As longer lifespans mean that more people may have to deal with Alzheimer’s and other dementias, researchers are working overtime to find new treatments. To leave no stone unturned, researchers at the University of South Carolina’s College of Pharmacy are scouring reams of data to see whether existing medications might be put to new use.
Using a combination of expertise in data science, artificial intelligence and pharmacology, the researchers are looking over decades of medical records to identify potential connections between approved treatments for a wide range of chronic illnesses and neurocognitive disease.
“What’s happening at the College of Pharmacy is unique,” says Joseph Magagnoli, clinical assistant professor in the Department of Clinical Pharmacy and Outcomes Sciences. “We bring together clinical pharmacy experts in drug use, data scientists like myself and basic scientists working with disease models, cell cultures and mouse models. By linking these departments, we can conduct the entire spectrum of analysis in-house.”
The team includes Functional Genomics Core Director Michael Shtutman, Clinical Pharmacy and Outcomes Sciences department chair Scott Sutton and biostatistician Tammy Cummings.
One challenge for the researchers is that comparing medical records is like comparing apples to oranges. Specific diagnoses, such as high blood pressure or diabetes, are often coded clearly for insurance reimbursement purposes. But other information must be parsed from clinician observations made in notes or “free text,” which can vary widely among individual physicians and medical systems.
“When pharmaceutical companies bring drugs to the market, they're typically for one indication. And that's based on the laboratory science and the drug development targeting a specific pathway. There could be side effects with those drugs, because they're hitting other pathways, that have unintended consequences. So, the question is how do we find those drugs that have clinically relevant off-target effects that could be beneficial?”
Finding connections in that mass of structured and unstructured information for millions of records is beyond human capabilities. That’s where artificial intelligence comes in.
“Clinical data don’t come neatly organized,” says Shtutman. “Physicians document care in many different ways and there’s no single standard. That means working with large volumes of unstructured notes and test results rather than a clean database. Making sense of that data is difficult, and our team has developed algorithms to help extract insights from it.”
Magagnoli explains that while a new drug is being tested, researchers obviously look for negative side effects that may outweigh the clinical value of the drug. But positive side effects may not be noted, especially since drug trials are short-term and people typically take medications for chronic conditions for long periods of time.
“When pharmaceutical companies bring drugs to the market, they're typically for one indication. And that's based on the laboratory science and the drug development targeting a specific pathway,” Magagnoli says. “There could be side effects with those drugs, because they're hitting other pathways, that have unintended consequences. So, the question is how do we find those drugs that have clinically relevant off-target effects that could be beneficial?”
Once the research reveals connections between certain drugs and these “off-target” outcomes, the team produces a paper documenting what it has found. Two such papers have already been accepted for publication. These will serve as a jumping off point for further lab study, which can, in turn, lead to clinical trials for existing medicines focused on a new use.
Magagnoli says the team can also study specific types of drugs and their impact on long-term issues like dementia. For example, if a drug that is used to treat high blood pressure also inhibits inflammation in the brain, he would expect to find that people who took that drug would have lower rates of dementia than people who took a different blood pressure medication. These findings can provide guidance for doctors as they choose the best treatment options for patients.
“If the scientists in the lab have a drug and it inhibits this enzyme, if it's clinically relevant, we should be able to see that effect in large numbers of patient data,” he says. “We can test that in the database.”
