OBJECTIVES: To assess the value and feasibility of using social media to understand patient perspective on medical product risks and living with different medical conditions.

METHODS: A third-party vendor acquired 12 months of publicly available posts from Twitter and Patient forums mentioning paclitaxel and adalimumab. Data were automatically processed to standardize drug names and vernacular symptoms, and to remove duplicates and noise. Posts were manually reviewed by analysts to confirm automated outputs and to apply sentiment tags. Binary Matrix Factorization (BMF), Bayesian Estimation (BE), Latent Dirichlet Allocation (LDA), and Regular Expression Filtering (REF) were then used to identify trends and associations between user sentiment and topics such as adverse events (AEs), quality of life, and experiences living with the certain diseases.

RESULTS: 192,902 posts were collected and processed; 4,742 were then manually reviewed. Topics pertaining to insurance, infection, employment, and emotional impact of treatment or disease were most commonly discussed. Results from the various methods implemented were mixed: LDA did not provide unique insights but BMF helped to ascertain patient preferences among different groups. Sentiment tags provided a valuable dimension in analysis methods; however, these may need to be applied at the concept level, rather than the post-level, to best reflect complex patient perspectives. Twitter provided a high volume of less specific statements, while Forums encouraged patients to provide rich narratives similar in detail to in-person interviews.

CONCLUSIONS: Social media listening is an important tool in understanding the patient journey. Obtaining meaningful insights computationally is difficult; however, automated methods can facilitate human analysis. Continued work in this area would require a comprehensive taxonomy to help identify a wide variety of topics, as well as a dedicated sentiment classifier for each product or therapeutic area of interest. Much work remains to determine best practices for using this rapidly evolving data source.