Eigenbehaviour as an Indicator of Cognitive Abilities

Area of Science:

• Digital Health
• Machine Learning
• Gerontology

Background:

• Digital biomarkers are increasingly vital in healthcare, particularly for continuous monitoring.
• Cognitive ability assessment is crucial for aging populations, especially for those living alone and in clinical trials.

Purpose of the Study:

• To introduce a new digital biomarker for cognitive ability using ambient sensor data.
• To evaluate its effectiveness in long-term monitoring and clinical assessment of older adults.

Main Methods:

• Utilized indoor location data from passive infrared sensors to construct a location matrix over several weeks.
• Calculated reconstruction error from the matrix's eigenvectors.
• Employed linear regression for cognitive score prediction and support-vector machines for classification.

Main Results:

• The reconstruction error demonstrated strong predictive performance for high cognitive ability.
• Achieved high discriminative performance (ROC AUC of 0.94) in classifying normal versus mild cognitive impairment, outperforming age alone.
• The method is entirely unobtrusive, imposing no patient burden.

Conclusions:

• Reconstruction error from ambient sensor data is a promising digital biomarker for cognitive health.
• It offers a non-invasive tool for binary classification of cognitive status.
• Potential for predicting inter-individual differences in cognition, though with less precision for lower cognitive levels.