Drug interactions are nothing new, but new ways of finding them are changing how fast the public learns about them. And you might be surprised to know that information from you and millions of other Americans, rather than clinical trials, is helping make these discoveries possible.
When you experience a side effect from a medication, you’re most likely to call your doctor for help. But what you do next can influence the lives of many others taking the same medication.
When you fill out an FDA report detailing not only the problematic medication and its adverse effects but also any other symptoms you noticed, any other medications you’re taking and any other conditions you have, this information goes into an FDA database. The database is maintained and analyzed continually to identify any patterns of side effects in the primary medication you listed. But until recently the rest of the information you put on the form has just been sitting there, unused.
Researchers have now found ways to tap into that information and identify medication interactions that were going unnoticed by using artificial intelligence (AI). At Columbia University, Nicholas Tatonetti, PhD, has trained his computer to look for patterns of side effects involving pairs of drugs matched to various medical conditions. His system, called Latent Signal Detection, assigns scores to drug pairs based on the likelihood for an adverse reaction from taking both drugs. A high score is the flashing red light for a potentially serious interaction.
At the top of the side effect list generated by the computer is a dangerous heart arrhythmia affecting people on both the antibiotic ceftriaxone and the heartburn medicine lansoprazole (Prevacid). To find the underlying cause of the irregular heartbeat, he collaborated with pharmacology researcher Robert Kass, PhD. They demonstrated that in combination, the two drugs led to a blockage in a channel that controls the rhythm of the heart (neither drug had shown this on its own when tested separately).
Another adverse effect uncovered by the system is the increase in blood sugar in people taking the common antidepressant paroxetine (Paxil) and the cholesterol-lowering drug pravastatin (Pravachol).
On the other side of the country at Stanford University, researchers have designed an AI system called Decagon. It has created a large graph of interactions between medications and the proteins in the body known to be affected by them. Decagon predicts adverse effects by analyzing both protein-protein interactions and drug-protein interactions.
Among its most notable findings is the painful muscle inflammation that can result from taking the cholesterol-lowering drug atorvastatin (Lipitor) and the hypertension drug amlodipine (Norvasc)…the kidney problem known as renal tubular acidosis from taking the acid reflux drug omeprazole (Prilosec) and the antibiotic amoxicillin…and breast inflammation from taking the hypertension drug aliskiren (Tekturna) and the vaginal antifungal tioconazole.
In all, these AI systems have identified dozens of drug combinations with potential adverse side effects. But there’s an obstacle to being able to use all that information—researchers need to identify how and why an interaction occurs to prove a direct relationship, and that’s not something AI can do on its own right now. (Some findings from both systems have been borne out in people-based research published around the world.)
While neither Decagon nor Latent Signal Detection is ready to be placed at your fingertips today, both teams of researchers are working hard to connect the dots and make that a reality in the not-too-distant future.