Patient reports in social media as safety evidence
What patients say in public about their medicines, on social networks, forums and websites built for reporting side effects, can hold reports of suspected adverse drug reactions and shows how patients perceive a medicine's benefits and risks. Whether it counts as evidence for a medicine's safety and benefit-risk balance turns on what the medium holds, what a posting would need to be a valid report, what the studies of it show, the channels built to collect patient reports, and the methods and guidelines still missing.
Why patient postings reach pharmacovigilance
Patients increasingly use social media to share their experiences with diseases, drugs, medical devices, vaccines and treatments, but as a source for adverse drug reaction reporting and pharmacovigilance this material is largely underutilised. It is underused partly because the data are unstructured and partly because the technology for assessing and analysing them lags behind.
Module VI of the European Medicines Agency's good pharmacovigilance practices states that marketing authorisation holders should regularly screen internet or digital media under their management or responsibility for potential reports of suspected adverse reactions. A report a holder finds in a digital medium it does not sponsor should still be assessed to determine whether it qualifies for reporting. Because patients increasingly post and discuss their experiences with diseases, drugs and treatments on social media, a vast amount of data is now generated about reactions to medicines that holders are required to monitor, assess and act on as appropriate.
Traditionally, adverse event reporting relies on physicians and drug safety groups, who serve as gatekeepers to validate the reports. Social media measurement and monitoring provides the opportunity to complement traditional surveillance, which relies on clinicians and patients reporting proactively through registries such as the FDA's MedWatch and typically experiences significant delays and systematic under-reporting.
What social media holds
Digital media covers audio, video and images in a computer-readable format, including websites, web pages, blogs, vlogs, social networking sites, internet forums, chat rooms and health portals. To these should be added the internet of things: uniquely identifiable embedded computing devices interconnected within the existing internet infrastructure, such as GPS devices, cars, cameras, coaching watches, sphygmomanometers, glucose measuring devices, insulin pumps and bathroom scales.
Estimates made in 2015 held that these websites would increase the amount of recorded data to 44 zettabytes by 2020, from 1 zettabyte in 2010, a massive amount of unstructured data against an estimated 5 zettabytes from structured sources by 2020. By the same estimates, 9% of all unstructured data would relate to healthcare, and half of that to drugs. Internet-based applications are easily accessible to large groups of people, and so can be updated immediately and with high frequency.
The information gathered through digital media is increasing exponentially, especially through data contributed by social media, and most of the data behind that growth are unstructured, such as tweets, comments on social media sites such as Facebook, and videos posted on websites such as YouTube. Current tools and techniques for analysing unstructured data, especially in pharmacoepidemiology and safety, are very limited. Extracting relationships between drugs and their effects from social media is a complex challenge, a literature review has reported, because social media texts, typically posts or tweets, usually contain many typographical errors and patients use lay terminology for diseases, symptoms and indications.
Social media measurement can be useful in identifying adverse drug events, alongside surveys of pharmacovigilance data and the scientific literature. Instagram influencers, for example, frequently discuss health problems, including diagnoses, prescribed drugs and side effects, through photos and comments.
Postings as safety evidence
The definition of an adverse drug reaction and the criteria for a valid individual case safety report are challenging to apply to social media reports of adverse drug reactions.
Discussions of the validity and reliability of data have traditionally excluded unstructured data provided by patients from the range of acceptable evidence for evaluating the benefit-risk balance of medicinal products. Consequently, very few studies have extracted knowledge from these sources, and there are methodological and technological gaps that must be filled. Marketing departments of pharmaceutical companies have begun using social media to understand patients' perceptions of their drugs, but safety and pharmacovigilance departments are still sceptical about the validity of the knowledge extracted from it.
Studies of patient postings
As of 2015, only five articles on PubMed discussed tracking adverse drug reactions with the help of social media. One study identified more than 4000 Twitter posts resembling adverse events and showed them to be significantly correlated with data from the FDA Adverse Event Reporting System by System Organ Class. A mixed-methods study examined posts about aromatase inhibitor side effects that breast cancer survivors made on twelve message boards between 2002 and 2010. Of 25,256 posts relating to aromatase inhibitors, 18.2% mentioned at least one side effect, 12.8% mentioned discontinuing the drug and 28.1% mentioned switching aromatase inhibitors.
Internet data on adverse drug reactions from consumers may also serve to identify areas for service improvement, or topics about which patients need more education or information. An analysis of 3785 items from five social media sites found that patients with glaucoma had stronger positive feelings towards complementary therapies and treatments with a poor evidence base than towards medically proven therapies, suggesting a lack of awareness or education about the clinically proven treatments.
Online communities may also highlight topics of concern to patients, such as medication convenience or packaging, and side effects that are not discernible in clinical trials. Beyond adverse reactions, social media data can also help assess patients' risk perceptions. For example, an analysis of patient narratives on popular French social media websites for health-related topics found that patients' perceptions changed drastically after the withdrawal of all medicines containing benfluorex: before the withdrawal date most posts concerned efficacy, while those after it discussed cardiovascular side effects.
