“People say they want placebo-controlled trials, but I always ask them, would you be willing to die to give a p-value?”

- Janet Woodcock, FDA, 2019

Most clinical trial participants have one chance at participating in a trial and receiving an experimental treatment that may help them. When participants are randomized to a placebo group, they unknowingly lose that opportunity. Having a high—often 50%—chance of receiving a placebo is a deterrent to enrollment in trials, and some participants will drop out of studies if they believe they are not receiving the treatment. The root of Dr. Woodcock’s argument is that innovative trial designs and programs like accelerated approval can lessen this burden and get lifesaving drugs to patients faster. More needs to be done.

The medical research community must give clinical trial participants the best chance to receive a beneficial treatment. At Unearn, we’ve built a novel type of randomized controlled trial (RCT) design called a TwinRCT which uses digital twins of clinical trial participants to allow sponsors to safely improve the power and speed of their clinical trials. In this post, I’ll explain how TwinRCTs can be used to maximize the fraction of participants randomized to the treatment group.

Across clinical development, RCTs are the gold standard for evidence. For most diseases, randomization is necessary to reliably measure safety and efficacy, which means that some participants must receive placebo. Clinical trial design for RCTs needs innovation to limit patient burden. Some novel trial designs, such as external controls, are unsuitable to generate reliable evidence because they lack randomization and cannot guarantee unbiased estimates of drug effectiveness. This is reflected in FDA guidance on externally controlled trials:

“In many situations, however, the likelihood of credibly demonstrating the effectiveness of a drug of interest with an external control is low, and sponsors should choose a more suitable design, regardless of the prevalence of disease.”

- FDA, Considerations for the Design and Conduct of Externally Controlled Trials for Drug and Biological Products Guidance for Industry

We developed TwinRCTs to let sponsors run Phase 2 and 3 trials with more power and efficiency. TwinRCTs are randomized trials that leverage digital twins of the trial participants to maximize the available power for a clinical trial. Sponsors can choose to use this power to increase the confidence in their trial’s results or to reduce the number of participants in the trial and speed up their study, bringing drugs to market faster.

This isn’t just theoretical. The statistical method that underpins TwinRCTs for continuous endpoints, PROCOVA, is qualified by EMA for use in pivotal studies and is fully supported by FDA and EMA guidance. We are expanding the diseases in which we can support TwinRCTs, with a growing slate of central nervous system diseases and other therapeutic areas coming soon.

Previous applications of TwinRCTs have focused on two different types of designs. The first design is more common for small studies, where TwinRCTs add power to trials while maintaining the total enrollment.  The second type of design shrinks the size of the control group while keeping the treatment group the same size and maintaining power.  Now, we are presenting a third design that keeps the total enrollment and power the same while maximizing the fraction of participants randomized to the treatment group.

Let’s start by considering standard RCTs that do not incorporate digital twins.  If we want to assign a greater fraction of participants to the treatment group, we need to increase the total enrollment to achieve the same power.  For example, let’s suppose we are going to run a 1000-participant phase 3 trial that is randomized 1:1.  Here’s how many participants we would need for some common randomization ratios at the same power:

As we shift more participants to active treatment, the sample size requirement increases.  While higher randomization ratios may be more beneficial to trial participants, added enrollment times can delay bringing drugs to market, which does not benefit the broader patient population.

TwinRCTs add power over standard RCTs by incorporating predictions from digital twins of trial participants.  For every participant, their baseline data is used to create a digital twin, which is a comprehensive, computational prediction of the participant’s outcomes in the study.  These digital twins are created by generative AI models that we call Digital Twin Generators (DTGs).  For each endpoint in the study, the digital twins provide a prognostic score used as a covariate to increase the power of the statistical analysis.

TwinRCTs allow sponsors to maximize the fraction of participants assigned to active treatment while keeping the overall sample size the same—and while keeping power the same, too.  This optimizes the trial for participants, minimizing the number required to be randomized to the placebo group.

In these TwinRCT designs, the ability to increase randomization to the treatment group depends on a metric of performance for the models that are used to generate digital twins.  This metric is the expected correlation between the predicted (from the digital twin) and observed endpoint values in the study.  In the table below, we show the correlation along with the randomization properties for the same ratios as the case of a standard RCT that does not use participants’ digital twins.  All of these designs have the same power:

As the digital twins become a better prognostic covariate for the outcomes, the ability to randomize more participants to the treatment group increases.  You can find specification sheets on our website for each of the disease areas in which we currently support TwinRCTs.  In the majority of trial designs in these diseases, we can support 2:1 randomization and easily support 3:2 randomization with significant added power.

At Unlearn, we are committed to developing statistical methodology and trial designs that allow sponsors to conduct trials that are faster, more powerful, and more beneficial for patients.  TwinRCTs are a type of randomized controlled trials that is suitable for Phase 2 and 3 applications for adding power, reducing sample size, or increasing the randomization ratio.  In this post, we showed how TwinRCTs can increase the randomization ratio to allow more participants to receive the experimental treatment.  We’re excited to work with sponsors who want to run TwinRCTs for their studies to capture the benefits of digital twins. Contact us today to learn more.