Quantifying the 3 Biases That Lead to Unintentional Overestimation of the Blood Pressure-Lowering Effect of Renal Denervation

Background: Studies of renal denervation report disparate results. Meta-analysis by trial design may allow quantitative estimation of sources and magnitude of biases in denervation studies.

Methods and results: One hundred forty nonrandomized, 6 randomized open-label, and 2 randomized blinded studies were analyzed for 2 outcomes: (1) blood pressure changes for nonrandomized, open-label randomized, and blinded studies; and (2) quantification of 3 biases potentially contributing to apparent antihypertensive effects: (a) regression to the mean, (b) asymmetrical data handling, and (c) true blood pressure drops caused by something other than the tested therapy (confounding). Nonrandomized studies and open-label randomized trials reported large reductions in office blood pressure of 23.6 mm Hg (95% confidence interval [CI], 22.0 to 25.3) and 29.1 mm Hg (95% CI, 25.2 to 33.1 mm Hg), respectively. They reported smaller reductions in ambulatory blood pressures (11.2 mm Hg; 95% CI, 10.0 to 12.4). The blinded trials found no significant reduction in blood pressure (2.9 mm Hg; 95% CI, -0.4 to 6.3). Analyses of these data indicate the magnitude of the 3 potential sources of bias to be regression to the mean, -1.01 mm Hg (95% CI, 4.24 to -6.27); asymmetrical data handling, -10.8 mm Hg (95% CI, -8.77 to -12.87); and confounding, -8.3 mm Hg (95% CI, -4.73 to -11.83).

Conclusions: Increasingly bias-resistant trial designs report effect sizes of decreasing magnitude. This disparity may be caused by asymmetrical data handling and confounding (eg, increased drug adherence). If these differences are caused by trial design and not by some other differences in patients or procedures, which happen to match the trial design, then randomization alone is not enough: blinding is also needed. This has broad implications across trials of medications and devices.