Socially adverse – inside the Web-RADR project

18 February 2016



The under-reporting of adverse drug reactions through traditional channels is a limitation of the current pharmacovigilance system. Reports on social media by patients about their experiences with certain drugs represent a new source of data for post-marketing safety surveillance. David Lewis, global head of safety at Novartis, and Phil Tregunno of the Medicines and Healthcare Products Regulatory Agency, explain how the Web-RADR project is exploring the feasibility of this approach.


Pharmacovigilance is undoubtedly one of the cornerstones of any clinical trial. Should participants experience any adverse drug reactions (ADRs), this will have important implications for whether the drug is approved and how safe it is determined to be.

Ideally, all ADRs should be reported straight away to ensure that patients' health is protected and that the trial continues to proceed along clear evidence-based lines.

While pharmacovigilance is criticalat the early stages of a trial, surveillance is still important during the post-marketing phase. If a country has a strong system of ADR reporting in place, it is more likely that promising drugs will be approved quickly, subject to the need for detailed pharmacovigilance in the years ahead.

It is imperative that any safety issues be recorded even in the case of older drugs. The better the surveillance system, the easier it will be for providers and patients to make informed decisions about a therapy. Through adequate reporting, pharma companies can find out how their drugs are used, how often ADRs occur and how safe a given medicine is in relation to its competitors.

Biased reporting

Unfortunately, pharmacovigilance is blighted by the problem of under-reporting, with many ADRs going unrecorded throughout the entire life cycle of a medicine.

"I joined the industry many moons ago, and under-reporting has been the bane of my whole working life," says David Lewis, global head of safety at Novartis. "There has been a series of papers to say that even in a clinical trial, where investigators are paid to look for specific adverse reactions, there is always a reporting bias.

"This may be to do with the doctor's specialism, what the patient chooses to tell the doctor or the fact that most healthcare providers are very busy. But there is almost invariably a significant level of under-reporting, and this applies everywhere, from the earliest phase of clinical trials to the point when a medicine is bought off the counter in a supermarket."

These issues are compounded by the fact that the level of under-reporting differs from the product. When a drug has widely publicised side effects, it is likely that most ADRs will be reported. Conversely, if the drug hasn't commanded much media attention, the actual number of ADRs may be far higher than first appears.

This variability means it's difficult to estimate the true impact of under-reporting, although a systematic review published in 2006 suggested a median under-reporting rate of about 95%.

As pharmacovigilance legislation grows tighter, these limitations are becoming more unacceptable. Therefore, a number of researchers are starting to wonder whether a different channel altogether - such as social media - might help improve post-marketing surveillance.

Lewis is deeply involved in these endeavours. At present, he is coordinating the involvement of pharmaceutical companies in a €2.3-million three-year public-private project called Web-RADR (Recognising Adverse Drug Reactions). The aim of this project is simple: to see whether social media and other digital technologies might be used for pharmacovigilance purposes.

Cut out the middleman

By September 2016, the project hopes to deliver a finalised mobile phone app for the reporting of suspected ADRs to EU regulators, as well as new technical tools for mining the public data shared on social media websites.

"I see everyone, from my kids to my father-in-law, at 90-something years of age, interacting on social and digital media," says Lewis. "Are we potentially missing a whole cohort of safety data that we can capture, collate or even mine for information purposes, with appropriate medical/scientific analyses to give us some sort of public health intelligence? I don't claim that I know there is something out there, but I think it would be crazy not to do this particular piece of research."

One of his partners in the project is Phil Tregunno of the Vigilance, Intelligence & Research Group at the Medicines and Healthcare Products Regulatory Agency (MHRA). Approaching these issues through a regulatory lens, Tregunno is interested in how technology can be used to facilitate improvements in public health, particularly when it comes to reporting ADRs.

"Social media is of paramount importance to us as an area that should be explored," he says. "We want to develop tools for new types of data and really understand their scientific impact before we make policy recommendations. This project will hopefully add robust scientific evidence that enables us to develop guidance for using social media data for pharmacovigilance purposes."

Researchers currently have many more questions than they do answers because this is a relatively new area of exploration. While social media may represent a valuable reservoir of information for pharmacovigilance, there are a few ethical and technical challenges remaining to be addressed.

"There's been quite a strong focus from the pharma industry on the more commercial, brand-oriented side of social media," says Lewis. "What we wanted to do here was to take a more considered, strategic look at the potential it gives us. To rush in would have been a foolish move in my view; it's important that the evidence will dictate the next steps."

