OBJECTIVES : Network meta-analysis is an evidence synthesis method by using direct and indirect evidence for all the treatment comparisons of interest. In many network meta-analyses, some studies have markedly different characteristics from the others. The inclusion of these “outlying” studies might lead biases and possibly produce misleading conclusions. The purpose of this study is to illustrate the influence of these outlying studies from an empirical evidence, a network meta-analysis of antihypertensive drugs (Sciarretta et al., 2011; Arch Intern Med 171: 384-394).

METHODS : We conducted network meta-analysis for 26 clinical trials (N=223,313) that compared α-blockers (AB), ACE inhibitors (ACEI), angiotensin II receptor blockers (ARB), β-blockers (BB), calcium channel blockers (CCB), conventional treatment (CT), diuretics, and placebo. The outcome was the incidence of heart failure. We assessed inconsistencies on the network of evidence using the Dias’s side-splitting method. We conducted sensitivity analyses that excluded candidates of outlying studies, which had markedly different characteristics.

RESULTS : Through the side-splitting method, the direct and indirect evidence for the comparison of ARB vs. CT was significantly inconsistent (direct OR (odds-ratio): 1.49, indirect OR: 0.91, P = 0.027). After excluding three trials that involved ARB vs. CT (VALUE, E-COST, Jikei Heart Study), the treatment ranking has changed significantly. In the original analysis, it was diuretic-ACEI-ARB-CT-CCB-BB-placebo-AB, while in sensitivity analysis it was diuretic-CT-ACEI-CCB-ARB-BB-placebo-AB. The efficacy estimate of ARB got smaller, and that of ACEI got larger in the sensitivity analysis. In addition, CCB vs. placebo was significantly different (OR 0.79, 95%CI 0.64, 0.99; P=0.038) only in sensitivity analysis.

CONCLUSIONS : Outlying studies possibly have influences to the overall conclusions of network meta-analyses, thus they should be carefully evaluated in practices. In addition, extreme profiles might be caused by certain biases such as industrial sponsorship bias. Sensitivity analyses are important to circumvent misleading conclusions for these comparative effectiveness reviews.