CitationSaxe, Glenn N.; Ma, Sisi; Morales, Lea J.; & Aliferis, Constantic (2018). Machine Learning and Causal Network Modeling for Research on PTSD and Suicide.
This Symposium will highlight the utility of machine-learning predictive classification—with a causal discovery feature selection—for research on child and adolescent PTSD and suicide. These methods enable both reliable and accurate prediction and the discovery of causal processes related to the etiology of mental illnesses.
Research with 2 cohorts of children will be presented. The first is from a longitudinal study of 163 acutely traumatized children to predict which children will acquire PTSD several months after trauma exposure. The second is from the National Longitudinal Study of Adolescent to Adult Health (ADD Health) dataset of 4799 adolescents assessed at wave 1 (age 16 years), with a broad range of risk variables to predict suicide attempts 1 year later. In both studies, machine-learning methods using both support vector machine and random forest classification were used to predict the targets of PTSD symptoms and suicide attempts, respectively. In addition, causal discovery feature selection methods were incorporated using Markov Boundary induction to identify causal variables, pathways, and networks related to the target variables.
In both studies, reliable and accurate predictive models of PTSD and suicide attempts could be obtained [areas under the curve (AUCs) approximately 0.8] and a set of causal variables could be detected within causal models for these respective mental illness outcomes. The causal models each contained a range of biopsychosocial variables, many of which are remediable through intervention.
Machine-learning predictive classification—with causal discovery feature selection—can enable accurate prediction and causal detection in research related to PTSD and suicide in children and adolescents.
Reference TypeConference paper
Book TitleAmerican Academy of Child & Adolescent Psychiatry 65th Annual Meeting
Author(s)Saxe, Glenn N.
Morales, Lea J.