Supervised learning approach a necessary and important step in monitoring drug-related social media posts
Social media, such as Facebook and Twitter, provides many users with an outlet to share opinions online among a broad audience. Some of these discussions concern their experiences with prescription and over-the-counter (OTC) medications. These conversations can have an outsized influence on the public perception of the products they discuss. The sentiment of individual posts regarding the efficacy and side effects of medications can be either positive or negative. Consumers may also disclose misuse or abuse of the product. Occasionally, widespread off-abel use may become apparent or even reveal anecdotal evidence of a drug’s efficacy for a new indication. While some posted content may be accurate, claims can be exaggerated, erroneous or unsubstantiated; regardless, anyone with access to the internet can be exposed to and influenced by this content.
Pharmaceutical manufacturers have no control over the content of posts by social media users, and it is nearly impossible to respond to every conversation concerning their products. Despite this (or because of this), it would be prudent for companies to monitor social media in order to capture the posts’ overall sentiment, potential trends in product use, or previously unreported adverse event types. Identifying unusual increases in tweets about a particular medication may lead to popular “threads”, where many users share their own experiences with the medicine or comment on a news report. Analysis of trending words and hashtags can give insight into trending themes and may also identify alternative spellings and words that Twitter users use to discuss particular medications. We hypothesize that monitoring social media can provide essential insights into unusual adverse events related to the medication or improper use of the medication by analyzing drug event pairs.
One of the most significant challenges of analyzing social media posts is the unstructured nature of the data. Mining for posts by keywords will often produce results that are unrelated to the medication of interest. A recent review of 10,000 tweets for Lyrica, for example, yielded only about 500 tweets (5%) that discussed the actual medication. Fortunately, advances in natural language processing and increased access to machine learning tools make it possible to train sophisticated classifiers to identify posts of interest. Such classifiers work behind-the-scenes, ensuring accurate analyses of social media posts for a particular medication. Once posts are analyzed and aggregated across topics and time, an interactive dashboard can be employed to deconstruct the data back into individual posts for qualitative analysis and identification of root causes that drive larger trends.
Veracuity’s Social Media Scanner operates in this manner. It utilizes natural language processing on aggregated social media posts specific to medications and summarizes key metrics. Customers will soon be able to inspect data on a dashboard in a user friendly and interactive environment.