Predicting cognitive scores from wearable-based digital physiological features using machine learning: data from a...

Area of Science:

• Neurology
• Biomedical Engineering
• Digital Health

Background:

• Continuous monitoring of cognitive function is crucial for managing mild cognitive impairment (MCI).
• Standardized tests are impractical for frequent assessment.
• Wearable sensors offer a promising avenue for remote cognitive function monitoring.

Purpose of the Study:

• To evaluate the predictive capability of digital physiological features from wrist-worn wearables for neuropsychological test scores in individuals with MCI.
• To explore the use of sensor data for tracking cognitive changes over time.

Main Methods:

• A 10-week clinical trial involving 30 individuals with MCI (aged 50-70) receiving a digital multidomain intervention.
• Cognitive performance assessed using the Neuropsychological Test Battery (NTB) before and after intervention.
• Physiological data (blood volume pulse, electrodermal activity, skin temperature) collected via Empatica E4 wearable; 106 features extracted.
• Supervised machine learning models trained to predict NTB scores using physiological features and demographics.

Main Results:

• Analysis of 96 data intervals from 17 individuals revealed strong correlations between physiological features and cognitive scores.
• Heart rate variability (HRV) showed the strongest correlation with executive function.
• The model achieved a correlation of r=0.69 for predicting actual executive function scores and r=0.61 for intra-individual changes.

Conclusions:

• Wearable-based physiological measures, particularly HRV, show significant potential for continuous cognitive function assessment in MCI.
• This technology can facilitate more personalized and responsive therapeutic interventions.
• Future research should further validate these findings in larger, diverse populations.