Detecting Suicide and Self-Harm Discussions Among Opioid Substance Users on Instagram Using Machine Learning

Background: Suicide and substance use disorder (SUD) pose serious public health challenges among young adults in the United States. Increasing social media use among these populations can be leveraged as an alternative method to detect characteristics of suicide-related topics and behavior among substance users. Objective: To detect and characterize suicide and self-harm related conversations co-occurring with SUD posts and comments on the popular social media platform Instagram. Methods: This study used big data and machine learning approaches to collect and classify Instagram posts containing 632 controlled substance-related hashtags. Posts were first classified for online drug diversion topics and then filtered to detect suicide and mental health discussions. Posts and comments were then manually annotated for SUD and mental health co-occurring themes. Associations between these characteristics were tested using the Chi-square test. Results: We detected 719 Instagram posts/comments that included user-generated discussions about suicide, substance use and/or mental health. Posts self-reporting SUD and mental health topics were also more likely to discuss suicide compared to those that did not discuss SUD and mental health topics, respectively (p < 0.001). Major themes observed included concurrent discussions of suicide ideation and attempts and low self-esteem. Conclusions: Our study results provide preliminary evidence of social media discussions about suicide and mental health among those with SUD. This co-occurrence represents a key health risk factor on a platform heavily utilized by young adults. Further studies are required to analyze specific patterns of suicide and self-harm ideations for the purposes of designing future suicide prevention campaigns through digital channels.