This talk presents methods and results of Opinion Mining applied to discussions of personal health. Currently 19%-28% of Internet users participate in online health discussions. 49% of those participants are looking for personal testimonials related to health and health care. This work studies sentiment and factual transitions on an online medical forum where users correspond in English.We work with discussions dedicated to reproductive technologies, an emotionally-charged issue. In several learning problems, we demonstrate that multi-class sentiment classification significantly improves when messages are represented by affective terms combined with sentiment and factual transition information (paired t-test, P=0.0011).