Tweet Your Reaction
The Future of Drug Monitoring
By Mark Dwortzan
Social media is often associated with trivial pursuits, but recent research at the Boston University College of Engineering has harnessed Twitter’s massive reach to improve a critical aspect of healthcare: how doctors and patients report adverse side effects of prescription drugs, vaccines and medical devices to regulators.
Clark Freifeld, a newly minted BME PhD and founder of the startup Epidemico, has leveraged the company’s expertise in public health and computer science to explore how the explosive growth of social media posts on adverse side effects could be exploited to replace or enhance the current paper-based reporting process and boost the volume of data that gets sent to the Food and Drug Administration. Moving more of the reporting process online could enable regulators and drug manufacturers to spot patterns of adverse side effects sooner and address them more quickly.
Under the traditional system, doctors must fill out a four-page form reporting a negative drug reaction, then send it in to the FDA. Not surprisingly, researchers estimate that only 10 percent of such reactions ever get reported. Freifeld’s project, funded by the FDA and described in a recent study published in the journal Drug Safety, could lead to significant improvements in prescription drug monitoring and safety.
In continuously monitoring all tweets (about 500 million per day) over a seven-month period,Freifeld and collaborators from Boston University, Boston Children’s Hospital, Harvard Medical School, the FDA and other research institutions found that for the most part, tweets of side effects matched up with reports collected the old-fashioned way. For each of 23 drugs examined, the most frequently mentioned adverse effects of that drug—from fatigue to stomach aches—were similar to those found in data manually reported to the FDA MedWatch drug monitoring system.
To identify tweets that reported adverse side effects of the 23 selected drugs, the researchers created two online dictionaries, one for drug product names, and the other for common symptoms. Using the drug name dictionary to perform searches on all Twitter posts over the study period, they obtained 6.9 million tweets that contained at least one mention of a drug name, collecting those tweets in a database. They manually inspected 60,000, finding 4,400 where the content really was about adverse side effects, and classified these by symptom using the symptom dictionary. Since submitting the Drug Safety paper, they’ve automated the process using a machine learning program.
“We found about a 70 percent correlation between symptoms reported on MedWatch and those posted on Twitter,” said Freifeld. “We also saw many more mild side effects in the Twitter data, as most patients and doctors wouldn’t fill out an extensive form just to report a headache, itch, or other mild reaction.”
In addition to monitoring social media for patterns in adverse side effect reporting, Epidemico is also advancing a user-friendly, iPhone and Android-compatible app called MedWatcher, which enables patients and physicians to upload a more streamlined version of the information that they’d normally mail or fax to the FDA’s MedWatch system. Users choose from a list of more than 10,000 drugs, vaccines and medical devices, indicate problematic symptoms, and then submit.
In development since fall 2010, MedWatcher promises to speed up an antiquated process and produce significantly more data that the FDA and pharmaceutical companies can use to monitor and address adverse side effects.
“The opportunity is to tap into the massive underreporting of side effects and capture that information,” said Freifeld. Epidemico’s data collection and analysis capabilities, which include the use of natural language processing and machine learning algorithms, have attracted investment not only from the FDA but also from pharmaceutical companies seeking more efficient and effective ways to gather and analyze online patient data.
Both the social media monitoring and automation of adverse side effect reporting were the focus of Freifeld’s PhD dissertation, which drew on skills in mathematics and computer science that he developed as an undergraduate at Yale and as co-founder of HealthMap, which tracks disease outbreaks around the globe. In recognition of his groundbreaking dissertation, he received a Societal Impact Award from the College of Engineering at the 2014 Commencement.
“Clark’s work enables us to continuously trace adverse effects caused by medications or medical products,” said Professor Simon Kasif (BME), Freifeld’s thesis advisor and a co-advisor on Twitter study. “This new way of engineering public health platforms has the potential to transform FDA procedures and guidelines and be used by pharmaceutical companies to track the effects of medications.”