Modeling the impact of age and sex on a dimension of poly-substance use in adolescence: a longitudinal study from 11- to 17-years-old

Abstract

We examined the conceptual utility of modeling use of multiple substances as a trait represented by a unified dimension throughout adolescence. Adolescents (710 males and 676 females) participating in a longitudinal community study were asked whether they had used alcohol, tobacco, marijuana, other controlled substances, or uncontrolled substances (i.e. over-the-counter medications and inhalants) at ages 11, 14, and 17. Using an item response theory framework, model fit indices demonstrated that although all substance use remained part of a single latent dimension, model parameters differed with age and according to sex. The impact of sex was observed at the level of the overall dimension, with reported substance use generally indicating a higher trait level (i.e. greater severity) in females than in males. While using these substances provided good information on individual trait level in mid- to late-adolescence, the trait was poorly characterized by substance use in early adolescence. Across ages and sexes, use of alcohol and tobacco tended to indicate lower trait levels than use of marijuana, other controlled substances, and uncontrolled substances. All substances provided a similar amount of information on the underlying dimension (except for uncontrolled substances, which provided the least information). This suggests that measurement and interpretation of adolescent substance use is enhanced by the consideration of a wide range of substances.

Figure 1

Item Response and Information Functions for each substance resulting from the best-fit model (Table 2, model 1, unconstrained except across twins) parameterization, separated by age and sex. The x-axis represents location, or the level of the trait in Z-score units. The y-axis associated with the Item Response Functions is the probability of endorsing use of a given substance or class at any given trait level.

Figure 2

Test Information (black line) and Standard Error (grey line) Curves (plotted on a scale ranging from zero to six) resulting from the best-fit model (Table 2, model 1, unconstrained except across twins) parameterization, separated by age and sex. Test Information is the sum of the Item Information Functions (from Figure 1). The x-axis represents location, or the level of the trait in Z-score units.