Computational Mechanisms of Coping with Distress: An Instantiation of Dynamic Action Selection Theory in Suicidal Youth

Background

Suicide is a global public health crisis and the third leading cause of death among U.S. youth. A major barrier to prevention is the lack of a mechanistic account of how maladaptive coping strategies such as suicidal ideation and self-injury are selected, reinforced, and repeated. Although many individuals report that these behaviors provide short-term relief from intense emotions, suggesting a negative reinforcement process, this phenomenon has not been formalized within a computational decision-making framework. Understanding how suicidal behavior becomes a learned response to distress is essential for developing interventions that are both timely and personalized.

Aims

Our objective is to examine how adolescents evaluate and update the perceived utility of coping strategies and how these processes contribute to suicide risk.

Aim 1: Develop a computational model of coping-as-action-selection that formalizes how individuals weigh and update the costs and benefits of different coping behaviors.

Aim 2: Integrate model-derived parameters with clinical risk factors and neurocognitive indices to identify coping vulnerability subtypes and evaluate their ability to predict suicide-related outcomes over a three-month follow-up.

Aim 3: Apply the reinforcement learning model to real-time ecological momentary assessment (EMA) data to quantify how suicidal adolescents learn from the emotional consequences of their coping choices, including suicidal ideation and self-injury.

Methods We will generate synthetic datasets and run simulations using a k-armed bandit framework, treating each coping strategy as an option whose expected value must be learned. Simulations will build on an adapted version of Dr. Franks OpAL* reinforcement learning model, which distinguishes learning from costs and benefits and incorporates beliefs about environmental reward richness. Our extensions will include (1) a perseveration parameter to capture habitual reuse of coping strategies and (2) a temporal-decay mechanism to test whether the time since a strategy was last selected influences value updating.

We will then collect three months of EMA data from 90 suicidal adolescent inpatients. Participants will repeatedly report which coping strategies they used and how their mood changed afterward, enabling us to estimate trial-by-trial prediction errors and value updates in daily life. Baseline clinical risk factors and neurocognitive task parameters derived from reinforcement learning and Bayesian modeling will be combined using mixture modeling to identify coping subtypes. These subtypes will then be tested as predictors of suicidal ideation, self-injury, suicide attempts, and treatment change at follow-up.

Impact and Future Directions: By integrating a neurocomputational model of learning with real-world assessments of coping, this project provides a mechanistic account of how suicidal thoughts and behaviors become reinforced over time. Identifying when maladaptive coping strategies begin to yield reinforcing emotional effects may reveal emerging periods of heightened risk that are not captured by traditional assessments. The resulting subtypes can inform more precise, individualized interventions that address each youths unique learning patterns and coping tendencies. Ultimately, this work advances suicide prevention by offering a dynamic, personalized framework for understanding and intervening on risk.