Building an AI Army of Digital Twins to Fight Cancer

A patient has cancer. It’s decision time.

Clinician and patient alike face, really, the ultimate challenge when making those decisions. They have to consider the patient’s individual circumstances, available treatment options, potential side effects, relevant clinical data such as the patient’s genetic profile and cancer specifics, and more.

“That’s a lot of information to hold,” said Uzma Asghar, PhD, MRCP, a British consultant medical oncologist at The Royal Marsden Hospital and a chief scientific officer at Concr LTD.

What if there was a way to test — quickly and accurately — all the potential paths forward?

That’s the goal of digital twins. An artificial intelligence (AI)–based program uses all the known data on a patient and their type of illness and creates a “twin” that can be used over and over to simulate disease progression, test treatments, and predict individual responses to therapies.

“What the [digital twin] model can do for the clinician is to hold all that information and process it really quickly, within a couple of minutes,” Asghar noted.

A digital twin is more than just a computer model or simulation because it copies a real-world person and relies on real-world data. Some digital twin programs also integrate new information as it becomes available. This technology holds promise for personalized medicine, drug discovery, developing screening strategies, and better understanding diseases.

How to Deliver a Twin

To create a digital twin, experts develop a computer model with data to hone its expertise in an area of medicine, such as cancer types and treatments. Then “you train the model on information it’s seen, and then introduce a patient and patient’s information,” said Asghar.

Asghar is currently working with colleagues to develop digital twins that could eventually help solve the aforementioned cancer scenario — a doctor and patient decide the best course of cancer treatment. But their applications are manifold, particularly in clinical research.

Digital twins often include a machine learning component, which would fall under the umbrella term of AI, said Asghar, but it’s not like ChatGPT or other generative AI modules many people are now familiar with.

“The difference here is the model is not there to replace the clinician or to replace clinical trials,” Asghar noted. Instead, digital twins help make decisions faster in a way that can be more affordable.

Digital Twins to Predict Cancer Outcomes

Asghar is currently involved in UK clinical trials enrolling patients with cancer to test the accuracy of digital twin programs.

At this point, these studies do not yet use digital twins to guide the course of treatment, which is something they hope to do eventually. For now, they are still at the validation phase — the digital twin program makes predictions about the treatments and then the researchers later evaluate how accurate the predictions turned out to be based on real information from the enrolled patients.

Their current model gives predictions for RECIST (response evaluation criteria in solid tumor), treatment response, and survival. In addition to collecting data from ongoing clinical trials, they’ve used retrospective data, such as from the Cancer Tumor Atlas, to test the model.

“We’ve clinically validated it now in over 9000 patients,” said Asghar, who noted that they are constantly testing it on new patients. Their data include 30 chemotherapies and 23 cancer types, but they are focusing on four: Triple-negative breast cancer, cancer of unknown primary, pancreatic cancer, and colorectal cancer.

“The reason for choosing those four cancer types is that they are aggressive, their response to chemotherapy isn’t as great, and the outcome for those patient populations, there’s significant room for improvement,” Asghar explained.

Currently, Asghar said, the model is around 80%-90% correct in predicting what the actual clinical outcomes turn out to be.

The final stage of their work, before it becomes widely available to clinicians, will be to integrate it into a clinical trial in which some clinicians use the model to make decisions about treatment vs some who don’t use the model. By studying patient outcomes in both groups, they will be able to determine the value of the digital twin program they created.

What Else Can a Twin Do? A Lot

While a model that helps clinicians make decisions about cancer treatments may be among the first digital twin programs that become widely available, there are many other kinds of digital twins in the works.

For example, a digital twin could be used as a benchmark for a patient to determine how their cancer might have progressed without treatment. Say a patient’s tumor grew during treatment, it might seem like the treatment failed, but a digital twin might show that if left untreated, the tumor would have grown five times as fast, said Paul Macklin, PhD, professor in the Department of Intelligent Systems Engineering at Indiana University Bloomington.

Alternatively, if the virtual patient’s tumor is around the same size as the real patient’s tumor, “that means that treatment has lost its efficacy. It’s time to do something new,” said Macklin. And a digital twin could help with not only choosing a therapy but also choosing a dosing schedule, he noted.

The models can also be updated as new treatments come out, which could help clinicians virtually explore how they might affect a patient before having that patient switch treatments.

Digital twins could also assist in decision-making based on a patient’s priorities and real-life circumstances. “Maybe your priority is not necessarily to shrink this [tumor] at all costs…maybe your priority is some mix of that and also quality of life,” Macklin said, referring to potential side effects. Or if someone lives 3 hours from the nearest cancer center, a digital twin could help determine whether less frequent treatments could still be effective.

And while much of the activity around digital twins in biomedical research has been focused on cancer, Asghar said the technology has the potential to be applied to other diseases as well. A digital twin for cardiovascular disease could help doctors choose the best treatment. It could also integrate new information from a smartwatch or glucose monitor to make better predictions and help doctors adjust the treatment plan.

Faster, More Effective Research With Twins

Because digital twin programs can quickly analyze large datasets, they can also make real-world studies more effective and efficient.

Though digital twins would not fully replace real clinical trials, they could help run through preliminary scenarios before starting a full clinical trial, which would “save everybody some money, time and pain and risk,” said Macklin.

It’s also possible to use digital twins to design better screening strategies for early cancer detection and monitoring, said Ioannis Zervantonakis, PhD, a bioengineering professor at the University of Pittsburgh, Pittsburgh.

Zervantonakis is tapping digital twin technology for research that homes in on understanding tumors. In this case, the digital twin is a virtual representation of a real tumor, complete with its complex network of cells and the surrounding tissue.

Zervantonakis’ lab is using the technology to study cell-cell interactions in the tumor microenvironment, with a focus on human epidermal growth factor receptor 2–targeted therapy resistance in breast cancer. The digital twin they developed will simulate tumor growth, predict drug response, analyze cellular interactions, and optimize treatment strategies.

The Long Push Forward

One big hurdle to making digital twins more widely available is that regulation for the technology is still in progress.

“We’re developing the technology, and what’s also happening is the regulatory framework is being developed in parallel. So we’re almost developing things blindly on the basis that we think this is what the regulators would want,” explained Asghar.

“It’s really important that these technologies are regulated properly, just like drugs, and that’s what we’re pushing and advocating for,” said Asghar, noting that people need to know that like drugs, a digital twin has strengths and limitations.

And while a digital twin can be a cost-saving approach in the long run, it does require funding to get a program built, and finding funds can be difficult because not everyone knows about the technology. More funding means more trials.

With more data, Asghar is hopeful that within a few years, a digital twin model could be available for clinicians to use to help inform treatment decisions. This could lead to more effective treatments and, ultimately, better patient outcomes.