This study focuses on improving information access in health-related social media services through opinion mining. The first contribution is an extensive evaluation of various features—lexical, syntactic, semantic, network-based, sentiment-based, and word embeddings—for representing patient-authored texts in polarity classification. The second contribution addresses polar facts, objective information with polar connotations, which have been traditionally neglected in polarity classification. The study demonstrates the existence and importance of polar facts in the polarity classification of health information.