The progressive nature of Parkinson's disease (PD) affords emphasis on accurate early-stage individual-level assessment of risk and intervention appropriateness. In PD, cognitive impairment (CI) may follow or precede motor symptoms but are generally underdetected. In addition to impeding daily functioning and quality of life, CIs increase the risk for later conversion to dementia, providing a pressing need to develop novel tools to detect and interpret them. Connectome-based predictive modelling (CPM) is an emerging machine-learning approach to individual prediction that holds translational promise due to its noninvasiveness and simple implementation. The aim of this study was to investigate CPM's potential to predict and understand CIs in PD. Resting-state functional connectivity from 58 patients with PD of varying cognitive status was used to train a CPM-model to predict a global cognitive composite (GCC) score. The model was validated using cross-validation, permutation testing, and internal stability analyses. The combined predictive strength of two brain connectivity networks, positive and negative, directly and inversely correlated with GCC, respectively, was assessed. The model significantly predicted individual GCC scores = 0.62, < .05. Separately, the positive and negative networks were similar in performance, s ≥ .58, s < .05, but varied in anatomical distribution. This study identified a connectome predictive of cognitive scores in PD, with features overlapping with established and emerging evidence on aberrant connectivity in PD-related CIs. Overall, CPM appears promising for clinical translation in this population, but longitudinal studies with out-of-sample validation are needed.

中文翻译：

---

基于连接组的预测模型评估帕金森病的个体认知状态

帕金森病 (PD) 的渐进性强调了对风险和干预适当性进行准确的早期个体层面评估。在 PD 中，认知障碍 (CI) 可能出现在运动症状之后或之前，但通常未被发现。除了妨碍日常功能和生活质量之外，CI 还增加了以后转化为痴呆症的风险，因此迫切需要开发新的工具来检测和解释它们。基于连接组的预测模型（CPM）是一种新兴的个体预测机器学习方法，由于其非侵入性和简单的实现而具有转化前景。本研究的目的是调查 CPM 预测和理解 PD 中 CI 的潜力。使用 58 名不同认知状态的 PD 患者的静息态功能连接来训练 CPM 模型，以预测整体认知综合 (GCC) 评分。该模型通过交叉验证、排列测试和内部稳定性分析进行了验证。评估了分别与 GCC 正相关和负相关的两个大脑连接网络的综合预测强度。该模型显着预测了个体 GCC 分数 = 0.62，< .05。另外，正负网络的性能相似，s ≥ .58，s < .05，但解剖分布不同。这项研究确定了预测帕金森病认知评分的连接组，其特征与帕金森病相关 CI 中已建立的和新出现的异常连接证据重叠。总体而言，CPM 在这一人群中的临床转化似乎很有希望，但需要进行样本外验证的纵向研究。