It seems like there’s an app for everything these days. You don’t have to leave your house for food, work, or social interaction (though we hope you still do on that last one).Now, we’re getting closer to testing drugs digitally, too.
Researchers published a study in npj Digital Medicine in May that used data from 59,000 Mayo Clinic patients’ electronic health records (EHRs), combined with computer modeling, to predict whether 17 existing drugs could help treat symptoms of heart failure.
The researchers checked their predictions against existing clinical trial results. They found their digital clinical trials predicted whether or not the drugs could improve several heart failure prognostic markers with about 89% accuracy, according to Nansu Zong, a biomedical informatician at Mayo Clinic and lead author of the study.
Because it uses existing data on real-world outcomes of drugs, an approach like this can’t predict outcomes of drug candidates that haven’t gone to market. But it could one day be a screening tool that helps researchers decide whether to repurpose an existing drug to potentially treat a new disease.
“I don’t think our intention will be to replace the actual clinical trial but to provide some signals so that it will be more effective and more efficient,” Cui Tao, the Nancy Peretsman and Robert Scully chair of the AI and informatics department and VP of Mayo Clinic Platform Informatics, told Healthcare Brew.
The researchers plan to test this model on other health conditions, too, and the “ultimate goal” is to predict how well any drugs will work on a disease, not just whether or not they’ll work, Nansu Zong, a biomedical informatician at Mayo Clinic and lead author of the study, told us. This can help them decide whether a clinical trial will be worth it.
“If you are low risk, low probability to achieve your expectation, maybe you’re going to change your drugs or you’re going to change your trials,” Zong said.
Why this matters. This is one of a growing number of studies using observational data to simulate randomized trials, a process called emulation. Randomized clinical trials are considered the most rigorous way to determine a cause-and-effect relationship, but they’re expensive to conduct.
Navigate the healthcare industry
Healthcare Brew covers pharmaceutical developments, health startups, the latest tech, and how it impacts hospitals and providers to keep administrators and providers informed.
Doing a virtual trial run like this can help save money. The R&D for bringing a new drug to market can cost $1.5 billion to $2.5 billion over an average of 10+ years. R&D for repurposing an existing drug is around $300 million and takes an average of three years, according to a 2024 report by Duke University research institute the Duke-Margolis Institute for Health Policy.
Emulation also allows researchers to track real-world drug outcomes after they’re approved and get data from patients typically underrepresented in clinical trials, like older people, minorities, or people with comorbidities, Tao said.
The details. According to the study, this research went beyond other clinical trial emulations by incorporating drug-target prediction, an analysis that predicts the probabilities a drug will affect specific genes.
In this case, the researchers predicted the probability each drug would work on genes associated with heart failure.
The researchers found this was a more accurate method than EHR-based emulation alone, which had about 70% predictive accuracy in this study according to Zong. The paper notes, however, “drawing definitive conclusions about the method’s effectiveness at this stage would be premature.”
The implications. The research “offers a great direction for helping to design more targeted trials,” Jimeng Sun, professor at University of Illinois Urbana-Champaign’s school of computing and data science and co-founder of Keiji AI—a generative AI platform for clinical research—told us.
“I think similar approaches can be expanded to other conditions like oncology if sufficient EHR data can be obtained,” Sun, who was not involved in this study, said.
One limitation, he added, is that outcome measures have to be data EHRs regularly include.
“Diseases with slow progression and without very clear outcome endpoints, like Alzheimer’s, will be difficult to adopt with these approaches,” he said.
The study’s researchers said they want to explore whether advanced simulation methods like AI can allow them to evaluate new, untested drugs as well.
“We are currently investigating the use of advanced simulation technologies to generate synthetic data, which may extend the framework to evaluating new drug candidates,” Zong said.