Mixture model applications in depression phenotyping: practices, challenges, and recommendations

Applications of mixture models are prevalent in studying psychopathology across development, particularly for identifying typical co-occurring symptom presentations (or phenotypes) in depression. Researchers have used both longitudinal and cross-sectional designs with varied statistical methods. The current study focused on studies that applied latent profile analysis, latent class growth analysis, and growth mixture models to phenotype continuously treated depressive symptoms. The current study aims to (a) provide a brief overview of common mixture models that are used in depression phenotyping, (b) review empirical applications of these methods in cross-sectional and longitudinal research of depression, (c) discuss the methodological considerations and recommendations in identifying phenotypes of depression when continuously treated symptoms are used. In 72 studies, we found heterogeneity in mixture model specification, selection, and interpretation. We identified three challenges in current practices: a “garbage in, garbage out” problem, inconsistent use and reporting of model selection criteria, and diverse, incomparable, and incomplete phenotype characterizations. We recommend that researchers: (1) select and justify measures and models based on the research question during model specification; (2) report BIC and bootstrapped likelihood ratio tests of all compared models, grounding model selection on the philosophy of science during model comparison; (3) provide all parameter estimates and use R² measures for class characterization during model interpretation.