Predicting Sleep Quality via Unsupervised Learning of Cardiac Activity.
Summary
This study introduces a new unsupervised method to predict perceived sleep quality using cardiac activity from polysomnography (PSG) data. The novel approach significantly improves prediction accuracy, suggesting new avenues for sleep analysis.
Area of Science:
- Cardiology
- Sleep Medicine
- Data Science
Background:
- Perceived sleep quality impacts mood, productivity, and performance but is difficult to predict using passive physiological and behavioral signals alone.
- Traditional methods for sleep quality modeling rely on feature-based approaches using established metrics.
Purpose of the Study:
- To propose and validate a novel, unsupervised method for predicting perceived sleep quality.
- To explore cardiac activity states derived from polysomnography (PSG) for improved sleep quality assessment.
Main Methods:
- An unsupervised method was developed to derive novel states of cardiac activity from over 6,800 polysomnography (PSG) recordings.
- The correlation between time spent in these cardiac states and perceived sleep quality was assessed.
- A longitudinal study over one month with 16 participants was used for validation.
Main Results:
- The proposed method achieved a balanced accuracy of 68% in classifying perceived sleep quality, outperforming prior feature-based methods.
- A strong correlation was found between the proportion of time in derived cardiac states and perceived sleep quality.
- The identified cardiac activity states appeared to oppose traditional sleep stages, despite being explainable by simple cardiac metrics.
Conclusions:
- The novel unsupervised method offers a significant advancement in predicting perceived sleep quality.
- The findings suggest the existence of under-investigated sleep processes impacting perceived sleep quality.
- This research may necessitate a re-evaluation of current sleep analysis techniques, particularly for subjective sleep quality assessment.
