This study aims to identify behavioral health phenotypes using Medicaid claims of adults with schizophrenia and machine learning. This research was performed using an Latent Dirichlet Allocation (LDA), a form of topic modeling with a 3-level hierarchical generative model that works well in analyzing discrete data and requires less data than other techniques, such as neural networks, to attain adequate fit. Using LDA with a 5-fold cross-validation, 5 behavioral health phenotypes were identified: depression, substance use, mania-mixed mood, anxiety-paranoid, and conduct disorder-developmentally delayed. Pairwise comparisons were able to find significant differences between phenotypes based on demographics (age, sex, race, etc.), such as Black/African American adults being significantly different across all phenotypes, except for mania-mixed mood and depression, while the highest proportion of White adults were in the mania-mixed mood phenotype. It was also found that about ¾ of the cohort were on more than one medication at a time, and based on phenotype, there was a significant difference between the class of medication prescribed (antipsychotic, mood stabilizer, hypnotic, etc.). The authors hypothesize that the most effective psychotropic treatments will vary by the behavioral health phenotypes established in this study, and will be further influenced by demographics.