Comparing classification vs. continuum models of the structure of substance dependence and abuse.

Abstract

Substance use disorders are classified as categorical disorders by prominent nosologies [1]. A bevy of structural equation models have suggested dimensional solutions to drug dependence and abuse criteria. However, it is well known that factor models can fit categorical structures, and class models can fit dimensional structures. Recent research has thus compared relative fits of both latent class and trait models, and in some cases mixtures of latent traits. Results have been inconsistent, in large part due to the level of analysis. We attend to both problems by fitting models in a large sample with high base rates of alcohol, marijuana, cocaine, and stimulant disorders, allowing us to fit complex models both at fine and coarse levels of analysis (e.g., only alcohol items versus all items from a variety of drug classes). In general, dependence and abuse items from different drug classes can be modeled with drug-specific factors, one per drug. When more complex models are fit only to alcohol items, the best fitting model is a mixture of latent traits that maps closely onto DSM-IV-TR [1] nosology.