Predicting Long-Term Chemotherapy-Related Cognitive Impairment

摘要 化疗相关的认知功能障碍（CRCI）影响了估计60%的患者， 影响生活质量。目前，还没有确定的方法来预测哪些患者会 发展CRCI。这些信息可以通过帮助临床医生进行治疗来改变实践 为患者个体做决策。我们已经证明，大脑网络（“连接体”）是 在CRCI患者中显著改变。因此，我们测量了患者的连接体， 任何治疗，并证明这些连接体特性可以与 机器学习预测化疗后1年的认知障碍，准确率为100%。的 拟议的项目旨在测试这个初步的预测模型在一个新的，更大的样本与 验证其用于临床实践的总体目标。我们将招募100名新诊断的患者 计划接受辅助化疗的原发性乳腺癌患者， 治疗，包括全身麻醉手术，化疗后1个月再次治疗 一年后。我们还将招募匹配的健康女性对照组， 的间隔我们将联合收割机将这些数据与我们在先前研究中获得的回顾性数据结合起来， 每组共150个样本。健康对照组的数据将用于确定损伤 乳腺癌患者的状态，并提供一个典型的连接体组织的模板， 对比我们假设我们的机器学习模型可以准确预测1年后的情况。 化疗认知障碍，它将比包括患者的模型更准确， 相关和医学变量。我们还将检查连接体的纵向变化 与损伤亚型相关的组织（即持续性与迟发性损伤）以及 特定功能网络的变化（例如默认模式、显著性、执行注意力和感觉注意力） 电机网络）。这些信息将为CRCI的神经机制提供新的见解 也可以帮助我们完善预测模型。

ABSTRACT Chemotherapy-related cognitive impairment (CRCI) affects an estimated 60% of patients, negatively impacting quality of life. Currently, there is no established method for predicting which patients will develop CRCI. This information could be practice-changing by assisting clinicians with treatment decision-making for individual patients. We have shown that the brain network (“connectome”) is significantly altered in patients with CRCI. Therefore, we measured the connectome is patients prior to any treatment and demonstrated that these connectome properties could be used in combination with machine learning to predict 1 year post-chemotherapy cognitive impairment with 100% accuracy. The proposed project aims to test this preliminary prediction model in a new, larger sample with the overarching goal of validating its use for clinical practice. We will enroll 100 newly diagnosed patients with primary breast cancer scheduled for adjuvant chemotherapy who will be assessed prior to any treatment, including surgery with general anesthesia, 1 month after chemotherapy treatment and again 1 year later. We will also enroll matched healthy female controls who will be assessed at yoked intervals. We will combine these data with retrospective data we obtained during a prior study for a total sample of 150 in each group. Data from healthy controls will be used to determine impairment status in patients with breast cancer and to provide a template of typical connectome organization for comparison. We hypothesize that our machine learning model will accurately predict 1 year post- chemotherapy cognitive impairment and that it will be more accurate than a model that includes patient- related and medical variables alone. We will also examine longitudinal changes in connectome organization associated with impairment subtypes (i.e. persistent vs. late onset impairment) as well as changes in specific functional networks (e.g. default mode, salience, executive-attention and sensory- motor networks). This information will provide novel insights regarding the neural mechanisms of CRCI and may also help us refine our prediction models.