assurance-package: Compute assurance

In scientific-computing-solutions/assurance: Perform assurance computations for a clinical trial.

Compute assurance

Description

Perform assurance computations for a clinical trial, comparing means, proportions, or hazards.

Details

Estimate the unconditional probability of a study achieving its desired goal by averaging the power of the study over the prior distribution of the treatment effect. This is termed ‘assurance’ by O'Hagan et al (2005), but also see Spiegelhalter et al (200X) and the references therein.

The information provided to the function is used to construct a prior density. For Gaussian data, the marginal density of the variance is taken to be scaled inverse \chi^2 and the conditional density of the mean is taken to be Gaussian (see pages 74 - 75 of Gelman et al).

For binomial data, the marginal proportions are taken to be beta distributions resulting from using Jeffreys' prior. That is, p_1 \sim B(x_1 + 0.5, m_1 - x_1 + 0.5) and similarly for p_2.

For time-to-event data, the log of the hazard ratio is taken to be Gaussian with known variance 1/x_1 + 1/x_2 (see, for example, Carroll, 2003). Treating the variance as fixed fails to account for some uncertainty, but in practice the loss is small because it is usual for the number of events to be prespecified as part of the study design.

Author(s)

Harry Southworth, Paul Metcalfe

Maintainer: Paul Metcalfe <[e-mail]>

References

A. O'Hagan, J. W. Stevens and M. J. Campbell, Assurance in clinical trial design, Pharmaceutical Statistics,4, 187 - 201, 2005

D. J. Spiegelhalter, K. R. Abrams and J. P. Myles, Bayesian Approaches to Clinical Trials and Health-care Evaluation, Wiley, 2003

A. Gelman, J. B. Carlin, H. S. Stern and D. B. Rubin, Bayesian Data Analysis (Second Edition), Chapman & Hall/CRC, 2004

K. J. Carroll, On the use and utility of the Weibull model in the analysis of survival data, Controlled Clinical Trials, 24, 682 - 701, 2003

scientific-computing-solutions/assurance documentation built on June 28, 2023, 12:31 p.m.