Predicting and Evaluating Cognitive Status in Aging Populations Using Decision Tree Models

Objective: To improve the identification of cognitive impairment by distinguishing normal cognition (NC), mild cognitive impairment (MCI), and Alzheimer's disease (AD). Methods: A recursive partitioning tree model was developed using ARMADA data and the NIH Toolbox, a multidimensional health assessment tool. It incorporated demographic and clinical assessment variables to predict NC, MCI, and AD. Model performance was evaluated using AUC, precision, recall, and F1 score. Robustness was tested through 5-fold cross-validation, sensitivity, scenario, and subgroup analyses. Results: The model achieved macro-AUC and micro-AUC scores of 0.92 and 0.91 (training) and 0.89 and 0.86 (testing). Key predictors included the Picture Sequence Memory Test and List Sorting Working Memory Test. Cross-validation yielded 70.22% accuracy and a Kappa of 0.52. Conclusion: Machine learning effectively uses a small set of assessments to distinguish NC, MCI, and AD, offering a valuable tool to support clinical decision-making. Future research should validate this model across diverse populations.