Abstract
Background: The large-scale identification of people at risk of transitioning from relatively lower-risk to higher-risk alcohol use (e.g. problem drinking) remains a public health challenge despite advances in the identification of risk and protective factors.Objective: This observational study used machine learning to identify Reddit (social media platform) posting activity associated with transitioning from lower- to higher-risk forms of alcohol use.Methods: We employed bottom-up and top-down approaches to identify lower- and higher-risk alcohol-related subreddits. Using a non-parametric negative control procedure, we estimated each of 10,006 Reddit communities' risk of progression from lower- to higher-risk alcohol-related communities and applied a random forest model to predict progression among individual Reddit members. Eligible Reddit members had posted on Reddit for two or more years before their first post in a lower-risk alcohol-related community and for three or more years after that (N = 4,160).Results: Our methodology identified 42 alcohol-related communities, four of which were suggestive of problem drinking. Five communities were significantly associated with progression. Random forests model's risk scores for individual members correlated with their progression to higher-risk communities at 0.30; the model predicted progression of individual Reddit members with a 0.92 area under the curve.Conclusions: Posting in communities dedicated to other substance use, depression, and occupation in the food service industry was associated with posting activity suggestive of problem drinking 3 years later. Posting activity on Reddit may be used for early detection of people at higher risk of transitioning from lower- to higher-risk forms of alcohol use.