Tracking Subjective Sleep Quality and Mood With Mobile Sensing: Multiverse Study.

Area of Science:

• Digital Health
• Computational Psychiatry
• Mobile Health (mHealth)

Background:

• Traditional sleep quality tracking methods are often burdensome and intrusive.
• Developing effortless and unobtrusive sleep monitoring is crucial for advancing sleep research and clinical practice.
• Sleep quality significantly impacts mood and mood disorders.

Purpose of the Study:

• To assess the feasibility of using mobile sensing data to infer sleep quality.
• To determine the predictive power of mobile sensing features for subjective sleep quality (SSQ), negative affect (NA), and depression.
• To analyze how different data processing and modeling choices influence prediction accuracy.

Main Methods:

• Utilized data from a 2-week trial involving 50 participants, collecting mobile sensing (accelerometer, light, activity, screen, Wi-Fi) and experience sampling data.
• Employed a multiverse analysis approach, varying sensor selection, feature extraction, preprocessing, and statistical models.
• Validated models using training, validation, and test sets to ensure robustness and prevent overfitting.

Main Results:

• Most models demonstrated limited predictive power on the validation set (R² ≤ 0).
• The best models achieved notable R² values on the training set: 0.658 for SSQ, 0.779 for NA, and 0.074 for depression.
• Test set R² values for the best models were 0.348 (SSQ), 0.103 (NA), and 0.025 (depression).

Conclusions:

• The study highlights the significant impact of methodological choices (feature selection, preprocessing, modeling) on prediction outcomes.
• Promising predictive performance for subjective sleep quality (SSQ) suggests potential for further investigation.
• Mobile sensing offers a viable, unobtrusive approach for sleep quality assessment, warranting continued research.