Most adults in the U.S. will take at least one prescription medication at some point in their lives, if they haven’t already. In 2021, the Centers for Disease Control and Prevention estimated that almost 65% of U.S. adults had taken a prescription medication within the previous 12 months.
Many people, especially those with chronic health conditions such as diabetes or high blood pressure, take long-term prescription drugs. It’s also common for patients to wind up taking multiple medications for various, overlapping health conditions.
But the more medications someone takes at once, the greater the risks become. Health care professionals watch these patients closely for side effects, drug interactions (which can change how medications act in the body), and other complications. But tweaks to a patient’s medication regimen can potentially tip the scales and cause unintended effects.
A University of Colorado Anschutz School of Medicine pharmacist is working on creating an artificial intelligence (AI) tool that can safely manage medications for patients taking multiple drugs. She wants to use this tool to create personalized treatment plans that apply advanced pharmaceutical knowledge to meet each patient’s unique needs.
“The advent of data science, and all the fun things that go with that, has really opened up a new era for what we can do with drug therapy and analysis,” says Andrea Sikora, PharmD, an associate professor in the Department of Biomedical Informatics.
Sikora, a newly minted National Academy of Medicine fellow, has also launched a startup called AIChemist Lab to develop this tool.
Why drug management matters — and where it can get complicated
Being on the right combination of medications, at the right dosages, can be game-changing for patients with chronic physical or mental health conditions. Getting to that point can take time, strategy, and a bit of luck.
Doctors prescribe medications to patients when they feel the benefits of the medication outweigh any potential risks or side effects. While any pharmaceutical can potentially cause adverse effects, the risk for most patients is usually low.
But different people can respond differently to the same drug. A drug that works well for one patient won’t always work for someone else. So, sometimes, patients will need to try one medication before switching to another, or will need to change their dosage, to get the desired effect.
Even before layering on additional medications, there are multiple factors that affect how, and whether, a particular drug will work for someone. Even when taking a simple over-the-counter drug like Tylenol, says Sikora, many different variables — such as the dosage, the number of pills taken, the timing, the frequency, and the maximum safe daily limit — come into play.
“Medications are very multi-dimensional. If I say I took two tablets by mouth one time for a headache, you need at least six or seven pieces of information to understand how that will affect me,” she says. “It becomes more complicated if I have kidney or liver issues or other underlying conditions.”
Accounting for all these factors, and layering them into a comprehensive treatment plan, is not simple. And the more medications one patient needs to take, the more intersecting factors need to be considered. In the intensive care units (ICUs) where Sikora has worked, she says patients often take an average of 13 to 14 medications each. Some take as many as 20 to 25 different drugs.
“That’s about 200 different variables just for one patient, at one point in time,” says Sikora.
Historically, it’s been challenging for health care providers to factor all of these variables into medication management. Sikora says that, in the past, there haven’t been solid enough statistical methods to reliably model how drug therapy affects patients.
That appears to be changing now. Modern tools such as large-language models and AI agents can handle these types of computational challenges. However, current mass-market LLMs such as ChatGPT are trained on the information available on the internet, and according to Sikora, there’s a lack of complete and accurate information about pharmaceuticals online.
Based on initial evaluations, Sikora says, these chatbots didn’t do well with tasks that involve reasoning and decision-making about drugs, and they made a lot of mistakes. Seasoned drug prescribers can spot these mistakes, but newer clinicians fresh out of school might not catch them as easily.
A step toward smarter, more comprehensive medication management
To overcome these challenges, Sikora is developing her own AI tool that will draw on much more specialized data and reasoning than a standard chatbot would have access to. She aims to create a tool that can provide the advanced decision-making and computing capabilities that are needed in medication management.
Although Sikora’s work is in its early stages, she’s hopeful her tool could one day empower care providers to confidently create comprehensive medication management plans that are tailored to each patient’s needs and that account for the most relevant variables.
She’s also experimenting with ways to create a system that makes flexible plans that can adjust based on how the patient is doing day-to-day. She anticipates such a tool could have an enormous impact on how drugs are prescribed.
“We believe a patient needs to have medication that is individualized to them in order to have the best outcomes,” Sikora says. “This tool helps provide that and could have ramifications for pretty much anywhere that drugs are used.”
