabetes in clinical trials of antihypertensive drugs: a network meta-analysis

Summary Background The effect of different classes of antihypertensive drugs on incident diabetes mellitus is controversial because traditional meta-analyses are hindered by heterogeneity across trials and the absence of trials comparing angiotensin-converting-enzyme (ACE) inhibitors with angiotensin-receptor blockers (ARB). We therefore undertook a network meta-analysis, which accounts for both direct and indirect comparisons to assess the effects of antihypertensive agents on incident diabetes. Methods We undertook a systematic review up to Sept 15, 2006, and identified 48 randomised groups of 22 clinical trials with 143 153 participants who did not have diabetes at randomisation and so were eligible for inclusion in our analysis. 17 trials enrolled patients with hypertension, three enrolled high-risk patients, and one enrolled those with heart failure. The main outcome was the proportion of patients who developed diabetes. Findings Intitial drug therapy used in the trials (and the number of patients with diabetes of the total number at risk) included: an ARB (1189 of 14 185, or 8·38%), ACE inhibitor (1618 of 22 941, or 7·05%), calcium-channel blocker (CCB, 2791 of 38 607, or 7·23%), placebo (1686 of 24 767, or 6·81%), β blocker (2705 of 35 745, or 7·57%), or diuretic (998 of 18 699, or 5·34%). With an initial diuretic as the standard of comparison (eight groups), the degree of incoherence (a measure of how closely the entire network fits together) was small (ω=0·000017, eight degrees of freedom). The odds ratios were: ARB (five groups) 0·57 (95% CI 0·46–0·72, p