Prediction of Escalation in E-cigarette Use Among Twitter Users

Electronic cigarettes have witnessed a growing market in the last few years. While they been advertised as a healthier alternative to combustible cigarettes, they can lead to nicotine addiction and weakening of the immune system. The most popular e-cigarette brand in the United States, Juul, was subject to a lot of public scrutiny due to its marketing practices targeted towards adolescents. There is a growing body of research in the usage of tobacco, cannabis, and illicit drugs, with studies suggesting that the usage of one drug leading to an increase in the likelihood of another drug, commonly referred to as a "gateway drug." However, little is known about whether we can predict the escalation from tobacco use to other drugs and whether e-cigarettes are contributing to such escalation. In this work, we collect data from Twitter for users who use Juul hashtags and look for temporal patterns that can predict the escalation from tobacco to cannabis. After filtering out the commercial and promoter users out from the data, we look for changes from first e-cigarette to first cannabis use. We reduce the problem to a supervised classification problem that can detect this escalation. We make predictions for different time intervals using several classifiers which consider lexical features from the tweets, together with user features, such as status count. Our findings have implications for making adequate public policies in the area of health and education.