Bayesian Borrowing and Real World Data: The Fundamentals

August 21, 2020

As uses of real world data become more familiar for trial design and regulatory submission, sponsors might become more interested in techniques related to Bayesian borrowing. Sophisticated uses of this technique have been applied to COVID-19 vaccines development and complex oncology trials. Simpler versions can be used to optimize across existing datasets, design trials with lower enrollment targets, accelerate the time to regulatory submission, and strengthen the power of a trial. Here we provide an overview of the potential benefits.

Data Optimization: As datasets become more easily accessible, borrowing from historical data can help to inform or augment data collected during an ongoing trial. Earlier trials can supply data that can be combined to create an informed prior, which is then updated during the course of a trial. Sometimes this is obtained simply by pooling data, though more typically such data is then adjusted with a weight (to create what is known as

a power prior). A meta-analytic prior can also be used to combine historical datasets with new data.

Lower enrollments: When historical data exists for a control group, it can be useful to decrease new enrollments into that group and save patients for the treatment arm. Sometimes, this might be the ethical thing to do, such as in an oncology trial. Alternatively, ensuring higher prospects of enrolling into the treatment arm might prevent patients from dropping out of the trial.

A recent paper by Cytel statisticians reveal that Bayesian Dynamic Borrowing can be used to reduce trial enrollment substantially, without compromising statistical results. One simulation revealed that control group enrollment could be reduced by up to 80% without affecting the findings of the trial. [1]

Accelerate Time to Submission:

The combination of beginning with an informed prior and reducing enrollments can lead to substantially accelerated times to submission. Examples include AstraZeneca’s Tagrisso which received approval within three years of the first enrolled patient, and the Ibrance study which required fewer than 25 patients.

Strengthen the Power of a Trial:

Perhaps the most important scientific reason to use historical or external data, is the potential to strengthen the power of a trial. Whether dealing with rare disease studies, where sample populations might have to be small, or illnesses so devastating that there are sound reasons to minimize enrollment into a control group, using historical data can often augment new findings. Whether or not such increases in power occur during a given trial require an evaluation of context. Working with a statistical consultant to determine your specific needs could help.


[1] Dron L, Golchi S, Hsu G, Thorlund K. Minimizing control group allocation in randomized trials using dynamic borrowing of external control data - An application to second line therapy for non-small cell lung cancer. Contemporary Clinical Trials Communications 2019; 16: 100446.