The amount of content on social media platforms such as Twitter is expanding rapidly. Simultaneously, the lack of patient information seriously hinders the diagnosis and treatment of rare/intractable diseases. However, these patient communities are especially active on social media. Data from social media could serve as a source of patient-centric knowledge for these diseases complementary to the information collected in clinical settings and patient registries, and may also have potential for research use. To explore this question, we attempted to extract patient-centric knowledge from social media as a task for the 3-day Biomedical Linked Annotation Hackathon 6 (BLAH6). We selected amyotrophic lateral sclerosis and multiple sclerosis as use cases of rare and intractable diseases, respectively, and we extracted patient histories related to these health conditions from Twitter. Four diagnosed patients for each disease were selected. From the user timelines of these eight patients, we extracted tweets that might be related to health conditions. Based on our experiment, we show that our approach has considerable potential, although we identified problems that should be addressed in future attempts to mine information about rare/intractable diseases from Twitter.

The amount of content on social media platforms such as Twitter is expanding rapidly. Simultaneously, the lack of patient information seriously hinders the diagnosis and treatment of rare/intractable diseases. However, these patient communities are especially active on social media. Data from social media could serve as a source of patient-centric knowledge for these diseases complementary to the information collected in clinical settings and patient registries, and may also have potential for research use. To explore this question, we attempted to extract patient-centric knowledge from social media as a task for the 3-day Biomedical Linked Annotation Hackathon 6 (BLAH6). We selected amyotrophic lateral sclerosis and multiple sclerosis as use cases of rare and intractable diseases, respectively, and we extracted patient histories related to these health conditions from Twitter. Four diagnosed patients for each disease were selected. From the user timelines of these eight patients, we extracted tweets that might be related to health conditions. Based on our experiment, we show that our approach has considerable potential, although we identified problems that should be addressed in future attempts to mine information about rare/intractable diseases from Twitter.