I probably don't have to explain the power of social media as a learning and communication tool. After all, you're here reading this blog post (and hopefully learning something from it).
Now researchers are learning how to mine online interactions to uncover information that would otherwise take months or years to document and publish in the medical literature. Dermatologists Kavita Sarin, MD, PhD, and Bernice Kwong, MD, collaborated with informatician Nigam Shah, MBBS, PhD, and researchers from Inspire – a company with a health care social network of more than one million patients and caregivers – to investigate whether it's possible to harness the power of online chatter to identify adverse drug reactions occurring after anticancer treatment.
They published their results this week in a letter in JAMA Oncology.
As Sarin explained in an email:
Drug reactions occur in the majority of patients undergoing anti-cancer therapy. Half of all serious drug reactions are detected after market approval, resulting in over 2 million injuries and $75 billion annual health care costs. Post-market drug surveillance platforms such as FDA monitoring rely on voluntary, spontaneous reporting and take time to identify trends. We sought to determine if we could mine internet health forums to more rapidly identify chemotherapeutic drug reactions.
Sarin, Shah and their co-authors, including medical student Julia Ransohoff and postdoctoral informatics scholar Azadeh Nikfarjam, PhD, chose to explore mentions of drug-related adverse skin reactions, including rash, blistering, nail changes and psoriasis flares in more than eight million past Inspire posts. (Faithful readers of this blog may remember that dozens of Inspire members have shared their health journeys here.)
Reviewing past posts, they found that forum participants receiving one of three chemotherapy drugs began discussing skin reactions to the treatments six to nine months before they were linked to the drugs in published scientific literature or reported by the Food and Drug Administration. They did so by designing a computer program to analyze the content of messages among forum members.
As Shah explained:
We first built a way to find mentions of drugs in the patient posts; then Julia and Azadeh trained a computer to recognize when a mention of a disease condition might describe an adverse drug event. We next took a set of drug-adverse event associations known to be true, and a set that is known to be not true to learn when the pattern of co-mentions of a drug and a specific adverse event are significant enough to consider the two to be truly associated.
Using this technique to examine past posts the team was able to see that patients began discussing potential adverse effects related to their chemotherapy treatments much sooner than they were formally linked by the FDA or in scientific publications. They also identified in 23 people a potentially new relationship between use of the drug erlotinib and a loss of ability to sweat properly.
The researchers are hopeful that this type of computational analysis of social media platforms may lead in the future to the more rapid identification of adverse events caused by newly approved drugs.
As Sarin told me:
We were surprised to find that we could detect specific skin drug reactions in internet health forums six to nine months before any published reports. We were also struck by the detail with which users describe their clinical symptoms. In addition, this pipeline detected a new drug reaction that has not yet been reported in medical literature. This demonstrates the robust potential of internet health forums to provide useful and timely information for doctors and medical researchers.
Photo by Anonymous Account
