Increasing efficiency of Randomised Clinical Trials via Bayesian Borrowing and Prognostic Covariates

Randomised Clinical Trials are considered to be gold standard in research, since they eliminate systematic bias between comparison groups. However, they are expensive and time consuming to carry out.

In order to maximise the benefit from trials, researchers increasingly try to combine the information from a new trial with the information already gained from a previous trial to supplement the data in a new trial.

This research will include two components in the combination of new and historical data. The first component is the Bayesian Borrowing, in which results from earlier trials are converted into informative priors on the parameters for the placebo arm of the current trial. This reduces the uncertainty on those parameters, increasing the efficiency of the current analysis.

The second component is to adjust for a Prognostic Covariate, which can be a traditional predictive covariate, or a "synthetic covariate" estimated using a flexible model based on the earlier trial, using a data-driven Machine Learning approach. Adjusting for Pronostic Covariates aims to reduce bias between historical and current studies.

This project aims to combine these approaches, using the concept of synthetic prognostic covariate to construct more efficient and informative priors for Bayesian borrowing.

Expected outcomes are new methods and software for using historical information in current trials involving rare and hard to study endpoints. This is achieved by developing a method which fully integrates more available external information into the current trial.

The goal is to achieve a better chance of making the correct decision to proceed to phase III trials with fewer patients in the trial, thereby accelerating the regulatory approval.