MenuBack to Poster-Presentations-Details
W 44: Social Listening for a New Product Launch and Beyond: How Does the Conversation Change Over Time?
Laurie S. Anderson
Safety Evaluation and Risk Management Scientist
GlaxoSmithKline United States
To describe characteristics of conversations in social media surrounding a new pharmaceutical product during the first four years of the product’s life cycle.
We reviewed de-identified English language social media posts from Facebook, Twitter, and several patient forums for a pharmaceutical product approved and launched in 2011. Each post was manually classified for its content, and results were summarized by quarter years.
We reviewed 4,095 social media posts from 1Q2011 around the time FDA approval was first obtained through 16 September 2015. Posts were provided and deidentified by a third party vendor searching under trade and generic names as well as common misspellings, then divided into quarters by months and curated by a team of trained health care providers. Each post was tagged if it belonged into one or more of 4 categories: adverse events, posts seeking information/asking a question, positive benefit discussions, and lack of effect discussions.
Over the first two years on the market, discussions of all four of these categories ranged from 0 to 43% of total mentions of the product across data sources. After two years on the market, however, these category ranges became much narrower and at lower rates, ranging from 1 to 18%. Specific numbers for adverse events were 14-43% in the first two years and 8-15% after two years; seeking information posts were 11-33% first two years and 8-18% thereafter; positive benefit discussions were 0-25% first two years and 7-16% thereafter; lack of effect discussions were 0-11% first two years and 1-4% afterwards.
Another interesting trend was in the overall volume of product discussions on Facebook, Twitter, and patient forums (dailystrength.org, healingwell.com, healthunlocked.com, and three indication-specific sites) over that same time period. The general trend was for discussions to take place most commonly on the indication-specific patient forums near launch, with Twitter picking up just over one year after approval and Facebook gaining more volume at around two years post-approval. Specifically, patient forums accounted for 80-100% of posts from 1Q2011 until 2Q2012, when Twitter accounted for 41% of the total volume. Twitter then stayed at above 40% for the remainder of the project while forums fell to under 20%. Facebook increased to 52% of the volume in 2Q2013, and remained primarily between 20 and 50% thereafter.
With social media continuing to rise in popularity as a communications tool, pharmacovigilance experts are striving to understand online discussions as a data source that might augment the current tools in use for drug safety surveillance. For this particular product, we showed a quite striking leveling out of social media discussions (as a percentage of total drug mentions) surrounding adverse events, information sought, positive benefit, and lack of effect at two years post-approval, which was sustained over nearly three years of continuing data. Clearly the findings may not be generalizable since this is the only product for which we currently have these data available. We are actively considering other new products on which to attempt replication of these results. If this general trend were true across many different products, it might serve as a baseline rate to aid in the identification of unusual spikes in the frequencies of discussions of these important events. This could be one of many steps necessary in pursuit of an automated approach to using social media listening for pharmacovigilance.
The second trend we noted involved detecting where the highest volume of posts were found over time (first in patient forums, then in Twitter and ultimately increasing in Facebook) which may help to quickly locate the bulk of early discussions in social media. Further research is needed to determine if either of these trends is generalizable, and to further characterize the nature of discussions on social media. The ability to automatically tag or classify posts will be key to increasing the scalability of the use of this technology.