Identifying Patients on Twitter and Learning from Their Personal Experience: The Case of IBD (Preprint)

BACKGROUND Social media serve as an alternate information source for patients, who use them to share information and provide social support. Though large amounts of health-related data are being posted on Twitter and other social networking platforms each day, research using social media data for understanding chronic conditions and patients' lifestyles is still lacking. OBJECTIVE In this research we contribute to closing this gap by providing a framework for identifying patients with Inflammatory Bowel Disease (IBD) on Twitter and learning from their personal experience. We enable the analysis of patients' tweets by building a classifier of Twitter users that distinguishes patients from other entities. The research aims to assess the feasibility of using social media data to promote chronically ill patients' wellbeing, by relying on the wisdom of the crowd for identifying healthy lifestyles. We seek to leverage posts describing patients' daily activities and the influence on their wellbeing for characterizing different treatments and understanding what works for whom. METHODS In the first stage of the research, a machine learning method combining both social network analysis and natural language processing was used to classify users as patients or not automatically. Three types of features were considered: (1) the user's behavior on Twitter, (2) the content of the user's tweets, and (3) the social structure of the user's network. Different classification algorithms were examined and compared using two measures (F1-score and precision) over 10-fold cross-validation. In the second stage of the research, the obtained classification methods were used to collect tweets of patients, in which they refer to the different lifestyle changes they endure in order to deal with their disease. Using IBM Watson Service for entity sentiment analysis, we calculated the average sentiment of 420 lifestyle-related words that IBD patients use when describing their daily routine. RESULTS The best classification results (F1-score 0.808 and precision 0.809) for identifying IBD patients among Twitter users were achieved by a multiple-instance learning approach, which constitutes the novelty of this research. The sentiment analysis of tweets written by IBD patients identified frequently mentioned lifestyles and their influence on patients' wellbeing. The findings reinforced what is known about suitable nutrition for IBD, and several foods that are known to cause inflammation were highlighted as words with negative sentiment. CONCLUSIONS Patients everywhere use social media to share health and treatment information, learn from each other's experiences, and provide social support. Mining these informative conversations may shed some light on patients' ways of life and support chronic conditions research.