Curtis B, Giorgi S, Ungar L, et al. AI-based analysis of social media language predicts addiction treatment dropout at 90 days. Neuropsychopharmacology 2023;48(11):1579-1585. doi:10.1038/s41386-023-01585-5
This article examines the ability of a large-language model to predict addiction treatment outcomes at 30, 60, and 90 days based on social media data. Participants were recruited from community-based drug-free outpatient treatment programs. Of the 504 individuals recruited, 269 met the language requirements in their social media posts, having written over 200 words on their Facebook page in the previous two years. At baseline, participants granted the research team access to their Facebook data and completed the abbreviated addiction severity index, 6th edition (ASI). Following the baseline assessment, participants began outpatient addiction treatment as usual and answered weekly text surveys about their substance use and abstinence. Digital phenotypes, 61 features per participant, were extracted from the social media data and used to predict drop-out risk, abstinence, and relapse. The digital phenotype predicted participants’ relapse, drop-out, and abstinence better than the ASI alone. The combination of the ASI and digital phenotype produced the most accurate predictions for drop-out. This was seen in the self-report data responses initially at 60 days and was clear by the 90-day endpoint. The authors suggest that this model, and models like it can help screen at the beginning of addiction treatment. Predicting relapse and drop-out may allow service providers to give additional attention and support to patients who are likely to relapse before the relapse occurs.