Web-based patient-reported outcomes can give holders and regulatory bodies an opportunity to understand the benefits and risks of medicines in the real world, but there are concerns about the validity of data from social media sites. This information could potentially help in the early detection of unrecognised side effects, and methodologies to collect, analyse and process it may serve as complementary tools for holders and the FDA to receive patient feedback. Knowledge of social media discussions will also help physicians understand how patients perceive their adverse reactions, manage the adverse effects of drugs, and develop strategies for improving treatment adherence.
These studies are small-scale, one-shot examples. Given the large volume of unstructured data on the internet, an efficient data solution is needed to detect adverse reaction reports on social media and bring them to the attention of marketing authorisation holders for validation and further action.
Channels for collecting patient reports
In 2015, few examples existed that could give insight into what is technically possible in detecting safety signals through the internet. One is AETracker, a module of IMS Health's Nexxus application suite that provides a cloud-based engine for monitoring adverse events, off-label usage and other legal, regulatory and reputational risks in company-sponsored digital assets, including social media accounts and mobile apps. Its pharmacovigilance experts review in real time, and confirm any false positives or alert the client within an hour of an adverse event being reported.
In a test project processing websites such as Twitter, Facebook, forums and Wikipedia pages, the system generated 281,971 records relating to a specific drug, flagged 15% of them as potentially interesting signals, and 1.7% were confirmed after analysis. The system flagged every record that looked like an adverse event, with 100% accuracy. The project showed that holders could save 85% of pharmacovigilance costs compared with a moderation approach, and that technological advances such as the Hadoop system, natural language processing and logical programming could provide useful methods for robust analysis of internet-based safety signals.
Beyond detecting safety signals, such tools allow holders and healthcare professionals to understand better the information exchanged about a drug and patients' and consumers' perceptions of it. In unpublished observations from social media conversations, use of the tool found that Copaxone was perceived as a highly effective first-line injectable for multiple sclerosis, followed by Rebif, Avonex and Betaseron in decreasing order of efficacy. Betaseron and Betaferon were perceived as easier to use and associated with fewer side effects, such as injection-site reactions, than Rebif and Avonex. Extavia and Avonex were perceived as more cost effective than Rebif, Copaxone and interferon β-1b drugs. According to these conversations, side effects were the main reason patients switched treatments.
In 2015 the WEB-RADR initiative, led by the MHRA in the UK, sought to investigate technologies for gathering adverse drug reaction data, and was to develop tools and recommend policy. It was a multi-stakeholder initiative that included developing a mobile app for collecting adverse reaction data, and the data collected through the app were expected to add to the established safety profiles of medicines, enable earlier detection of new signals, reveal new patterns or trends in reporting, and even provide a means for geo-pharmacovigilance. Technologies then in development, such as WEB-RADR, were to allow patients to report adverse reactions through apps or social media sites. In May 2016, 21 months after WEB-RADR's launch, the MHRA described the problems involved in dealing with the huge amount of data available and the approaches it was taking. By 2021 WEB-RADR, a project of the European Union's Innovative Medicines Initiative, had explored the value of social media and partnered with PatientsLikeMe, u-Motif and Voluntis, but it did not recommend the use of general social media for broad statistical analysis.
Ideally, reports collected through such technologies would be analysed alongside other sources of pharmacovigilance data, with any signals identified and regulatory action agreed. Feedback can also be provided directly to the patient, but this two-way communication should be handled with care because of its ethical and legal implications. Data gleaned from social media are also subject to validation, assessment and duplication issues. In 2015 a collaboration between IMS Health and Facebook allowed the majority of reports in AETracker to be deduplicated.
In 2012 RxISK.org was the first free website, sponsored neither by the pharmaceutical industry nor by advertising, for patients and their doctors to research and easily report drug side effects. It is operated by Data Based Medicine Limited, which the psychiatrist David Healy founded. While still under construction, the site stated that it would offer a medical timeline chart capturing essential information on treatment-induced problems, tag clouds to help convey the impact of problems on people's lives, and free access to the FDA's database of adverse events.
Methods, guidelines and capacity still lacking
Adverse drug reaction reporting methodologies are changing, and the amount of data generated through social media is growing rapidly. This gives holders and healthcare professionals an opportunity to use the information on adverse reactions shared there as a source of information and insight for the benefit-risk evaluation of drugs. There is an unmet and urgent need for policy, methods, guidelines and technology platforms describing the tracking of adverse reactions through social media, so that patients' voices there can be adequately heard and incorporated into the benefit-risk assessment of drugs. A big-data technology solution is required to assess and analyse this new data set and produce insightful and robust knowledge from it. Guidelines developed in consultation with industry, patients, regulators, academic groups and prescribers are urgently needed to suggest methodologies for collecting, analysing and processing this large pool of information. Governments and agencies have undertaken a number of initiatives to fill the gaps and unmet needs, and a first step would be the development of guidelines or position papers. A multi-stakeholder approach, including at least industry, patients, regulators, academic groups and prescribers, seems necessary for developing such guidelines.
As of 2021, although the potential value of social media data was increasing, neither research on social media-based pharmacovigilance nor social media monitoring and measurement more generally was in a position to supplant more traditional methods. Data from traditional health systems and patient-generated data, however, have complementary strengths and, combined, can lead to more robust public health systems.
One barrier to drug companies' use of social media is that analysing it, even for purposes other than drug safety, may generate a lot of adverse drug reaction information, which could go beyond their capacity to process individual case safety reports.