Taking a “better safe than sorry” approach, WVU professors are involved in a new research project to extract vital information related to drug reactions from social media chatter.
By creating computer algorithms, which can detect key patterns or relationships within online chatter on social media, the researchers can catalog the complaints about a drug and monitor them until they reach a high enough level to require regulatory attention.
The idea is for drug manufacturers, federal regulatory agencies, health care professionals and the public to be aware of such potential problems for a given drug, and perhaps take a deeper look at the drug,” said Donald Adjeroh (left), professor of computer science in the Benjamin M. Statler College of Engineering and Mineral Resources, who is working with a team of researchers studying the problem.
Normally, patients report negative reactions to their physicians, who then report them to the Federal Drug Administration and pharmaceutical companies. But are all the reactions being reported and is the FDA receiving the information quickly enough?
In a preliminary study involving 20 drugs already on the market, the researchers were able to detect potential adverse drug reactions 80 percent of the time. Adjeroh said they detected these drug reactions significantly earlier than the FDA-issued warnings. In more than 50 percent of the cases where adverse reactions were detected, the detection was more than three years before the FDA warning.
The National Science Foundation recently awarded the team of researchers from West Virginia University and the University of Virginia a $130,000 grant to launch a larger project, which will identify key patterns and relevant information through a coding system, capable of deciphering drug-related information from randomly posted comments.
“There are unusual side effects with reported incidence of less than one in 1,000, but they may lead to life-threatening conditions,” said Wanhong Zheng, psychiatrist and clinical assistant professor at the WVU School of Medicine, who is also involved in the project.
“There is no doubt that some side effects may be over- or misreported, mainly due to different confounding factors,” said Zheng. “A patient may report diarrhea as a side effect of a new medication but actually they started having this symptom after eating a bad salad. I believe the pieces can always be put together if enough information is provided and good computational methods are used.”
But how do you decipher which reported side effects are caused by the medication and which are caused by external, unrelated events? How about intentionally misleading postings? According to Adjeroh, “We have used control cases to check how often the method reported signals when none was expected. Ultimately, the system aggregates various pieces of information from online sources to generate hypotheses about potential adverse drug reactions. These are expected to be further studied by interested parties, such as regulatory agencies, or the drug manufacturers.”
Adjeroh noted that at this time, their work does not focus on the actual cause of the adverse event and the researchers do not necessarily look for intentionally misleading comments online. “However, Ahmed Abbasi, our collaborator and project co-leader at University of Virginia, has developed a method for fraudulent website detection, which will be leveraged in this project,” he added.
Marie Abate, professor in the WVU School of Pharmacy and director of the statewide West Virginia Center for Drug and Health Information, will monitor current medical literature for newly published reports of adverse drug reactions during the development phase of this project to determine the extent to which similar reports may appear on the internet.
“The nature of human-computer interaction has significantly changed over the past few years,” added Arun Ross, associate professor and assistant director for the Center for Identification Technology Research at WVU, who is also a member of the research team. “This work is a combination of crowdsourcing, human-based computation and automated parsing of social media. The near ubiquity of social media applications offers a compelling reason for advancing the research agenda suggested in our project.”
Richard Beal and William Mensah, two of Adjeroh’s graduate students, are also involved in the project. Adjeroh and his partners hope this research will lead to the population of a large database from which the medical community and pharmaceutical companies can tap into. The database will provide more information in a timelier manner so that in the case that investigations of the drugs are necessary, all parties involved have a much larger dataset to work with.
According to Adjeroh, the Federal Drug Administration and some large pharmaceutical companies are already indicating interest in the research.
CONTACT: Mary C. Dillon