Mathematical and Computational Modeling of Suicidal Thoughts and Behaviors

Abstract

Suicide is one of the most devastating, complex, and perplexing of all human behaviors. Unfortunately, despite centuries of scientific inquiry, suicidal thoughts and behaviors remain exceedingly difficult to understand, predict, and prevent. This dissertation aims to develop and integrate methods from across the clinical and computational sciences, with the goal of advancing our ability to capture and model the immense complexity of suicidal thoughts and behaviors. Paper 1 harnesses recent advances in real-time monitoring technology to monitor patients’ momentary suicidal thoughts during psychiatric hospitalization, and models dynamic fluctuations in these data to predict suicide attempts during the high-risk time period following hospital discharge. Paper 2 takes an idiographic approach, building personalized machine learning models to predict suicidal thoughts as they unfold in the real world, in real time. Paper 3 takes a complementary theory-driven computational approach to construct and evaluate a formal mathematical model of suicide as a dynamical system, using systems of differential equations. Together, findings from this dissertation suggest that real-time monitoring data can advance our ability to predict suicidal thoughts and behaviors (at both the group and individual level), particularly when combined with data-driven computational methods such as machine learning. In addition, this dissertation demonstrates the promise of formal mathematical modeling in advancing the understanding of suicide, by allowing for direct observation of a theorized system’s behavior over time. Future directions for both data- and theory-driven computational methods in suicide research are presented in the general discussion.