Debunking Top 5 Myths about Digital Twins in Clinical Trials

Addressing misconceptions as industry becomes more familiar with AI and ML.

Artificial intelligence and machine learning (ML) have the potential to transform clinical research. They hold the key to generating digital twins—also known as virtual patients—to act as external control arms in clinical trials. AI and LM make it possible to generate datasets based on existing ones to apply the power of realistic simulation to human modeling. In turn, studies based on digital twins hold the promise of accelerating approvals, expanding labels, and speeding up research.

Digital twins are already being used to advance clinical research today. In 2017, the FDA approved Brineura for CLN2 disease (a form of Batten disease) based on data from 22 pediatric patients studied in a single-arm trial versus an independent external control group data with 42 untreated patients. Meanwhile, in Europe, the EMA granted the lung cancer drug Alecesna an expansion of label based on a study employing digital twins. Using an external data set of 67 patients—rather than a Phase III trial—to prove efficacy sped up approval by 18 months in 20 European countries—a life changing difference for sick patients.

Digital twins offer the best of both worlds: the reliability of a randomized controlled trial (RCT) along with the efficiency of a synthetic control arm. Despite this, some myths persist about using digital twins in clinical trials. Here’s five common ones—and what you should know about each.

Myth 1: A digital twin is an exact replica of the human counterpart

In order to create a digital twin, existing data is synthesized to create new “patients”. The availability of large historical datasets of longitudinal patient information and artificial intelligence (AI) technologies make digital twins possible. Since patient data is not mapped one-on-one, it is not an exact replica of any particular patient.

By reusing statistically correlated distributions from previous projects, the new data 'behaves' the same way in analyses as the original data. The results are more than a simple medical profile, but creates matches based on demographics, lab tests, and biomarkers. This means there is less of a need to find similar actual pairs of people—actual twins, even—to run tests and controls. A digital twin mirrors the operation of some elements of a trial, performing the testing of these conditions “in silico”. Each resulting digital twin can then predict how that “patient” would likely evolve over the course of the trial if they were to be given a placebo.

Myth 2: Digital twins will fully replace human participants in clinical trials

Digital twins are not meant to replace human participants in a clinical trial, although they may reduce or eliminate the need for participants to be given placebos. Digital twins can therefore solve a longstanding ethical issue within clinical research of placebo arms.

There are limitations to when digital twins can be used within clinical research. Since they rely on existing clinical trial datasets to build personalized machine learning models for a specific disease, there must be adequate quality data to pull from. So if researchers are studying a rare disease (or one which few studies have been conducted about so far), there may not be enough data available to create digital twins. Therefore, synthetic control arms using digital twins require that the studied disease is predictable and that its standard of care is well-defined and stable.

Myth 3: There are no risks, since there are no humans involved in the study

Using digital twins within clinical research carries some inherent risks. Although the FDA and other regulators have signaled interest in encouraging life science companies to invest in these new methodologies, there are still regulatory concerns.

First, there is the issue of obtaining informed consent. For example, will patients providing data for one clinical trial be asked for consent before their data is reused in a different clinical trial? The approach to data reusability may vary between EU and US regulators. Regardless, digital twins may raise new ethical concerns about patients’ rights to decide what happens with their data after a trial ends.

Also, if used incorrectly, digital twins could cause researchers to unintentionally compare apples to oranges. Researchers need to ensure they are comparing similar patient populations and aren’t relying on too narrow of a patient population to create any digital twins. To avoid this risk, researchers need to know where any data used to create any digital twins came from. Traceability of reused data is therefore paramount to receiving regulatory approval of a new therapy.

Myth 4: Digital twins will increase clinical study complexity

Digital twins may significantly simplify studies. The potential benefits of incorporating into clinical trials are massive, especially when it comes to reducing patient burden. Reusing previously collected data from patients—while eliminating the need for patients in a placebo arm—reduces recruitment needs and may represent significant cost savings.

Studies built using digital twins may also increase study diversity and amplify the statistical power of a single study. With reduced requirements for patients and sites, digital twin studies can target selection of “real” trial subjects to focus more on indications for personalized products based on simulation data and a richer real-world dataset. For these same reasons, this model may also accelerate clinical trial timelines so potential new therapies can reach patients faster.

Myth 5: Any data will do to create a digital twin

The statement of “garbage in, garbage out” holds true with digital twins. It’s not simply a case of throwing enough data at a model to create digital twins within a study—low quality data will skew results. According to a 2020 study published in Clinical Epidemiology, “it is important to consider the process of the original data collection, compare the populations of the datasets that are being compared, and the reliability and comprehensiveness of the datasets.”

Valid conclusions can only be drawn from digital twins generated from relevant, high quality datasets. Digital twins should therefore be informed by external control data from studies that are 1) recent, 2) answering as similar a question as possible, and 3) using similar designs and implementation processes.

It will take time and effort to test these new models sufficiently against more conventional trial processes to demonstrate viability. At this point, the widespread use of digital twins within clinical trials remains largely hypothetical, with many challenges for organizations who want to experiment. But the work to overcome these challenges will be well worth it since the results of these studies have the power to move clinical research forward by leaps and bounds.

In the future, digital twins fueled with sufficient digital data may even be used to model the active arm of future studies. The model will take years in development and refinement for use on trials. However, the potential is clear. As companies continue to experiment with ways to deliver value in research, preclinical studies, and conventional trial processes, we can expect to see the full power of digital twins unleashed.

Derk Arts, CEO & Co-Founder, Castor