Using Computational Modeling to Test Reinforcement Learning as a Predictor of Response in Family-Based Treatment for Adolescent Anorexia Nervosa

项目总结/摘要 神经性厌食症（AN）与致命的医疗并发症的重大风险和每年的费用有关， 美国约112亿美元。虽然青少年AN的家庭治疗（FBT）已经证明， 在针对AN症状的有效性方面，高达60%接受FBT的个人没有完全缓解。值得注意的是， 以前的工作没有探索FBT反应的神经认知预测因子，这可能有助于促进FBT反应。 确定治疗机制和制定针对无应答者的靶向治疗方案。当 考虑到哪些神经认知过程可能与FBT反应有关，越来越多的研究表明， 成人AN的特征可以在于强化学习的改变。此外，以其他形式开展工作， 精神病理学表明，强化学习可以预测对行为治疗的反应。然而，在这方面， 到目前为止，很少有研究测试了强化学习和治疗结果之间的变化， 已经探索了强化学习和FBT结果之间的关联。目前的调查将 利用认知神经科学和计算建模的方法来探索强化学习， AN青少年（n = 58）和健康对照受试者（n = 58），以及其与治疗的关系 FBT的结果。我将测试以下假设：目标1：与成人现有数据一致，AN组 与HC相比，在学习任务中表现较差，学习损失减少， 利用先前学到的信息。目标2：在AN组中，从损失中学习的比率也较低， 由于较低的探索/利用参数值将与1个月和6个月随访时较差的结果相关， 可操作为较低的体重和更大的进食障碍认知症状。在导师的指导下 五位专家横跨生物统计学，青少年临床研究，计算建模和认知 神经科学，目前以病人为导向的职业发展奖将允许我获得培训， 在这些领域的交叉点促进独特的专业知识。从短期来看，目前的调查数据将 产生可用于了解AN症状持续性的见解，并确定潜在的方法， 改善治疗效果。从长远来看，目前的项目将允许我开始我的职业生涯，并采取下一个 在一个程序化的研究路线的步骤融合互补的专业知识，在神经认知，计算 方法和青少年干预发展。

PROJECT SUMMARY/ABSTRACT Anorexia nervosa (AN) is associated with significant risk for deadly medical complications and an annual cost to the US of around $11.2 billion. Although Family-Based Treatment (FBT) for adolescent AN has demonstrated effectiveness in targeting symptoms of AN, up to 60% of individuals who receive FBT do not remit fully. Notably, no prior work has explored neurocognitive predictors of FBT response, which may help to facilitate the identification of treatment mechanisms and formulation of targeted treatments for non-responders. When considering what neurocognitive processes may be implicated in FBT response, increasing work suggests that adult AN may be characterized by alterations in reinforcement learning. Further, work in other forms of psychopathology suggests that reinforcement learning may predict response to behavioral treatments. However, few studies to date have tested alterations between reinforcement learning and treatment outcome, and none have explored associations between reinforcement learning and FBT outcome. The current investigation will leverage methods from cognitive neuroscience and computational modeling to explore reinforcement learning in adolescents with AN (n = 58) and healthy control subjects (n = 58), as well as its associations with treatment outcome in FBT. I will test the following hypotheses: Aim 1: Consistent with existing data in adults, the AN group will demonstrate poorer performance in the learning task compared to HC, decreased loss learning, and poorer exploitation of prior learned information. Aim 2: Within the AN group, lower rates of learning from loss, as well as lower explore/exploit parameter values will relate to poorer outcomes at 1- and 6-month follow-ups, operationalized as lower body weight and greater eating disorder cognitive symptoms. With the mentorship of five experts across biostatistics, adolescent clinical research, computational modeling, and cognitive neuroscience, the current patient-oriented career development award will allow me access to training that will facilitate unique expertise at the intersection of these fields. Short-term, data from the current investigation will yield insights that can be used to understand the persistence of AN symptoms and identify potential methods to improve treatment outcomes. Long-term, the current project will allow me to launch my career and take the next steps in a programmatic line of research merging complementary expertise in neurocognition, computational methods, and adolescent intervention development.