Why can’t we agree on how to define digital twins in healthcare?

Why can’t we agree on how to define digital twins in healthcare?

The term “digital twins” has found its way across the healthcare industry to describe everything from AI-based simulations of disease progression to mechanistic models of cellular function. With so many meanings and applications in circulation, it seems that people aren’t operating on a clear definition of what digital twins are — as well as what they are not. To clarify the term’s usage in healthcare, we recently brought together a panel of industry leaders from the digital twin frontier for an Endpoints webinar. The panelists included Charles Fisher, founder and CEO of Unlearn; François-Henri Boissel, Co-founder and CEO of Nova Discovery; Jonathan Baptista, Co-founder and CEO of DeepLife; and Simon Sonntag, founder, and CEO of Virtonomy. Read some of the highlighted excerpts below and stream the on-demand webinar to listen to the full conversation. 

Is a single definition for “digital twins” possible?

Charles: I hope so, and I think that it should be. The key aspect when we talk about digital twins broadly is that there are some core things we think of digital twins as providing. People are often using the term, unaware that it's borrowed from engineering. What I think of as digital twins include things like a digital twin of a person's heart, which should be a simulation of how a particular person's heart operates. You can create digital twins of people. You can create digital twins of devices. But I've also heard people referring to digital twins as a patient's electronic health record, as in spreadsheet data. They assume if it's in a computer, it's digital, so it's their digital twin, but that is a very different usage than how that term has been used in engineering. That use, which is becoming more widespread within the field, is detrimental to people's understanding of digital twins. 

Simon: I really like the definition of digital twins that was provided by IBM as an environment of linked data and models that reproduce an accurate virtual representation of real-world entities of processors and uses simulation, machine learning, and reasoning to help decision-making. 

Jonathan: To define digital twins, I would say it involves numerical modeling of a system that we can leverage to mimic the temporal evolution of its real-world counterpart. And in that being said, there are two things that are important — causality in the model and temporality. 

Francois: I fully agree with Charles’ opening comments. There are many different ways of approaching the digital twin space within healthcare.

Who, how, and what needs to be done to get to a consistent definition of the approach so that a digital twin doesn't get lost in the evolving, digital health/health tech, ecosystem?

Charles: This is step one—to create a digital twin consortium in which we try to come up with some sort of broad definition of digital twins.

Simon: It’s useful to have to arrive at a definition for digital twins because it's so broad and used across so many disciplines. It may be necessary to have subcategories such as “digital patient twins” or “digital devices” to define digital twins in the healthcare domain. 

Charles: That's actually a fantastic point. The term “digital twins” is meaningless by itself—you have a digital twin of something, and you have to say that whole phrase because it is a digital twin of something. For instance, we have digital twins of people and digital twins of cells, and digital twins of tissues. If you just say, “digital twin,” then it’s not clear what you're talking about. 

Some people use digital twins for diagnostic purposes, others for prognostic. Do you think digital twins accurately describe both?

Charles: Digital twins is a broad term, but there is some particular thread that connects all of the definitions, and that thread is simulation. And so there are two different ways that you could think about doing that. The first is diagnostic. For example, you could have a patient you've measured certain kinds of things about, and you may want to understand more about that patient's current state. So you're using a computer to peer into and ask questions about these unseen things. The other thing is using a computer model to simulate not only what will potentially happen in the real world in the future but also in the multiverse in the future — different scenarios. For instance, what would happen if you give a patient’s digital twin both treatment A and treatment B — what would the outcome be? So those are two different ways that you can think about what digital twins are doing and how they can be used. Human experimentation is time-consuming and sometimes impossible. Sometimes you literally want to know about something that you can't measure; we can do all of those things on a computer. 

While the panelists may not have reached a singular definition for ‘digital twins,” the discussion elucidated the need for further dialogue as the diversity of digital twin applications continues to evolve. The panelists also debated misconceptions surrounding synthetic versus simulated data and the ethics of utilizing digital twins to run completely simulated clinical trials in the future. Today, Unlearn’s digital twin technology is laying the foundation for that future. Unlearn is the only company creating prognostic digital twins that enable smaller control arms and produce regulatory-qualified evidence equal to that of larger randomized controlled trials. The company’s vision is to advance AI to eliminate trial and error in medicine. 

To view or add a comment, sign in

More articles by Unlearn.AI

Insights from the community

Others also viewed

Explore topics