A new computer program has found thousands of new side effects and problematic drug interactions, researchers reported March 14 in the journal Science Translational Medicine. On average, the program found 329 side effects for every drug, when most drug labels list an average of 69 effects. The new program, developed at Stanford University, combs through millions of doctor and patient reports to the U.S. Food and Drug Administration, Canada's MedEffect and similar databases.
Even after they're sold in pharmacies, new prescription drugs are still, in a way, in test phase. Of course, drugs in development undergo experiments before they make it to the market. The side effects found in those experiments are what people see listed on drug labels. "The problem with those is that they're fairly incomplete," one of the new program's developers, Nicholas Tantonetti, said. Tantonetti is a doctoral student at Stanford who studies large medical datasets.
There's no test that can recruit enough volunteers to catch extremely rare side effects, Tantonetti explained. So researchers often don’t learn of rare side effects until drugs are sold widely and many people take them. Pre-market tests also don't often study how a new drug will interact with other drugs someone is already taking. After all, it would be impossible for any human scientist to check all the possible drug interactions out there. But what a human can't do, a computer command can.
Though it seems obvious to check the FDA's report database for new side effects, it's been difficult to do because the people in the database are too different from one another to conduct proper scientific studies. Outside of lab, real life is messy. "It's hard to compare one patient with another because each patient is so unique in their medical history, in their personal health," Tantonetti told InnovationNewsDaily. Because people in the database aren't comparable with each other, it's difficult to find true, new side effects.
Other computer algorithms have tried. But to make valid, scientific comparisons, those programs had to ignore many reports. The new method puts people on a "continuum," Tantonetti said, so that if someone in a database doesn't have a perfect match somewhere else in the dataset, she might still have a pretty close match.
Theoretically, that means the matches the new algorithm makes aren't as close as those from other programs. When Tantonetti checked his program, however, he found his results were as accurate or better than others', because his program doesn't need to ignore as much data, he said.
He and his colleagues have put their findings about 1,332 drugs in databases called OFFSIDES, for side effects, and TWOSIDES, for drug interactions, which they plan to update once or twice a year. These are research databases that highlight side effects that need FDA attention, Russ Altman, the senior scientist in the study, said. "Patients should not change medications based on these findings, but instead talk to their physicians," Altman wrote in an email to InnovationNewsDaily. "Physicians need to use their clinical common sense to decide if the side effects and interactions we are finding are clinically significant."
Altman is an advisor to the FDA's Science Board. The research team will present their findings to the agency and see if it will use their program to find new side effects. In the past, the FDA has added algorithm-found side effects to drug labels, Tantonetti said, though usually there is other corroborating evidence as well.
In this digital age, many very large datasets are available online. "We are partly inspired by many of the recent discussions about the value of 'big data' in medicine and elsewhere," Altman wrote. He and Tantonetti hope their algorithm can also help them find the hidden patterns in electronic health records. "Electronic medical records are a treasure trove of potential discoveries," Altman said.
Related on InnovationNewsDaily:
Copyright 2012 InnovationNewsDaily, a TechMediaNetwork company. All rights reserved.