more likely to belong to the users or the non-users class. For these subjects the training variable equals one, indicating that class membership needs to be estimated. Potential future users and true long-term survivors are distinguished during model fitting based on similarity of patterns of the covariates. If a covariate of an individual without reported first use is similar to a reported first user then this individual will have a higher probability of belonging to the user class. It should be noted though that given the small expected differences between non-users and future users with respect to our covariates, it is likely that at least a proportion of future users will have high probabilities of belonging to the long-term survivor class. A simple comparison of class proportions and prevalence rates of reported first use was used to evaluate this issue.
