(PS2-C52) Using Classification Trees Discern Risk Factors for Multiple Suicide Attempts vs. Risk Factors for Single Suicide Attempts Among Colorado Adolescents

For the past decade, researchers have attempted to identify individually salient predictors of suicide attempts in order to prevent future suicide attempts (e.g., Edgcomb et al., 2020; Yıldız et al., 2018). The results of these previous studies have limited capacity for identifying the multitude of factors that predict suicidal behaviors because these models circumvent the nuanced interactions between individual risk factors and the intersectionality of individuals who may be at risk for multiple suicide attempts (Hjelmeland & Knizek, 2016; Nock, 2016). Furthermore, these research studies often lack the sample size to draw comparisons between single suicide attempts and multiple suicide attempts. The purpose of this study is to fill these gaps by identifying and comparing the etiological risk factors that predict single and multiple suicide attempts among Colorado adolescents.

We examined a diverse sample of 45,013 Colorado adolescents (12-18-years-old; M_age = 15.69; Sex_Female = 50%, Sex_Male = 50%; 4.24% attempted suicide one time, 3.92% attempts suicide more than one time) to investigate through classification trees the differing risk factors for suicide across samples of individuals who attempted suicide one time and those who attempt multiple times in the last 12 months. Classification trees are person-centered analyses that use a machine learning algorithm to recursively evaluate predictors of a categorical outcome. Example variables that were be considered in this study include previous self-injurious thoughts and behaviors, substance use, demographic identites, and sexual behavior.

Preliminary classification trees indicated that suicidal thoughts and non-suicidal self-injury were salient predictors of both single and multiple suicide attempts. Individuals who attempted suicide multiple times were more likely to endorse having multiple sexual partners and lower frequency of using nicotine products in the last 30 days. The model generally performed well (xval error = 0.904±0.022). The classification tree investigating the predictors of one suicide attempt in the last year compared to those who did not attempt suicide did no perform well without overfitting, which could indicate that there were no predictors in this dataset that explained enough variance to predict single suicide attempts. Additional classification trees will be run on similar datasets collected at different times to investigate if these findings are consistent overtime.

The results of this study will be a crucial first step to creating and implementing specific prevention plans that target risk factors for single and multiple suicide attempts among adolescents. It is also hoped that optimized prevention plans will create a space in which we raise community awareness and decrease the taboo of suicide discussions through the identification of at-risk individuals (WHO, 2018).