A Content Analysis of Patient-Reported Medication Outcomes on Social Media

The small number of patients enrolled in clinical trials to test new drugs and the relatively short trial durations make it paramount to monitor drugs' effectiveness and risks after they are approved by the regulatory agency. A thorough evaluation of a drug's effectiveness, side effects, and social and economic influences can prevent serious health damage to the public and shed light on new drug discovery and development. Past research has examined spontaneous reporting systems and electronic health records systems as data sources to study medication outcomes. However, both data sources are not able to provide complete and unbiased pictures of patients' care, making it necessary to integrate new data sources, such as increasingly prevalent social media data. In this study, we compared and evaluated four social media sites, in terms of data coverage and quality using 11 disease-drug pairs of careful selection. We found some patients reported serendipitous new indications for the drugs they were using for comorbid conditions, which is truly valuable information for drug repositioning. We also identified five cases of informal use of English on social media that can be challenging for computers to process, including comparative sentiment, sarcasm, grammar errors, pronoun reference and semantic reference, and emoticons. Our study suggests that social media can be a complementary new data source for studying medication outcomes, and reliable natural language processing and text mining methods are needed to automatically mine social media data on a large scale.