Finding better ways to reduce serious drug side effects
big data methods may yield improvements
By Jeffrey Norris
Many of the medicines we depend on to treat disease—and even to save our lives—pose potentially serious risks along with their benefits. Data from the U.S. Centers for Disease Control and Prevention indicate that about 40,000 deaths yearly in the United States may be attributable to the side effects of drugs, a number that rivals the toll of traffic accidents.
Brian Shoichet, PhD
Image credit: Elisabeth Fall
Two UCSF School of Pharmacy faculty members, whose research and clinical practice have focused on the identification and minimization of side effects, have pinpointed routes for improvement.
Predicting potential side effects is a complicated task. Even after extensive testing, some adverse effects often don’t show up until drugs are taken by a broad range of patients, said Marilyn Stebbins, PharmD, a faculty member in the School’s Department of Clinical Pharmacy. “Clinical trials are not done in the general population; they are done in a small population under controlled conditions,” she noted.
For decades, Stebbins helped physicians manage treatment as part of a team at an outpatient medical practice that served many low-income patients. Like many other health care providers, she relies on the Food and Drug Administration Adverse Event Reporting System (FAERS) both for reporting new unintended reactions and for assessing known side effects. Physicians, pharmacists, nurses, and other providers contribute to the public database. Patients, family members, and even attorneys can also submit reports.
“FAERS is potentially a gold mine, but the nuggets often are buried under informational rubble…” —Brian Shoichet, PhD
But while FAERS is growing by a more than one million reports a year, the data is neither curated nor standardized, and doesn’t link drugs to their chemical structures. Treating the database as a big data project provides a way of sorting through all the noise, according to a new study published August 8 in the online journal eLife, co-led by Brian Shoichet, PhD, a faculty member in the School’s Department of Pharmaceutical Chemistry, along with researchers from Novartis.
“FAERS is potentially a gold mine, but the nuggets often are buried under informational rubble that needs to be cleared away to realize the potential of the database,” Shoichet said.
The cumulative number of reports in FAERS is shown in the top panel; the bottom panel shows the number of new reports per quarter.
Image credit: eLife 2017
In a computational analysis of more than 8.7 million reports, carried out primarily by postdoctoral fellow Mateusz Maciejewski, PhD, guided by Shoichet and by Laszlo Urban, MD, PhD, Global Head of Preclinical Safety Profiling at Novartis, the researchers found that nearly one percent of reports in the database are duplicate entries of the same adverse event in the same patient, a problem that can overstate the significance of a side effect.
They also found that a disease—or even a death due to a lethal disease—was erroneously listed as a side effect of drug treatment in five percent of reports. For instance, FAERS shows the drug thalidomide associated with the adverse side effect of myeloma multiplex, when it actually is used to treat that cancer. Not surprisingly, death itself is an over-reported outcome relative to its actual prevalence among drug side effects; less serious side effects are reported less often.
Making data more useful
A “tangle of drug synonyms” is also holding back the database, according to the researchers. Most drugs in FAERS are entered under their brand name or their non-proprietary name, meaning there can be hundreds of names referring to the same active ingredient. Instead of using drug names, researchers organized the database by the chemical structure—which is the key to understanding a drug’s therapeutic effect—and corrected for duplicate entries.
Only about half of the reports in the database were made by health care professionals, the study found. Reports made by attorneys, which make up about 3 percent of the database, were more highly correlated with a bias related to news coverage. Reporting of side effects was affected by news and announcements regarding specific drugs, and other similar-acting drugs often were affected by the same announcements.
For example, reporting of strokes and heart attacks associated with the arthritis treatment, celecoxib, a COX-2 inhibitor, spiked around the time another COX-2 inhibitor, rofecoxib (Vioxx) came under investigation and was eventually withdrawn from the market. However, reports of these celecoxib side effects later fell back to insignificant background levels. Conversely, without pooling drugs for analysis by active chemical ingredient, even some known links to side effects did not appear to be significant, such as the sexual side effects associated with use of anti-depressant SSRIs.
Shoichet has had a longstanding interest in using chemical information and computational methods to predict drug side effects and to avoid side effects during early stages of new drug design, and even to leverage these off-target effects to repurpose drugs for new disease indications. To Shoichet, the FAERS database seemed like a great place to mine data to improve understanding of adverse events, and to find off-target interactions that revealed the potential for drugs to be repurposed to treat other diseases. However, Shoichet said, “We were surprised at how little chemistry there was in FAERS, and how hard we had to work to even organize the many brand names in the database into coherent active ingredients that could be treated as chemical structures.”
Shoichet is hopeful that the method he used in the study could make the massive database more useful. “Researchers who are analyzing FAERS to identify signals that link drugs to important side effects need to really be aware of the confounding factors, and we have laid out a way to do that.”
“You cannot put people on multi-drug regimens and just let them go.” —Marilyn Stebbins, PharmD
While FAERS may prove invaluable for identifying potential drug side effects, it is difficult to identify side effects due to drug interactions using the database. With more individuals in our aging population on multi-drug regimens for more than one chronic condition, identifying such interactions is increasingly important. In addition, not all patients respond similarly to the same drug taken as prescribed, and it remains difficult to predict an individual’s response.
Marilyn Stebbins, PharmD
Image credit: Elisabeth Fall
The bottom line, according to Stebbins, is that medication use should be monitored for individual patients. In many cases, adverse drug events are due to failures to take medication as prescribed—known as nonadherence. Patients may misunderstand directions, or fail to fill prescriptions they think they cannot afford. At UCSF, Stebbins works in a program for newly discharged hospital patients to respond and track patient concerns about their medications. “Patients often get into trouble with side effects due to miscommunications about medications, especially pain medications and diabetes drugs,” Stebbins said.
For outpatients coping with chronic conditions, she envisions a larger role for community pharmacists to help manage medication use. “You cannot put people on multi-drug regimens and just let them go,” she said. “If we were able to engage patients and evaluate their medications on a more regular basis, we could probably avoid a whole lot of adverse drug reactions, drug interactions, emergency department visits, and hospitalizations … and we could improve patients’ quality of life.”
source: University of California, San Francisco