Scrolling through Twitter may feel like a casual downtime activity, but a recent Wall Street Journal article describes social media’s growing utility as a medical research tool. Increasingly, researchers are using AI and other data-culling techniques, to sift through social media comments and gather real-time, unfiltered feedback on patient experiences—from the patients themselves.
In an interview with the WSJ, Dr. Graciela Gonzalez-Hernandez, an Associate Professor at the University of Pennsylvania’s Perelman School of Medicine said, “Collecting abundant social-media data is cost-effective, does not involve burdening participants, and is available in real time.” She also points out that it may be more inclusive of populations often underrepresented in biomedical trials or cohort studies. And, most importantly, social media provides researchers direct access to unfiltered patient commentary on their experiences. “Healthcare providers report what they deem important, such that serious events are overrepresented, while bothersome side effects that may be of great importance to patients and lead to nonadherence and non-persistence are underrepresented,” Gonzalez-Hernandez said.
The article describes how researchers studying an opioid withdrawal drug have benefited from social media scans. Dr. Abeed Sarker, an Assistant Professor of Biomedical Informatics at Emory University, and his team used natural language processing algorithms to search Reddit for comments on opioid withdrawal experiences. Dr. Sarker’s team found that Reddit contributors frequently described concerns about withdrawal drugs causing “precipitated withdrawal,” or extreme withdrawal symptoms, among those who have used fentanyl. Importantly, many of these users’ comments expressed concern that their medical providers did not understand this condition. The study also found that the quantity of posts about fentanyl and withdrawal have increased over a seven-year period. These findings have played a key role in bringing to light the urgency of this issue for physicians researching withdrawal assistance drugs.
In France, a similar study took place to better understand breast cancer patients’ quality of life. The researchers used an algorithm to scan comments left on Facebook and a French online forum for breast-cancer patients. Researchers compared these insights with the European Organization for Research and Treatment of Cancer’s quality of life questionnaire to ensure it was inclusive of the most pertinent topics. The online comments brought to light two additional topics that were top of mind for patients—nonconventional treatments and patients’ relationships with their families—which will be added to future iterations of the questionnaire for formal study.
Many researchers seem intrigued by the possibility of using social media data in combination with traditional research. Su Golder, an Associate Professor at the University of York in England, is working on a medical literature review on HPV vaccines, which will incorporate Twitter and WebMD comments. “It’s important to research what the public is worried about,” Dr. Golder said. She notes that the benefits of social media data—patient perspectives and real-time information—can be meaningful, and that its shortcomings—comments that may confuse association with causation—can be overcome by combining the findings with traditional research efforts.
Social media data might also work in combination with electronic medical records to improve patient care. A recent study by Dr. Munmun De Choudhury, an Associate Professor at Georgia Tech, created an algorithm to predict psychosis relapses and re-hospitalizations using the social media posts of 50 adult and adolescent psychosis patients who suffered relapses and hospitalization, along with their medical records. The study found that in the month preceding the relapse, there were distinct changes to social media usage in the form of language patterns—words related to “anger, death, or emotional withdrawal”—and the number of posts between midnight and 5am. The algorithm was able to accurately predict about 71 percent of relapses. Dr. De Choudhury is currently working with the state medical system to determine how to incorporate the algorithm into physicians’ clinical care processes.