Understanding Adolescent Alcohol Consumption: A Decision Tree Analysis of YRBSS Data

Introduction

Alcohol is currently the most common drug used by adolescents in the United States. The 2019 Youth Behavior Surveillance System (YRBSS) data reported 30% of high school students had consumed alcohol in the last 30 days. Alcohol consumption among adolescents can have negative effects on their mental and physical health. To address these concerns, it is beneficial to identify the risk factors and interactions that contribute to alcohol usage among adolescents, to develop effective prevention strategies. In this study, we used a decision tree analysis to identify the risk factors and interactions associated with alcohol usage among adolescents.

Methods

This study analyzes the 2019 YRBSS data focusing on 13,231 adolescents with an average age of 16 years. Of this group, 56.53% reported consuming alcohol. A classification decision tree was used to model the data, which was trained and tested using 75% and 25% of the total sample, respectively. To prevent overfitting, the model was fine-tuned using bootstrapping and the best complexity parameter was selected using the simplest tree by the cost complexity hyperparameter, with accuracy as the metric. The model incorporates 12 variables that relate to suicidality, sexual abuse and sexual behaviors, bullying and cyberbullying, drugs in school, vaping, and ethnicity.

Results

The decision tree comprised of 15 nodes, including the root node, and four levels. The model demonstrated strong performance in classifying alcohol use, with high sensitivity (0.76) and specificity (0.77), as well as a high kappa value (0.53) and overall accuracy (0.77). The tree split the sample into two main branches, with using an electronic vapor product and having sexual intecourse as the splitting variables. The sexual intercourse branch further subdivided into sub-branches based on the presence or absence of hopelessness, suicide ideation, being offered drugs in school, sexual abuse, and identifying as Hispanic with increasing proportions of alcohol consumption when these variables are present and interacting. The second branch comprised of adolescents who had used an electronic vapor product show that 82% of them reported alcohol consumption. Overall, the decision tree underscores the crucial role of electronic vapor use, hopelessness, being Hispanic, sexual intercourse, suicide ideation, and sexual abuse in adolescent alcohol consumption.

Discussion

The decision tree algorithm demonstrated high accuracy in predicting alcohol use among adolescents, with an accuracy of 76%. The study highlights the importance of identifying and considering interactions between risk factors when developing prevention and intervention to reduce adolescent alcohol use. Decision tree methods have been used in mental health research given it automatizes detection of main effects and interactions because they can handle complex relationships between predictors and outcomes while providing an intuitive visualization that can be communicated to clinicians. The study identified suicidality, sexual abuse, and hopelessness as risk factors for alcohol consumption, which is consistent with previous literature. Further research is needed to validate this study and to explore the effectiveness of targeted interventions in reducing alcohol consumption among adolescents.