Cognitive function and eye movement patterns during reading tasks in older adults

Progressive cognitive decline associated with aging can be measured by observing alterations in eye movement behavior during reading. This study explores the association between eye movement profiles and cognitive performance in older adults during reading-based tasks in virtual reality.

Sixty older adults with Alzheimer's disease (AD), mild cognitive impairment (MCI), and healthy controls (HC) wore a virtual reality (VR) headset during a reading-based task. Recorded eye movements yielded four gaze features (F₂-F₅): fixation duration, region of interest (ROI) entry count, regression count, and saccade amplitude. With the reading-task total score (F₁), these formed a five-feature set (F₁-F₅) for logistic regression under leakage-safe stratified 10-fold cross-validation. Concurrent validity was assessed against paper-based Mini-Mental State Examination (MMSE).

The reading-task total score was strongly correlated with MMSE (r = 0.92, p < 0.001). Eye-tracking features showed significant group differences, with longer fixation duration, more ROI entries, more reading-order regressions, and larger saccade amplitude in cognitively impaired groups. The three-class combined model reached macro area under the curve (AUC) = 1.000, accuracy = 0.997, and macro F1 = 0.997; for MCI versus HC, eye-tracking-only and combined models reached AUC = 1.000 compared with AUC = 0.895 for the reading-task score alone.

Integrating gaze features with the reading-task total score can improve cognitive assessment by capturing process-level information. Participants with AD and MCI exhibited longer fixation durations, more ROI entries, and more reading regressions than controls, with the largest effects in AD. The combined model showed its clearest value in the three-class setting.