Immediate word recall in cognitive assessment can predict dementia using machine learning techniques

Area of Science:

• Gerontology
• Cognitive Science
• Artificial Intelligence

Background:

• Dementia is a growing public health concern, with increasing prevalence in aging populations.
• Existing machine learning (ML) models for dementia prediction often achieve high accuracy but suffer from low sensitivity.
• The utility of specific cognitive assessment data, particularly word-recall features, for dementia prediction using ML remains underexplored.

Purpose of the Study:

• To investigate the effectiveness of word-recall cognitive features in developing ML models for dementia prediction.
• To assess the sensitivity performance of ML models in predicting dementia.
• To determine the optimal combination of sample person (SP) and proxy responses for dementia prediction.

Main Methods:

• Nine experiments were conducted using data from the National Health and Aging Trends Study (NHATS).
• Four ML algorithms (KNN, decision tree, random forest, ANN) were employed to build predictive models.
• Experiments focused on 'word-delay,' 'tell-words-you-can-recall,' and 'immediate-word-recall' tasks, analyzing SP and proxy responses.

Main Results:

• The 'immediate-word-recall' task, using combined SP and proxy responses, achieved a perfect sensitivity of 1.00 across all four ML models.
• 'Word-delay' and 'tell-words-you-can-recall' tasks demonstrated poor predictive performance.
• Combining SP and proxy responses in the 'word-delay' and 'tell-words-you-can-recall' tasks yielded the highest sensitivities of 0.60.

Conclusions:

• The combination of responses in an immediate-word-recall task from both SP and proxies is clinically valuable for dementia prediction.
• Immediate-word-recall cognitive assessment is a reliable method for predicting dementia.
• 'Word-delay' and 'tell-words-you-can-recall' tasks are not reliable for dementia prediction based on this study's findings.