Pantelis Vlachos, Principal, Strategic Consultant at Cytel, conducted a webinar to introduce the capabilities of East AlloyTM. East Alloy is a new East environment that enables rapid access to innovation with the trust and support you have come to expect from Cytel. The cloud-native software makes it practical to apply computationally intensive Bayesian methods. Download the brochure to learn more.
This blog is a part of the new blog series on technology and Bayesian decision-making by Pantelis. Continue reading to learn about the methods and capabilities, such as, Bayesian meta-analytic priors, Bayesian MAMS, adaptive dose-finding and others, available to all East Alloy users.
Cytel’s new web-native extension of East, East AlloyTM, makes it practical and sustainable to adopt innovative and computationally intensive designs. This software as a service (SaaS) delivery model expedites release of innovation from Cytel’s incubator and makes it easier to deploy and access new methods. It also reduces the risk and burden of adopting open-source innovation by curating and verifying public packages; strengthening them with commercial grade UI, training, support and maintenance. This extension of East will be accessible anytime, anywhere, without burdening your IT group.
Bayesian methods are not new to East. Bayesian probability of success calculations through the concept of assurance has been available in East since early 2013 for fixed designs. In group sequential settings, Bayesian predictive power has been reported when uncertainty is assumed for the treatment effect in the form of a prior distribution. Moreover, Bayesian approaches have also been used in dose escalation designs like the BLRM or the CRM, which are alternatives to the classical algorithmic methods like the 3 + 3 design.
Methods available in East Alloy
In East Alloy we are complementing the capabilities of East with a couple of different modules. These modules generally deal with sequential types of designs including tools for group sequential and MAMS clinical trials with normal, binomial, or time-to-event endpoints.
Bayesian meta-analytic priors
Creation of prior distribution is done through Meta-Analytic-Predictive (MAP) method which uses historical metadata synthesized to historical studies. Both studies can have considerable heterogeneity which dictates how much information will be shared from these studies for the design and analysis of the future studies. There are different types of endpoints that can be considered; currently Normal, Binomial and Survival endpoints are available.
The prior will be synthesized from historical studies and will be converted to a parametric representation through a mixture of Normals. The number of these mixture components will be selected automatically using the AIC criterion and we can also manually specify the components for that mixture. The benefit of using this method is that it helps in reduction of within-study placebo-treated number of subjects and increase of study power. Additionally, easy communication of prior information through parametric mixture density leads to fast and accurate analytical procedures to evaluate properties of trial designs.
Bayesian Group Sequential Capabilities
This module is a simulation platform for Bayesian group sequential designs. Instead of using spending functions as in the classical approach to dictate how a trial can stop because of efficacy or futility, it uses multiple decision-making criteria for early stopping either for efficacy or futility, which are based on thresholds of the posterior distribution. The operating characteristics which are resulting from the simulation capabilities will include probabilities of success and futility at each interim analysis, and the expected sample size. The advantage here is that multiple such efficacy and futility stopping criteria based on these posterior probabilities can be created.
Bayesian Survival Capabilities
Within this simulation module for Bayesian Group Sequential designs, specific capabilities tailored for time-to-event endpoints are available. Blinded or unblinded data to predict landmark event times for this time-to-event endpoint can be used, utilizing different types of distributions like Exponential, Weibull or Piece-wise exponential Modelling.
In the case where the unblinded data are censored, this process will complete the censored data based on posterior draws of the hazard function. Posterior probabilities for the hazard ratio and predictive probability of success (POS), from both Bayesian and Frequentist points of view, can be calculated at the end of the trial.
Bayesian Multi-arm Multi-Stage Capabilities
This module can run Multi-Arm Two Stage design simulations. For now, only Normal endpoints can be entertained, and the sample size can be calculated per arm, given the initial information about the anticipated treatment effect. The simulations can help in efficiently evaluating the multiple treatments as ineffective treatments are quickly eliminated through the use of stopping or elimination criteria based in the posterior distribution.
This module will be released later this year. The capabilities will include both Frequentist and Bayesian methods. It will help with adaptive dose-finding and allocating patients dynamically to optimal doses. It will also help in decreasing the allocation to the ineffective or unsafe doses and thus, enable efficient estimation of the dose response relationship and provide options to stop for efficacy and futility. In the first release of the software several Frequentist type of methods like the T-test up and down design as well as the Maximizing design will be included. From the Bayesian front the Emax model will be included. At a later release, capabilities dealing with Bayesian 4PL and NDLM will be added.
To learn more about these methods that are available in East Alloy and to watch a demonstration by Pantelis, access the on demand webinar.
About Pantelis Vlachos
Pantelis is Principal/Strategic Consultant for Cytel, Inc. based in Geneva. He joined the company in January 2013. Before that, he was a Principal Biostatistician at Merck Serono as well as a Professor of Statistics at Carnegie Mellon University for 12 years. His research interests lie in the area of adaptive designs, mainly from a Bayesian perspective, as well as hierarchical model testing and checking although his secret passion is Text Mining. He has served as Managing Editor of the journal “Bayesian Analysis” as well as editorial boards of several other journals and online statistical data and software archives.