Since the turn of the 21st century, the rise of the social web has brought about phenomenal changes – and opportunities – in the healthcare world. Online platforms have empowered patients to take an active role in their own treatment and allowed them to express themselves more frankly than they might in the presence of a physician. Simultaneously, providers and pharmaceutical companies alike have embraced healthcare consumerism and the tenets of patient-centred care.
Over the course of this tectonic shift, Treato has capitalised on the development of big data analytics to capture the unfiltered ‘patient voice’. To date we have collected, organised and analysed more than 2.3 billion user-generated, health-related posts from forums, blogs and social media sites. Pharma companies and healthcare organisations can utilise this data for strategy development, competitive research and patient segmentation, among other applications.
One of the many areas in which social data is employable is in the identification and assessment of adverse drug reactions (ADRs), events which lead to thousands of hospitalisations and deaths each year. What remains unclear, however, is how social data analytics compare in terms of reliability to more established processes for clinical data collection.
To answer this question, we partnered with researchers at Brigham and Women’s Hospital, Harvard Medical School, and Partners HealthCare System to conduct a pilot study on the differences and similarities between ADRs stored in electronic health records (EHRs) and concerns shared by patients on social media. Published in Drug Safety, the journal of the International Society of Pharmacovigilance, the study emphasises the growing role social data will play in personalising care and impacting health outcomes.
Drugs and databases analysed
The study focused on two drugs, aspirin and Lipitor, that had a relatively high prevalence among the ADR reports in both the clinical and the social media databases analysed. As explained above, our social media database contains user posts from health-related forums, blogs and social media sites, as well as the Treato.com discussion platform. These posts are collected using a patented algorithm that extracts relevant information through a natural language processing system combining medical ontologies and patient language dictionaries. It should be noted that the adverse events that appeared in the social media database were vetted by the research team based on clinical validity.
Boston’s Partners HealthCare System, which includes Brigham and Women’s Hospital and Massachusetts General Hospital, provided its Partners Enterprise-wide Allergy Repository (PEAR) as the clinical database. PEAR consists of allergy records corresponding to more than 700,000 patients at all the Partners’ hospitals and outpatient clinics. The allergy information was inputted by physicians who either chose from a list of reactions or entered the reactions as they were described by patients.
What the numbers showed
The analysis revealed significant overlap between the clinical and social media databases, as well as certain discrepancies that warrant further research. To begin with, the two most common reactions to aspirin in both data sets were hives (or another type of rash) and gastrointestinal upset. These reactions appeared at a higher frequency in the clinical data than the social data, though. Furthermore, the fourth- and fifth-most common concerns on social media, tinnitus and Reye’s syndrome, landed outside the top 10 and the top 20, respectively, in the clinical data.
It appears that tinnitus was in fact under-represented in the EHRs. As the authors explain, “aspirin studies support the social media data, with reports of up to 30% of patients experiencing hearing problems with long-term aspirin use, while our clinical data showed only low tinnitus prevalence.” Reye’s syndrome, on the other hand, is very rare in the general population. As such, its prevalence was likely overstated by its ranking among patient concerns on social media.
This data can offer key insights into how drugs perform outside of a carefully crafted clinical environment
In both data sets, musculoskeletal pain and cramps were among the three most common reactions to Lipitor. While musculoskeletal pain appeared nearly 50% of the time in the allergy records, however, it appeared in only 10% of patient concerns on social media. Conversely, fatigue and depression were highly prevalent in the social data but much less so in the clinical data. This suggests that patients and providers do not perceive Lipitor’s effects on mood and energy level in the same way, or that patients are reluctant to discuss mental health issues with their physicians.
A challenge presented by analysing social data for adverse drug events is establishing causality. In a post taken from Treato’s discussion platform, a patient reported experiencing itchy hands “days after starting atorvastatin [Lipitor] 10mg.” Without more context, it’s difficult to determine other potential explanations for this symptom. Similarly, in other cases patients might have been expressing apprehension about a side effect they heard or read about rather than one they actually experienced themselves. As for the frequency of each reaction, researchers cannot rule out the possibility that the same patient posted about a reaction on several different sites. Future advances in natural language processing will help neutralise these variables.
Beyond adverse drug events
Research by Treato has demonstrated the utility of social media analysis in areas of pharmaceutical development besides the collection and assessment of ADRs. One of these areas is the planning and execution of clinical trials. In recent years, opportunities for patients to become aware of and engage with clinical trials through social media have increased exponentially. While trial participants are in theory not supposed to discuss their experiences with others, many patients skirt the boundaries of confidentiality (or are perhaps unaware of just how public their social media profiles are).
Non-disclosure agreements notwithstanding, researchers can use social analysis to glean insights into how a trial is progressing. Last year, while cystic fibrosis treatment Orkambi was in the final stages of development, Treato found that 58% of trial participants who discussed the drug on social media “reported beneficial effects.” Moreover, 20% of social posts were written by patients who were awaiting the drug’s approval. This kind of information would have thrilled executives at Vertex Pharmaceuticals, the makers of Orkambi – and worried their competitors.
Looking ahead
Information shared by patients online, what I like to call the ‘human side of healthcare’, differs from clinical data in its unvarnished, unsolicited nature. This data offers key insights into how drugs perform outside a carefully crafted clinical environment: off-label usage, drug adherence and patient’s perception of a drug’s effects.
A drug tested on thousands could eventually be taken by tens of thousands more. Accordingly, it is imperative that clinical data be combined with social data to accurately portray the kaleidoscopic spectrum of patient experiences. In terms of reducing and preventing adverse drug events, the next step involves utilising social data analytics to generate hypotheses and test them in a clinical setting.
The findings of this study further solidify the clinical validity of data extracted from healthcare consumers’ online conversations. Given that the FDA receives more than a million adverse event reports each year, the current reporting system will inevitably fail to capture all of the events that occur, or as our results suggest, misinterpret a portion of the events it does capture. However, a new day is approaching in drug safety specifically and healthcare more broadly. In the next five years, social data analytics will revolutionise pharmacovigilance processes. The result will be better drugs and healthier lives.
