Extracting information from mental health records can be useful for large-scale clinical studies (e.g., to predict medication adherence or to understand medication effects) in this clinical specialty largely underserved by the Natural Language Processing (NLP) community. Vocabularies that contain medical terms for specific clinical use-cases, such as signs, symptoms, histories, social risk factors, are valuable resources for the development of NLP systems that aid clinicians in extracting information from text. Substance abuse is an important variable for many clinical use-cases, but, to our knowledge, there are no publicly available vocabularies that cover these types of terms. In this study, we apply and combine three methods for generating vocabularies related to substance abuse. We propose a simple and systematic method to generate highly relevant vocabularies and evaluate these vocabularies with respect to size and content, as well as coverage and relevance when applied to authentic psychiatric notes.