New confidence bounds for QT studies

Abstract

The proposed guidelines for the assessment of the effect of new pharmaceutical agents on the QT interval (beginning of QRS complex to end of T wave on the electrocardiogram) are based on the maximum of a series over time of simple one-sided 95 per cent upper confidence bounds. This procedure is typically very conservative as a procedure for obtaining a 95 per cent bound for the maximum of the population parameters. This paper proposes new bounds for the maximum, both analytical and bootstrap-based, that are lower but still achieve correct coverage in the context of crossover and parallel designs for the most realistic portions of the parameter space.

Fiducial inference on the largest mean of a multivariate normal distribution

“…The other two methods attempt to account for the bias of the largest mean but have other shortcomings. Simulations suggest that our upper tailed fiducial interval tends to be shorter than the interval created by Eaton et al [4] and, when the sample size is small, slightly longer than the interval from Boos et al [1]. When the sample size is large the fiducial interval is shorter than all the competing methods.…”

“…This method produced two intervals using a normal approximation and intervals using a parametric and nonparametric bias adjusted bootstrap technique. We compared our method to those of the intersections-union, [4], and the bias adjusted method using a normal approximation from Boos et al [1]. All of the methods from Boos et al [1] had similar coverage rates, and the one discussed here created the shortest upper tailed interval.…”

“…We compared our method to those of the intersections-union, [4], and the bias adjusted method using a normal approximation from Boos et al [1]. All of the methods from Boos et al [1] had similar coverage rates, and the one discussed here created the shortest upper tailed interval. It seems reasonable that a Bayesian solution could also provide a viable method.…”

“…Another technique, seen in [4], approximates the maximum with a smooth function and uses the delta method with a bias adjustment to create a one sided upper-tailed interval. Additionally, Boos et al [1] uses a linear model to account for the variation and the bias of the largest mean. This method produced two intervals using a normal approximation and intervals using a parametric and nonparametric bias adjusted bootstrap technique.…”

A Comparison of Methods for Thorough QT Analysis for the Assessment of Cardiac Safety

“…There were some attempts to find a direct solution to the problem of a confidence interval for the maximum [4,5], but testing the null hypothesis that the placebo corrected change from baseline in QTc exceeded 10 ms for at least one time-point against the alternative that this quantity was below 10 ms for all time-points has been adopted almost universally [6].…”