Recovery Speaks: Language Use Within a Peer-to-Peer Online Discussion Forum Predicts Relapse in Alcohol Use Disorder

• In Event: Mobile Health Communication: Health Apps, Activity Tracking, Texting, Online

Abstract

This paper investigates whether language used within a peer-to-peer discussion board can predict relapse for individuals treated for alcohol use disorder (AUD). 104 AUD sufferers accessed a mobile-based relapse-prevention system and reported their incidence of risky drinking at four, eight, and twelve months. Their patterns of language used were assessed using LIWC, a text analysis software. A logistic regression model was built to predict likelihood of engaging in risky drinking within a year based on language use in the discussion board, while controlling for baseline characteristics and utilization of the mobile system. Results show that whereas baseline characteristics and system use explained little variance in relapse, a small number of linguistic features including swearing and cognitive mechanism words contributed over two thirds of explained variance. The model achieved 84% accuracy in classifying relapse. As online communication expands, models that leverage user-submitted text toward predicting relapse will be increasingly scalable.

Authors

• Rachel Kornfield, University of Wisconsin - Madison

• Catalina Laura Toma, U of Wisconsin-Madison

• Tae Joon Moon, U of Wisconsin-Madison

• Dhavan Shah, Univ. of Wisconsin - Madison

• David H Gustafson, U of Wisconsin-Madison