Chu Chenyin, Wang Yihan, L H Huynh Andrew, Ng Ka Weng, Liu Shu, Ji Guangyan, Doecke James, Fripp Jurgen, Masters Colin L, Goudey Benjamin, Jin Liang, Pan Yijun
Faculty of Medicine, Nursing and Health Sciences, Monash University, Melbourne, Victoria, Australia.
Department of Medicine, Austin Health, Heidelberg, Victoria, Australia.
Age Ageing. 2025 Jul 1;54(7). doi: 10.1093/ageing/afaf198.
Cognition monitoring is crucial for care planning in people with mild cognitive impairment (MCI) and Alzheimer's dementia (AD).
To develop a machine learning model to assist cognition monitoring.
Florey Fusion Model (FFM) was constructed and validated in two phases: (i) model development and cross-validation using data collected via the Australian Imaging, Biomarker, and Lifestyle of Ageing (AIBL) study, and (ii) simulation and missing data trials with 30 new participants.
This prognostic study recruited 238 participants in the AIBL study. Support vector machine, gradient boosting and random forest were trialled to develop the FFM. Cognitive decline was assessed via changes in Clinical Dementia Rating Sum of Boxes (CDR-SB) and Mini-Mental State Examination (MMSE) scores. Model performance was evaluated by cross validation and compared against baseline models.
The FFM achieved a median area under receive character curve (AUC-ROC) of 0.91 (IQR 0.87-0.93) for MCI-to-AD progression prediction. A mean absolute error (MAE) of 1.32 (IQR 1.30-1.33) for CDR-SB and 1.51 (IQR 1.50-1.52) for MMSE was achieved for 3-year cognition forecast. Simulation and missing data trials yielded up to 94% accuracy for MCI-to-AD conversion and MAEs of 1.27-2.12 for CDR-SB score prediction.
The FFM holds the potential to facilitate cognition monitoring in people with MCI/AD; however, a larger trial will be required to refine it as a clinical grade tool.
认知监测对于轻度认知障碍（MCI）和阿尔茨海默病（AD）患者的护理规划至关重要。
开发一种机器学习模型以辅助认知监测。
弗洛里融合模型（FFM）分两个阶段构建和验证：（i）使用通过澳大利亚衰老成像、生物标志物和生活方式（AIBL）研究收集的数据进行模型开发和交叉验证，以及（ii）对30名新参与者进行模拟和缺失数据试验。
这项预后研究在AIBL研究中招募了238名参与者。试用支持向量机、梯度提升和随机森林来开发FFM。通过临床痴呆评定框总和（CDR-SB）和简易精神状态检查表（MMSE）评分的变化评估认知衰退。通过交叉验证评估模型性能，并与基线模型进行比较。
FFM在预测MCI向AD进展方面，接收特征曲线下面积（AUC-ROC）的中位数为0.91（四分位间距0.87-0.93）。在进行3年认知预测时，CDR-SB的平均绝对误差（MAE）为1.32（四分位间距1.30-1.33），MMSE为1.51（四分位间距1.50-1.52）。模拟和缺失数据试验在MCI向AD转化方面的准确率高达94%，在CDR-SB评分预测方面的MAE为1.27-2.12。
FFM有潜力促进MCI/AD患者的认知监测；然而，需要更大规模的试验将其完善为临床级工具。
