Related Experiment Video
Updated: Jul 6, 2026
Methodology for Establishing a Community-Wide Life Laboratory for Capturing Unobtrusive and Continuous Remote Activity and Health Data
Published on: July 27, 2018
Integrating actigraphy with demographic data enhances cognitive performance prediction: a multimodal UK biobank
Mohammad Mahdi Ghiasi1, Ryan Stanley Falck1, Teresa Liu-Ambrose2
1School of Biomedical Engineering, University of British Columbia, Vancouver, BC, Canada; Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, British Columbia, Canada; Centre for Aging SMART, Vancouver Coastal Health Research Institute, Vancouver, British Columbia, Canada.
Wrist-worn actigraphy significantly improves machine learning prediction of cognitive performance, outperforming demographic factors alone. Raw activity data enhances models, highlighting the need for sex-aware cognitive assessment strategies.
Area of Science:
- Gerontology
- Computational Neuroscience
- Biomedical Engineering
Background:
- Wrist-worn actigraphy offers continuous, non-invasive monitoring of rest-activity patterns, relevant for cognitive health.
- Current utility of actigraphy for predicting cognitive function beyond demographics requires large-scale validation.
Purpose of the Study:
- Evaluate actigraphy's effectiveness in enhancing machine learning prediction of cognitive performance (Digit Symbol Substitution Test).
- Identify key predictors for model explainability and assess performance fairness across sexes.
Main Methods:
- Utilized UK Biobank data (N=42,707) with 24-hour actigraphy, demographics, and DSST scores.
- Trained Extra Trees (ET) classifiers using demographics alone, then integrated raw or sine-transformed actigraphy features.
- Evaluated model performance using ROC AUC and PR AUC on a held-out test set.
Main Results:
- Raw actigraphy features significantly improved ET model performance (ROC AUC 0.76, PR AUC 0.83) compared to demographics alone (ROC AUC 0.66, PR AUC 0.75).
- Age, income, and education were top demographic predictors; activity timing (7-8 AM, 5-10 PM) was most informative from actigraphy.
- Sex-stratified analysis showed nuanced differences in predicting cognitive performance.
Conclusions:
- Raw actigraphy data demonstrably enhances predictive models for cognitive performance beyond demographics.
- Supports integrating wearable-derived activity data into cognitive assessment frameworks.
- Highlights the necessity of sex-aware modeling for equitable cognitive health evaluations.
