Discovering Barriers to Opioid Addiction Treatment from Social Media: A Similarity Network-Based Deep Learning Approach
42 Pages Posted: 27 Mar 2019
Date Written: March 4, 2019
Opioid use disorder (OUD) refers to the physical and psychological reliance on opioids. OUD costs the US healthcare systems $504 billion annually and poses significant mortality risk for patients. Understanding and mitigating the barriers to OUD treatment is a high-priority area for healthcare and IS researchers, practitioners, and policymakers. Current OUD treatment studies largely rely on surveys and reviews. However, the response rate of these surveys is low because patients are reluctant to share their OUD experience for fear of stigma in society. Such social stigma significantly limits the representativeness of the patient population in surveys. In this paper, we explore social media as a new source of data to study OUD treatments. Drug users increasingly participate in social media to share their experience anonymously. Yet their voice in social media has not been utilized in past studies. We develop the SImilarity Network-based DEep Learning (SINDEL) to discover barriers to OUD treatment from the patient narratives and address the challenge of morphs. SINDEL significantly outperforms state-of-the-art baseline models, reaching a precision of 85.31%, a recall of 70.14%, and an F1 score of 76.79%. Thirteen types of OUD treatment barriers were identified and verified by domain experts. This study contributes to IS literature by proposing a novel deep-learning-based analytical approach with impactful implications for health practitioners.
Keywords: deep learning, text mining, opioid addiction, computational data science, design science
Suggested Citation: Suggested Citation