Extracting Medication Nonadherence Reasons with Sentiment-Enriched Deep Learning

Abstract

Medication nonadherence (MNA) refers to the behavior when patients do not take medications as prescribed. Adverse health outcomes of MNA cost the U.S. healthcare systems $290 billion annually. Understanding MNA and preventing harmful outcomes are an urgent goal for researchers, practitioners, and the pharmaceutical industry. Past years have witnessed rising patient engagement in social media, making it a cost-efficient and heterogeneous data source that can complement and deepen the understanding of MNA. Yet, such dataset is untapped in existing MNA studies. We present the first study to identify MNA reasons from health social media. Health social media analytics studies face technical challenges such as varied patient vocabulary and little relevant information. We develop the Sentiment-Enriched DEep Learning (SEDEL) to address these challenges. We evaluate SEDEL on 53,180 reviews about 180 drugs and achieve an F1 score of 90.18%. SEDEL significantly outperforms state-of-the-art baseline models. This study contributes to IS research in two aspects. First, we formally define the MNA reason mining problem and devise a novel deep-learning-based approach; second, our results provide direct implications for healthcare practitioners to understand patient behaviors and design interventions.