Co-occurring Problem Behaviors Assessing Suicide Risk Among High Risk Adolescents: A Classification And Regression Tree Model

Considerable research confirms that at-risk youth evidence clusters of problem behaviors. For intervention studies with adolescents, the assessment of suicide-risk is often critical and challenging. Suicide risk behaviors often emerge in the research setting because they are linked to other problems including substance use, depression, and delinquency. Sorting through the relevance of putative risk and protective factors, as well as determining the future likelihood of involvement in a given risk behavior remains a challenge for prevention scientists. In this presentation we evaluate the application of CART (Classification And Regression Tree) models for depicting the role of protective and risk factors on suicide risk. Data are from a stratified random sample of over 1000 high school youth, over-sampling high-risk youth (potential school dropouts). Youth completed a questionnaire capturing a range of behaviors/attitudes related to school, substance use, mood, family, peers, and self. In addition, each youth received a brief face-to-face interview assessment (Screen for Youth Suicide Risk) measuring suicide risk behaviors. CART is used to establish best predictors of suicide behaviors (e.g. attempts, ideation, threats) by searching through sets of known risk and protective factors through a sequence of exploring all bivariate associations. Each sequence generates a set of more and more homogenous branches, with greater ability to identify individuals who are at greater risk than others within a given branch. Comparisons are made with alternative predictive approaches including multiple regression, logistic regression and discriminant analysis. The analyses provides insight into the role of specific protective and related risk factors, as well as, identifies clinically interesting subtypes of at-risk individuals.