Digital Twins Could Be the Crystal Balls of Healthcare

There’s a scene in the iconic 1995 film “Apollo 13” in which NASA aerospace engineer Ken Mattingly (played by actor Gary Sinise) is awoken in the middle of the night. A catastrophic oxygen tank explosion has just damaged Apollo 13’s engine and fuel cells, leaving the spacecraft floating above the far side of the moon, with “barely enough electricity to power a coffee machine for nine hours.”

Mattingly puts on his shirt and tie and makes his way to an Apollo simulator module to see whether he can help figure out a way to get the ship back safely.

“I need the sim cold and dark,” Mattingly orders. “Give me the exact same conditions they’ve got in there now.”

The scene marks a turning point in the film, but the same concept might also mark a turning point for healthcare. NASA’s use of simulators to solve the Apollo 13 riddle is now remembered as the first use of “digital twins.” The space agency used real-world data to power simulations that allowed them to better understand why the accident occurred and how to prevent similar accidents in the future. Now, somefive decades later, scientists are attempting to construct digital twins of human beings, with the goal of predicting the progression of individual patients’ diseases, and modeling whether and why particular therapies might help cure them.

The front lines of a new frontier

Jun Deng, Ph.D., is in the vanguard of the effort. He helped launch the Digital Twins for Health Consortium after signing up for a workshop held by the National Cancer Institute and the U.S. Department of Energy in 2020. The goal of the workshop was to bring scientists together to build a cancer patient digital twin.

Jun Deng, Ph.D.

“I was wowed by the whole concept,” says Deng, a professor of therapeutic radiology at the Yale School of Medicine. Yet, he quickly understood the scope of the challenge before his group. The team was doing nothing less than attempting to model perhaps the most complicated system imaginable — the human body. “It’s unknown territory,” he says. “...So anything we do here will be pioneering.”

In the years since, digital twin pioneers have set off down a number of paths. Some efforts aim to build digital twins of individual organs of particular patients. Others are pursuing the goal of creating digital versions of a patient’s entire body.

In the case of Unlearn, an artificial intelligence (AI) firm in San Francisco focused on healthcare, digital twins are being used to optimize clinical trials. The company developed a statistical methodology that utilizes digital twins to predict a participant’s likely clinical outcome had they been assigned to a trial’s control group. By doing so, investigators can more precisely understand the impact of the therapy being evaluated.

Jess Ross, Ph.D.

“Where humans may make outsized judgments regarding the impact of particular factors due to their social and environmental context, the power of AI lies in its ability to reliably gauge the relative impact of multiple pieces of (occasionally conflicting) information to the clinical trajectory of a particular person, pulling the signal from the noise,” says Jess Ross, Ph.D., the company’s senior government affairs lead.

Earlier this year, Unlearn announced that it had received comments from the FDA indicating that the company’s proposed digital twin methodology aligns with the agency’s current guidance on statistical methodology, essentially giving the methodology a seal of validity.

The challenge of behavior

Unlearn’s progress with the FDA represents a vote of confidence for digital twins, but like so much in the era of big data, digital twins are only as valuable as the data used to create them. One challenge in making them reliable is the fact that their subjects — human beings — are not always reliable themselves. Human health and the efficacy of a particular therapy can be significantly impacted by factors such as missing doses of medicine or engaging in unhealthy behaviors.

Reinhard C. Laubenbacher, Ph.D., a professor of systems medicine at the University of Florida, has studied digital twins extensively and is involved in a project to use them in the context of pulmonary hypertension. He says the problem of human behavior can be overcome. “It is certainly possible to incorporate imperfect adherence to medication protocols if there are data about prevalence, frequency or individual behaviors,” he says. One example, he says, is “self-reporting” pills, which are designed to send a signal when they enter the stomach, enabling a truly accurate indication of whether and when patients take their medications.

Deng notes that because digital twins are modeled after particular patients, the models can be personalized in ways that account for specific factors, such as a particular person’s risk factors or habits. But he adds that there will be an element of personal responsibility with digital twins. If patients conceal health data or do not follow a plan of care, they might affect their twin’s accuracy while also affecting their own health. “It’s your data and your model,” he says.

Ethical questions linger

Still, digital twins are not solely reliant on an individual’s data. They are also built with the help of AI and machine learning trained on big data. All of this raises certain ethical questions, says Brandon Ferlito, M.Sc., a bioethicist at Ghent University, in Belgium.

He argues that although technology like digital twins may ultimately improve human health, there is a risk that technological momentum will steamroll important ethical considerations, including a lack of representation in the data used to build the models.

“If the data mainly [come] from middle-aged White male [patients], the DT [digital twin] may not accurately predict disease progression or treatment responses for women or people of different ethnic backgrounds,” says Ferlito, who is completing his Ph.D. at the Bioethics Institute Ghent.

Privacy and consent are also important issues, he says. For instance, it may prove difficult to ensure patients have a clear understanding — and control — of what their data will be used for and by whom it will be used. Furthermore, he worries the technology could be limited to wealthy patients and regions and thus exacerbate existing health disparities.

“We also need to keep talking to patients, healthcare professionals and ethicists to navigate the ethical issues and create policies that promote fairness and inclusivity in healthcare technology,” he says.

Ross, at Unlearn, says privacy concerns and the novelty of the technology are currently limiting factors for digital twins. Without consumer trust, she says, there will not be sufficient data to develop the kind of precise models that could lead to truly personalized care.

“For this reason, we are hopeful that adoption of digital twins in clinical trials will demonstrate value to patients and improve trust in the technology, enabling greater data availability,” she says.

A multidisciplinary task

Deng agrees that ethicists should be part of the digital twin discussion, noting that the concept is by its very nature interdisciplinary.

“It overlaps a lot with AI, but it also relies on other experts from other fields and backgrounds: mathematics, physics, biology, medicine and engineering, in addition to AI/computer science,” he says.

In addition to multidisciplinary subject matter experts, the project will also require significant resources, something Laubenbacher says has been challenging. “This type of project is extensive — essentially bench to bedside — and the current funding models at NIH [National Institutes of Health] and other agencies are not well suited for this kind of project,” he says.

Still, Deng is optimistic that funding will follow the enthusiasm arounddigital twins. He sees the emergence of technology around AI, big data and advanced computer chips as coming at an opportune time for healthcare to take advantage of digital twins.

“So definitely, this is the right time and the right moment,” he says.