Social media offers a trove of information for medical researchers

Summary
- What are patients talking about on Facebook, Reddit, Twitter and other places? It could be revealing.
Medical researcher are turning to social-media posts to improve patient care.
Using machine-learning algorithms to sift through social-media posts, researchers can get insights into patients’ experiences that are often overlooked or difficult to attain when relying mainly on data from medical reports and doctors’ charts. It also provides data more quickly than traditional epidemiological or medical studies, which can take years to complete.
“Collecting abundant social-media data is cost-effective, does not involve burdening participants, and is available in real time," says Graciela Gonzalez-Hernandez, an associate professor at the University of Pennsylvania’s Perelman School of Medicine. It also might do a better job of capturing voices of underrepresented populations who are often absent in biomedical trials and traditional cohort studies, she adds.
One such area, she says, is in the way that healthcare providers report problems with prescription medication to the Food and Drug Administration’s Adverse Event Reporting System.
“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 nonpersistence are underrepresented," Dr. Gonzalez-Hernandez says.
Opioid withdrawal
One recent study, for instance, mined social-media posts to learn more about the effects of buprenorphine, a drug for helping opioid users get through withdrawal. The study showed a concern among Reddit subscribers that the drug would cause extreme withdrawal symptoms—known as precipitated withdrawal—for people who had used fentanyl, a drug that is increasingly mixed with heroin and is helping fuel the opioid crisis.
“Many Reddit subscribers express frustration about their medical service providers not understanding precipitated withdrawal," says Abeed Sarker, an assistant professor of biomedical informatics at Emory University and co-author of the resulting paper. Natural-language-processing algorithms in the study searched 267,136 Reddit posts for expressions about precipitated withdrawal, fentanyl and micro dosing. They also found a substantial increase in posts about fentanyl and withdrawal over a seven-year period.
Ashish Thakrar, a fellow at the University of Pennsylvania’s Perelman School of Medicine who is studying the effects of buprenorphine on people withdrawing from fentanyl, says that Dr. Sarker’s study has helped underscore the urgency of the fentanyl problem.
In France, researchers designed a machine-learning algorithm to help healthcare providers better address the needs of women with breast cancer. The program identified topics raised by breast-cancer patients posted on Facebook and cancerdusein.org, a French online breast-cancer forum, and then compared those topics with a questionnaire used by the European Organization for Research and Treatment of Cancer to gauge quality-of-life issues experienced by breast-cancer patients. The researchers found there were subjects of concern to patients that the questionnaire wasn’t exploring.
The questionnaire asked patients to share thoughts about all sorts of issues, including their body image, sexual enjoyment and systemic therapy side effects. The researchers found several topics in social-media posts that weren’t included in the questionnaire and suggested two—nonconventional treatments and the relationships that patients have with their families—be included in future questionnaires.
Some researchers are looking at how data gleaned from social-media posts can be combined with more-traditional research. Su Golder, an associate professor at the University of York in England, is working on a review of medical literature about all the adverse side effects of HPV vaccines against human papillomavirus. To buttress that work, she and her co-researchers are using natural-language processing on social-media platforms, like Twitter and WebMD’s forum, to help identify what people are saying about the vaccines online.
“It’s important to research what the public is worried about," Dr. Golder says. She says that social-media posts and data found in traditional research complement each other. Social media reveal patient perspectives and real-time information, while traditional sources can help overcome some of social-media data’s shortcomings. Data extracted from social media, for example, tend to show association, not causality. And the demographic characteristics and points of view of social-media posters may not reflect an entire cohort, Dr. Golder says.
Suicide rates
Researchers from the Centers for Disease Control and Prevention and the Georgia Institute of Technology have combined data from social media and more-traditional epidemiological sources in an effort to improve the accuracy and timeliness of estimates for national suicide rates. Currently, national suicide statistics are compiled using death certificates from more than 2,000 medical examiners, and can take more than a year to generate. That makes it difficult to plan effective suicide-prevention programs.
By adding social-media posts and other data to the statistical mix, the CDC and Georgia Tech researchers found they were able to estimate national suicide rates accurately on a weekly basis. The algorithms being used comb through public posts about suicide on Reddit, Twitter and Tumblr, and search trends from Google and YouTube, emergency-room visits for suicide attempts and ideation, calls to the National Suicide Prevention Lifeline and calls to U.S. poison-control centers as captured in the National Poison Data System.
Other experiments include combining data from social media and electronic medical records to improve care. In a paper published in Nature, Munmun De Choudhury, an associate professor at Georgia Tech, and her co-authors were able to predict psychosis relapses and hospitalizations about 71% of the time.
In the study, the authors trained algorithms to sort through roughly 50,000 Facebook posts from about 50 young adult and adolescent psychosis patients who suffered relapses and rehospitalizations. Patients in the study gave the researchers consent to see both their medical records and social-media posts. The algorithms detected distinct changes in the months before a relapse, including: changes in language patterns, especially the use of more words associated with anger, death or emotional withdrawal; changes in the number of posts between midnight and 5 a.m. and the frequency of tagging or friending.
Dr. De Choudhury is working with the Feinstein Institutes for Medical Research at Northwell Health to see if these insights can be used clinically to improve care. Clinicians could see how many depressive Facebook posts a patient had posted recently and a chart showing how that number may have changed over time. The dashboard could possibly allow clinicians to have a better sense of what is happening with patients between sessions.
This story has been published from a wire agency feed without modifications to the text