Predictive Validity of Latent Classes for Middle and High School Polysubstance Use

Background/Significance:Substance use among adolescents is associated with multiple risk factors including anxiety, depression, aggression, violence, poor parent and peer relationships, academic problems, and negative neighborhood perceptions. Given these associations, it is possible to test the predictive validity of the latent class solutions for middle and high school students by comparing the classes on these characteristics. This paper answers the following question: are the classes identified in the previous analysis associated with known outcomes and predictors of youth success?

Methods:Classes were ordered according to the findings of the previous study to reflect a pattern of increasingly severe polysubstance use. Youth were assigned to their highest probability class. Classes were examined in relation to known correlates of substance use including mental health problems, delinquency and antisocial behavior, academic problems, community and neighborhood characteristics, and social stress. The most severe class was compared to the abstainer class on these factors using t-tests and chi-square tests. Because of the non-linear pattern in the high school model, the top two classes – user and severe user – were compared to the abstainer class. The classes were also subsumed into an ordinal variable of increasing severity of use, and Pearson and point-biserial correlations of this ordinal variable with these factors were also estimated.

Results:Among high school students, youth in the two most severe classes (users with high cigarette and marijuana use and severe users with the highest use of alcohol, inhalants, and prescription drugs) showed significantly worse levels on all of the factors described above. Among middle school students, youth in the severe class had worse outcomes on most measures compared to youth in the abstainer class (excluding relational aggression and anxiety). On the correlation with the ordinal severity of use variable, these factors had the hypothesized signs, with high positive associations on measures of youth delinquency and negative associations with predictors of youth success. The correlations for delinquent friends and neighborhood crime were large for both middle and high school youth.

Conclusions/Implications: The results support the predictive validity of the polysubstance use latent class solutions and inform potential avenues of intervention among rural youth. The validity of these classes implies that they can be used to classify youth according to lifetime substance use patterns and can be used in predictive models to identify youth at risk for different patterns of substance use. Given that youth with high cigarette and marijuana use and those with high levels of alcohol, inhalant, and prescription drug use showed significantly worse psychosocial outcomes, and given that these analyses are cross-sectional and observational (and the direction of causality is unclear), this study has direct implications for substance abuse treatment by identifying these risk factors to target. Youth with severe polysubstance use would likely benefit from additional services, including family counseling, mental health screening, and opportunities to engage with prosocial youth. Thus, the polysubstance use latent class solutions not only provide a means of categorizing levels of use among adolescents, but also provide insight into potential co-occurring problems.