A Bayesian Design for Agreement Studies for Medical Devices

Agreement studies for a new medical device measuring a certain physiology aim to validate this device with respect to a reference or a predicate device. The traditional (Frequentist) approach is to use the Bland-Altman plot along with confidence intervals for the Limits of Agreement to test the null hypothesis for poor agreement. In the absence of prior data it might be difficult to properly power these studies and may increase the risk of an under or overpowered trial.  

In this article, we explore how adaptive design can help in mitigating this risk. We also analyze the use of a Bayesian adaptive design in absence of established adaptive methodology for agreement studies. In such cases, we propose the use of the Bayesian predictive power for making interim decisions such as deciding on the final sample size. 

We illustrate our proposed design with an example using simulated data and using weakly informative priors for the parameters of interest – the bias and the variability.