TY - GEN
T1 - Extracting Medication Nonadherence Reasons with Sentiment-Enriched Deep Learning
AU - Xie, Jiaheng
AU - Liu, Xiao
AU - Zeng, Daniel
AU - Fang, Xiao
N1 - Publisher Copyright:
© 2019, Springer Nature Switzerland AG.
PY - 2019
Y1 - 2019
N2 - 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.
AB - 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.
KW - Deep learning
KW - Health analytics
KW - Medication nonadherence
KW - Text mining
UR - http://www.scopus.com/inward/record.url?scp=85076754130&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85076754130&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-34482-5_26
DO - 10.1007/978-3-030-34482-5_26
M3 - Conference contribution
AN - SCOPUS:85076754130
SN - 9783030344818
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 294
EP - 301
BT - Smart Health - International Conference, ICSH 2019, Proceedings
A2 - Chen, Hsinchun
A2 - Zeng, Daniel
A2 - Yan, Xiangbin
A2 - Xing, Chunxiao
PB - Springer
T2 - 7th International Conference for Smart Health, ICSH 2019
Y2 - 1 July 2019 through 2 July 2019
ER -