[2310.18027] Bayesian Prognostic Covariate Adjustment With Additive Combination Priors


Obtain a PDF of the paper titled Bayesian Prognostic Covariate Adjustment With Additive Combination Priors, by Alyssa M. Vanderbeek and Arman Sabbaghi and Jon R. Walsh and Charles Okay. Fisher

Obtain PDF
HTML (experimental)

Summary:Efficient and speedy decision-making from randomized managed trials (RCTs) requires unbiased and exact remedy impact inferences. Two methods to handle this requirement are to regulate for covariates which are extremely correlated with the end result, and to leverage historic management info through Bayes’ theorem. We suggest a brand new Bayesian prognostic covariate adjustment methodology, known as Bayesian PROCOVA, that mixes these two methods. Covariate adjustment in Bayesian PROCOVA relies on generative synthetic intelligence (AI) algorithms that assemble a digital twin generator (DTG) for RCT contributors. The DTG is educated on historic management information and yields a digital twin (DT) chance distribution for every RCT participant’s consequence underneath the management remedy. The expectation of the DT distribution, known as the prognostic rating, defines the covariate for adjustment. Historic management info is leveraged through an additive combination prior with two elements: an informative prior chance distribution specified primarily based on historic management information, and a weakly informative prior distribution. The combination weight determines the extent to which posterior inferences are drawn from the informative part, versus the weakly informative part. This weight has a previous distribution as effectively, and so your complete additive combination prior is totally pre-specifiable with out involving any RCT info. We set up an environment friendly Gibbs algorithm for sampling from the posterior distribution, and derive closed-form expressions for the posterior imply and variance of the remedy impact parameter conditional on the burden, in Bayesian PROCOVA. We consider effectivity good points of Bayesian PROCOVA through its bias management and variance discount in comparison with frequentist PROCOVA in simulation research that embody totally different discrepancies. These good points translate to smaller RCTs.

Submission historical past

From: Arman Sabbaghi [view email]
Fri, 27 Oct 2023 10:05:06 UTC (1,487 KB)
Thu, 9 Nov 2023 17:39:59 UTC (3,610 KB)
Thu, 23 Nov 2023 00:57:01 UTC (3,609 KB)
Wed, 28 Feb 2024 18:57:15 UTC (3,605 KB)

Supply hyperlink


Please enter your comment!
Please enter your name here