A Literature Review of Social Media-Based Data Mining for Health Outcomes Research

Abstract

Patient-generated health outcomes data are health outcomes created, recorded, gathered, or inferred by or from patients or their caregivers to address a health concern. A critical mass of patient-generated health outcome data has been accumulated on social media websites, which can offer a new potential data source for health outcomes research, in addition to electronic medical records (EMR), claims databases, the FDA Adverse Event Reporting System (FAERS), and survey data. Using the PubMed search engine, we systematically reviewed emerging research on mining patient-generated health outcomes in social media data to understand how this data and state-of-the-art text analysis techniques are utilized, as well as their related opportunities and challenges. We identified 19 full-text articles as the typical examples on this topic since 2011, indicating its novelty. The most analyzed health outcome was side effects due to medication (in 15 studies), while the most common methods to preprocess unstructured social media data were named entity recognition, normalization, and text mining-based feature construction. For analysis, researchers adopted content analysis, hypothesis testing, and machine learning models. When compared to EMR, claims, FAERS, and survey data, social media data comprise a large volume of information voluntarily contributed by patients not limited to one geographic location. Despite possible limitations, patient-generated health outcomes data from social media might promote further research on treatment effectiveness, adverse drug events, perceived value of treatment, and health-related quality of life. The challenge lies in the further improvement and customization of text mining methods.