A study last year found that many people reported adverse drug events on Twitter.

Online patient community PatientsLikeMe has found another partner for its massive repository of patient-generated data on health conditions and treatments, but it's not another pharma company or retail pharmacy. PatientsLikeMe has announced a research partnership with the FDA: The agency will assess the platform's feasibility as a way to generate adverse event reports, which the FDA uses to regulate drugs after their release into the market.

“Most clinical trials only represent the experience of several hundred or at most several thousand patients, making it impossible to anticipate all the potential side effects of drugs in the real world," PatientsLikeMe Co-Founder and President Ben Heywood said in a statement. "Patient-generated data give a more complete picture about a drug’s safety by providing a window into patients’ lives and healthcare experiences over time. We’re very encouraged by the FDA’s action to evaluate newer sources of data to help identify benefits and risks earlier.”

FDA already collects adverse event data in a number of ways including MedWatch, the FDA's voluntary reporting system, and the Sentinel Initiative, which consists of partnerships with health insurers and electronic medical records to get access to their aggregated data sets.

"Those are all coded data," Gerald Dal Pan, Director of the Office of Surveillance and Epidemiology at the Center for Drug Evaluation and Research at the FDA, told MobiHealthNews. "Here, we’re looking to see if we can supplement that information with more information about the patient’s voice, having the patients tell us their experience in their own words. ... We’re interested in going into a social media environment to see what we might be able to learn. We don’t know what we’re going to find." 

According to PatientsLikeMe, it has collected more than 110,000 adverse event reports on 1,000 different medications, data that the FDA will now be able to access and analyze in addition to its existing sources of information. Dal Pan stressed that the partnership is exploratory, but PatientsLikeMe could provide the right mix of data that is sufficiently detailed to be useful while also being sourced from a large group of consumers. That's important because the FDA needs to know a lot about the context of adverse events.

"When we look to detect adverse events, we’re trying to see if there’s some new thing we don’t know about," he said. "[But] we’re also trying to see, if there’s some new thing, is it in everybody? Is it in some people? Is it an interaction between the drug we’re interested in and some other drug? Are people using the medicine correctly? Is there a pattern of misunderstanding directions for use? We’re trying to learn about a lot of different aspects."

Most importantly, he said, the research collaboration will show the FDA whether PatientsLikeMe data is detailed enough to make it clear that adverse events really were caused by the drug being evaluated.

"One very basic thing we look at is 'Can we tell if the adverse event happened after the patient started the medicine, or was it there before the patient started the medicine?'" he said. "If we can’t do something like that, than it would limit the value of these types of systems. So the first step is really feasibility, and then we’ll see if we have interpretable information that may lead us to look at data from other sources that we have, or further study."

A study published last year found that Twitter could potentially be a lucrative source of otherwise unreported adverse drug events, but that the data is noisy and hard to parse. Dal Pan didn't comment on the likelihood of the FDA working with large, public social media platforms, but he did express some concern that they wouldn't be able to produce the level of granular data the FDA needs.

"If you look at the MedWatch report form, there’s a lot of boxes you can check, and then there’s a narrative space," he said. "The value of that MedWatch form to us is, how much information is in there that’s useful for an independent medical verification of whether the adverse event was caused by the drug or not. It doesn’t have to be very long, but it probably always will take more than 140 characters to make that independent judgement that there’s a reasonable association of a causal relationship between the drug and the adverse effect."