Reward Learning in Late-Life Suicidal Behavior

摘要 美国正面临自杀率上升的问题。然而，我们对为什么有些人，而不是 其他人则从考虑自杀到试图自杀。在过去的资助期间，我们已经表明， 决策受损的抑郁老年人更有可能从自杀意念发展到 行动上具体来说，使用决策实验，计算模型和功能磁共振成像，我们发现可复制的， 腹内侧和背外侧前额叶抽象改变导致的学习和选择过程缺陷 学习信号在这次更新申请中，我们建议通过研究人们在 自杀风险在认知和情感需求下做出决定，这些需求更能代表 自杀危机在我们的计算框架中，这些需求包括（i）高信息负载和（ii） 时间压力和迫在眉睫的威胁对信息处理造成的限制。我们已经开发 并验证了研究信息处理瓶颈的新实验和计算方法 在决策过程中。具体来说，我们的强化学习计算模型适用于行为和 神经成像数据，使我们能够检查人们如何使用他们有限的神经计算资源， 高信息负载下的良好决策。我们的初步研究表明，在这种情况下， (i)依赖于资源合理的策略来管理信息负载，（ii）受到背侧注意的影响， 扣带回-鳃盖神经网络，（iii）自杀未遂时可能中断，（iv）异常 背侧注意网络对信息负载的反应。因此，我们建议测试的一般假设， 有自杀风险的人容易出现信息处理瓶颈， 皮层网络我们将执行决策实验和认知计算模型（目标1）， 发现样本和非重叠复制样本（各n = 200），以确保发现结果对以下各项具有稳健性 自杀行为的临床和认知异质性两个样本将最大限度地包括个人 自杀受害者的代表，即老年抑郁自杀者，其中约一半幸存 近乎致命的尝试操纵信息负荷的功能性神经成像实验将询问 背侧注意网络和扣带回-鳃盖网络在决策过程中的神经计算动力学 在一个样本中制造（n = 200，目标2）。对精神病理学，人格，认知， 精神暴露和自杀未遂造成的脑损伤能让我们控制住关键的混乱。的 跨学科团队拥有自杀行为机制（Dombrovski）、决策神经科学 （Dombrovski，McGuire，Hallquist）、成像方法（Hallquist）和自杀风险管理（Szanto， Dombrovski）。这项工作符合NIMH关于自杀的优先研究议程的一个关键目标： 识别与自杀风险相关的认知功能障碍/神经回路概况”，并利用 强化学习和RDoC框架的有限能力结构。

ABSTRACT The US is facing rising suicide rates. Yet, we have only a limited understanding of why some people, but not others, progress from contemplating to attempting suicide. In the past funding period, we have shown that depressed older adults whose decision-making is impaired are more likely to progress from suicidal ideation to action. Specifically, using decision experiments, computational modeling, and fMRI, we have found replicable deficits in learning and choice processes paralleled by altered ventromedial and dorsolateral prefrontal abstract learning signals. In this renewal application, we propose to extend these findings by examining how people at risk for suicide make decisions under cognitive and emotional demands that are more representative of the suicidal crisis. In our computational framework these demands include (i) a high information load and (ii) constraints on information processing imposed by time pressure and impending threats. We have developed and validated new experimental and computational methods for studying information-processing bottlenecks during decision-making. Specifically, our reinforcement learning computational model applied to behavioral and neuroimaging data, enables us to examine how people use their limited neurocomputational resources to make good decisions under high information load. Our preliminary studies show that decision-making in this context (i) relies on resource-rational strategies for managing information load, (ii) is subserved by dorsal attention and cingulo-opercular networks, (iii) is likely disrupted in attempted suicide, (iv) a deficit paralleled by abnormal dorsal attention network responses to information load. We thus propose to test the general hypothesis that people at risk for suicide are prone to information-processing bottlenecks arising from alterations in these cortical networks. We will perform decision experiments and cognitive computational models (Aim 1) in a discovery sample and a non-overlapping replication sample (n = 200 each) to ensure that findings are robust to the clinical and cognitive heterogeneity of suicidal behavior. Both samples will include individuals maximally representative of suicide victims, namely older depressed suicide attempters, about half of whom survived near-lethal attempts. Functional neuroimaging experiments manipulating information load will interrogate the neurocomputational dynamics of the dorsal attention network and cingulo-opercular network during decision- making in one sample (n = 200, Aim 2). A careful characterization of psychopathology, personality, cognition, psychotropic exposure and brain damage from suicide attempts will allow us to control for key confounds. The interdisciplinary team has expertise in mechanisms of suicidal behavior (Dombrovski), decision neuroscience (Dombrovski, McGuire, Hallquist), imaging methods (Hallquist), and suicide risk management (Szanto, Dombrovski). This work aligns with a key objective of the NIMH’s prioritized research agenda on suicide: “to identify cognitive dysfunction/neural circuitry profiles … associated with suicide risk” and taps into the reinforcement learning and limited capacity constructs of the RDoC framework.