Learning Hormonal Therapy Medication Adherence from an Online Breast Cancer Forum

Abstract

Online health communities (OHCs) have become popular online environments for patients seeking and sharing treatment experiences. These platforms enable us to move beyond traditional sources of clinical information for learning about a patient’s long-term adherence to treatment. In spite of this opportunity, large-scale self-composed online free text brings challenges in processing and understanding patients’ health-related behaviors. Additionally, it has been shown that social support from trusted relationships (e.g., family and friends) positively influences treatment adherence in offline environments, but much less is known about the online setting. In this chapter, we showed that user activities in online health communities can be applied to learn about their medication adherence. Specifically, we focused on a 5-year hormonal therapy, a highly prevalent long-term treatment for breast cancer, with varying completion rates, in the breastcancer.org OHC. We characterized online user activities with emotion of self-disclosure and social interaction and focused on learning how these activities are associated with different adherence behaviors. To do so, we first designed a machine learning classifier to extract three types of adherence behaviors (taking, interruption, and completion of hormonal therapy), and then studied how emotions differ when patients mentioned different adherence behaviors. To examine the effect of social interaction, we relied on reciprocity (specifically in the form of reciprocal response to each other) to measure the social support between each other in this OHC. We first examined how reciprocity is related to time active in the OHC and the tones communicated by authors in their posts (e.g., emotions, writing styles, and social tendencies), and then assessed if such reciprocity is associated with treatment adherence. We found that patients in online health communities tend to exhibit fear with taking events, anger with interruption events, and joy (with a tinge of sadness and disgust) with completion events. We also found that the volume of the reciprocity is positively associated with completing the 5-year protocol, rather than the rate of the reciprocity or the fraction of the posts that received replies. We anticipate that our methodology can be applied to study treatment adherence for other diseases using online self-reported information.