The idea of using social media as a tool for monitoring drug safety is not new. What is new, however, are the innovative methods now under development to gain meaningful insights from such a large volume of data.
Social media data contribute greatly to this volume. For instance, Facebook receives more than 900 million unique visits each month,1 and according to the eBizMBA website, which constantly ranks traffic on communication websites, the top 15 sites receive a combined total of 2.24 billion unique visits monthly.2 Twitter boasts more that 245 million users and LinkedIn almost 300 million.
These sites are popular for sharing information about living with a chronic disease requiring treatment. Patients can search for healthcare and disease-based information, patient experiences with physicians and hospitals, prescribed medications and reactions to those medications, both good and bad.
Other tools such as Google Insights for Search provide data on web traffic related to a disease and a drug or symptoms and adverse events in conjunction with specific drugs. The prospect of tapping into these rich sources of open information for drug safety monitoring is attractive.
But, is it possible to extract drug safety information from social media’s sea of communicated data? More importantly, is it feasible to extract scientifically meaningful information that can be used to better understand the effect of a specific medication on the health of particular individuals or groups? These two distinct questions were explored at the DIA 2014 Pharmacovigilance Conference, held in Washington, D.C., and the answer to both is “yes and no.”
At the conference, Nabarun Dasgupta, MPH, PhD, a quantitative epidemiologist and president of the health data analytics company Epidemico, spoke about an information gathering tool still in development called MedWatcher Social. Supported by FDA funding, the tool searches Facebook, Twitter, and selected online patient community sites for potential safety signals, which are identified by the reference to specific drug names—by brand, generic, variations and International Nonproprietary Names (INNs). A natural language algorithm is used to sort all instances of a drug name and surrounding text into one of two categories: “Potential adverse event” (e.g., “Ever since I started taking Drug X, my hands are so swollen.”) and “not an adverse event” (e.g., “Waited for 20 minutes at the pharmacy just to pick up Drug X!”).
The posts identifying potentially drug-related adverse events (AEs) could be further evaluated to determine if they may be true AEs. In a study of one drug, a comparison of the AEs identified in clinical trials with the potential AEs detected for that same drug acquired by mining Facebook and Twitter data with the MedWatcher Social tool, showed close alignment in the AEs identified and the rank order of frequency of those AEs.
Challenges of Using Social Data
But how accurate are the data produced by such tools? Andrew Rut, CEO of MyMeds&Me, which creates technology solutions for direct patient input of AEs and product complaints, compared Google Flu Trends (GFT) data, based on Internet search algorithms, with CDC data on influenza-like illness in four different time periods. While the GFT data demonstrated changes in patterns of flu-like illness, their accuracy was not high. And variability from the CDC calculated rates fluctuated with the season, region of the country, and age distribution of the affected population.
According to Rut, these results highlight the challenges of working with social media data. Internet search algorithms have been used successfully to gather consumer data for commercial purposes, and methodologies for tracking adverse events must continue to evolve. But if the tools to harness social media data for drug surveillance are available, is it feasible to derive meaningful information by applying them to the data as they currently exist?
Experts at the DIA meeting noted the limitations and challenges of working with social data. Of the large numbers of social media users, only a small percentage are “doing the talking,” says Rut, who estimates that in online communities, 1% of users supply most of the regular commentary, 9% contribute occasionally, and the remaining 90% simply read or observe.
Yet, many elderly individuals don’t use social media—and this is important because it creates a strong user bias for pharmacovigilance applications—this demographic is a large user of prescription medications. Unfortunately, the data are prone to interpretation bias due to the fragmented, exaggerated or false nature of many communications. Also, it’s often impossible to validate suspected AEs because the patient can’t be identified—and the sheer volume of data that must be culled for potential problems can be overwhelming.
Advantages of Using Social Data
Social data offers some advantages over traditional adverse event reporting data or data mined from health and reimbursement records. Social reports are rapid, usually occurring in close proximity to the event, making them the closest thing to real-time data. They come directly from the patient and are potentially a richer source of information than reports filtered through health professionals.
Dr. Henry Francis, Director for Data Mining and Informatics Evaluation and Research at the Food and Drug Administration (FDA) Center for Drug Evaluation and Research (CDER), agrees that while social data have limitations, all data types have strengths and weaknesses, and a variety of data sources must be used together for drug safety surveillance. Traditional methods will remain a part of the pharmacovigilance “tool box,” but social media will be an added tool to cull data in real time, making it an early indicator of potential issues for further examination.
The FDA is also actively exploring a number of social media-based strategies that include using Google search tools, mining blogs and microblogs (such as Twitter), creating virtual networks with avatars that could send and receive data for emergency systems, and optimizing data gathering from content communities and collaborative sites. According to Dr. Francis, all could be possible additions to the safety monitoring toolbox of the near future.
Before industry can be comfortable with the use of social data for drug safety surveillance, however, new regulatory paradigms will be needed and more questions will need to be raised, such as:
- What are the boundaries of industry’s responsibility for collecting and reviewing social data once the tools are available?
- What will be acceptable practices for follow up on potential signals within the context of data privacy?
- What are the protocols for data interpretation and reporting of follow-up results?
There is much work to do before the power of social media can be applied to drug safety questions—but the concept clearly has everyone’s serious attention.
References:
1. Maeve Duggan and Aaron Smith, Pew Research Center, January 2014, “Social Media Update 2013.” Available at: http://pewinternet.org/Reports/2013/Social-Media-Update.aspx.
2. eBizMBA.com: http://www.ebizmba.com/articles/social-networking-websites.
