Detection of Cognitive Impairment From eSAGE Metadata Using Machine Learning.

Area of Science:

• Neurology
• Computer Science
• Gerontology

Background:

• Mild cognitive impairment (MCI) and dementia (DM) diagnosis relies on cognitive assessments.
• Digital cognitive tests offer opportunities for enhanced data collection and analysis.
• Machine learning (ML) presents a powerful tool for analyzing complex health data.

Purpose of the Study:

• To investigate the utility of machine learning methods for predicting mild cognitive impairment (MCI) and dementia (DM).
• To leverage metadata from the digital Self-Administered Gerocognitive Examination (eSAGE) to improve diagnostic accuracy.

Main Methods:

• Utilized eSAGE scores and behavioral metadata (e.g., time per page, drawing speed, stroke length) from 66 patients (normal cognition, MCI, DM).
• Trained logistic regression (LR) and gradient boosting models to detect cognitive impairment (CI).
• Evaluated model performance using 10-fold cross-validation and metrics including AUC, accuracy, precision, recall, and F1 score.

Main Results:

• Logistic regression with feature selection achieved an AUC of 89.51% for CI detection using both behavioral data and scores.
• LR models demonstrated high AUCs for detecting MCI from normal cognition (84.00%) and DM from normal cognition (98.12%).
• Average stroke length emerged as a key predictive feature, enhancing CI detection AUC to 92.06% when combined with other scoring features.

Conclusions:

• eSAGE scores and associated metadata are valuable predictors of cognitive impairment.
• Combining eSAGE scores with ML-analyzed metadata enables more accurate and efficient detection of CI.