Donate to Science & Enterprise

S&E on Mastodon

S&E on LinkedIn

S&E on Flipboard

Please share Science & Enterprise

AI Process Devised for Drug Repurposing

Network pointer

(Gerd Altmann, Pixabay)

5 Jan. 2021. A biomedical informatics lab developed techniques with artificial intelligence to find new uses for existing drugs from millions of electronic health records. A team from Ohio State University in Columbus describes its process in yesterday’s issue of the journal Nature Machine Intelligence (paid subscription required).

Researchers led by Ohio State computer science and biomedical informatics professor Ping Zhang are seeking alternative, yet reliable methods for identifying new uses of currently approved drugs that can bypass randomized clinical trials. Using current drugs for new types of diseases can take advantage of the drugs’ established safety, thus speeding-up and lowering the cost of these new applications. And while clinical trials are the so-called gold standard for testing safety and efficacy of current drugs with other types of diseases, those studies are time-consuming and expensive, delaying opportunities and reducing any savings from repurposing the drugs.

In their study, Zhang and colleagues sought to replicate to the extent possible the scientific rigor of clinical trials, but with real-world experiences captured in electronic medical records. The real-world evidence for this study comes from 90 million electronic health records from 2012 to 2017 in databases provided by MarketScan, a division of IBM. MarketScan gathers electronic health records from insurance claims, including Medicare and Medicaid, employers, health plans, and health care providers.

From these records, the researchers identified 1.2 million individuals with coronary artery disease, a narrowing or blockage of arteries carrying blood to the heart, and a leading cause of heart failure and stroke. The team used coronary artery disease as a test case, but the researchers say their process can be applied to many other types of diseases.

Controlling real-world confounders

With these data, the researchers built a deep-learning algorithm, a form of A.I., to emulate clinical trial testing of different drugs to treat coronary artery disease. The algorithm matches patients in the database by age and gender, and tracks the drugs taken by patients as well as their clinical outcomes over two years. A key objective of the algorithm is to statistically control for the host of real-world variables such as co-morbidities and other prescribed drugs, called confounders, that in clinical trials can be controlled through the enrollment process.

“This is the reason we have to introduce the deep learning algorithm, which can handle multiple parameters,” says Zhang in a university statement. “If we have hundreds or thousands of confounders, no human being can work with that. So we have to use artificial intelligence to solve the problem.”

Their analysis highlighted 55 drugs as potential repurposing candidates, which the algorithm calculated patient outcomes over two years, against matched groups of patients not taking the drugs. From the algorithm, the researchers identified nine current drugs as effective treatments preventing heart failure and stroke. Of those nine drugs, six medications are not yet approved for coronary artery disease: metoprolol, fenofibrate, hydrochlorothiazide, pravastatin, simvastatin, and valsartan. The analysis also shows a combination of the diabetes drug metformin and anti-depressant escitalopram can lower the risk of heart failure and stroke.

Zhang emphasizes that the important outcome of the study is the model for identifying alternative uses of current drugs, not just the drugs and combinations to treat heart disease. “The general model could be applied to any disease if you can define the disease outcome,” notes Zhang.

More from Science & Enterprise:

Disclosure: The author owns shares in IBM.

*     *     *

Comments are closed.