Technically, the big question is how social media can be adequately mined for ADR reports. While a disgruntled patient might take to Twitter to complain about their reaction to a new medicine, they'd be unlikely to express that view in regulator-friendly terminology.

Researchers are, therefore, being faced with huge volumes of data, phrased in a vernacular language they're not used to interpreting through this framework. Through the use of machine learning algorithms, they hope to be able to determine whether people are talking positively or negatively about a product and whether they're referring to side effects.

They will also need to address any issues around repeat data, particularly in cases where people post the same information on multiple social media sites, or when they're retweeting or quoting someone else's viewpoint.

"You've got challenges in terms of identification and interpretation of data. Then, once you've identified these cases, you've got challenges in signal detection - which of these issues truly constitutes something new? The project's got activities in all these areas," says Tregunno.

"When we go through formal curation, it astounds me as to how much information can be yielded in the form of possible case reports," adds Lewis. "Even for products that are many tens of years on the market, we are seeing tens, hundreds or even in rare cases thousands of postings that could equate to case reports.

"What we're doing in the second part of Web-RADR is to try to identify any possible ADRs and compare that with what we see with traditional methods. That's where the scientific comparison comes in. The real longer-term ambition is to use statistical analyses to see whether we can find signals of safety concern from this data."

Yellow gold

Another big question that steps into ethical territory is whether tweets and Facebook posts might be treated as ADR reports in their own right. If not, should the people who post these items be encouraged to submit a formal ADR report? And who should be tasked with getting in touch to make that suggestion?

While the Web-RADR project is only concerning itself with publicly available data and is not breaching any social media privacy settings, there is still a fear that a regulatory agency mining posts for information may come across as somewhat creepy.

"My personal concern is also an industry-wide concern - I don't want to be viewed as Big Brother trying to pry into personal data or read medical information I'm not allowed to look at," says Lewis. "So I need some form of consent, and I want to engage as many people as possible to want to use the available tools."

One of these tools is an app that patients can use, rather less contentiously, to submit reports in the appropriate format. This part of the Web-RADR project has already come to fruition, with the first version of the Yellow Card smartphone app launched in July 2015.

The app is based on the Yellow Card Scheme, which was introduced more than 50 years ago in the wake of the thalidomide disaster. Supplementing the MHRA's existing Yellow Card website, where patients can submit ADR information, it can be downloaded on iOS and Android platforms, and simplifies the process of drug reporting. The app also enables users to receive news and alerts about specific medicines of interest.

"The app doesn't just use data the patient is willing to freely share with us; it also brings the potential that the data is of good quality and comes directly from the source - it's not modified by what a doctor has heard," says Lewis. "It's freely shared and, potentially, it gives us the chance to feed back, which is something we don't do very well at the moment."

The next step for the app will be to develop versions suited to different countries. The team is hoping to introduce a Dutch-language version early in 2016, and has also started working with HALMED, the regulatory agency in Croatia. It is then intended to make the app's code available as freeware so that it can be easily adopted elsewhere, functioning exactly the same as a case report from a patient or doctor.

The next phase for social media will be to develop a dashboard that can be used to screen social and digital media sites. This would incorporate a signal-detection algorithm that enables a comparison with traditional reporting mechanisms.

"We can identify whether we're augmenting things and finding new things out, or whether we're just repeating the same things we've done under traditional methods," says Lewis.

Once the technologies have evolved sufficiently, the team want to focus on regulations to cultivate clear policy recommendations about where this data is and isn't useful. Tregunno believes that, as they move into the final phase of the Web-RADR project, a look at what could constitute good pharmacovigilance practice for social and digital media will be the main focus.

"I think that area of strong, robust scientific analysis leading to appropriate policy recommendations is where we'll see the most activity behind the scenes," he says.

For Lewis, projects such as Web-RADR present a key opportunity to explore what is currently uncharted territory.

"We know that patients and healthcare professionals are already changing the way they use technology to interact with pharmaceutical companies and regulators," he says. "We know there's a whole corpus of care data generated through social media, but we don't know the value of the data that's been posted. It's simply not been established. So we want to transform pharmacovigilance in a traditional sense to pharmacovigilance in a digital world."

Under-reporting of adverse drug reactions is a huge issue for pharmacovigilance.
Web-RADR can mine public data from social media sites, but those running it are weary of how this practice can come across and are keen to find ways to properly engage the public.


Privacy Policy
We have updated our privacy policy. In the latest update it explains what cookies are and how we use them on our site. To learn more about cookies and their benefits, please view our privacy policy. Please be aware that parts of this site will not function correctly if you disable cookies. By continuing to use this site, you consent to our use of cookies in accordance with our privacy policy unless you have disabled them.