Sleep is crucial for overall health, particularly mental health. Insufficient sleep is linked to negative health outcomes, including increased depression and anxiety. Monitoring sleep is essential for effective healthcare, but current methods like sleep logs and actigraphy face challenges such as burden, cost, and low engagement. Smartphones, ubiquitous and accessible, present a promising alternative for scalable sleep monitoring. While prior research explored smartphone-based sleep monitoring, challenges include reliance on self-reported data and the use of proprietary or unavailable apps. This study aimed to address these limitations by using the open-source mindLAMP app to develop and validate smartphone-based sleep measures in a population of college students experiencing elevated stress levels. The researchers hypothesized that passively collected smartphone sensor data could accurately estimate sleep duration and contribute to digital phenotyping of sleep habits.

The literature highlights the critical role of sleep in overall and mental health, emphasizing the need for effective and scalable sleep monitoring methods. Existing methods, such as sleep diaries and actigraphy, suffer from limitations concerning user compliance, cost, and accessibility. Wearable devices offer a potential solution, but their high cost and limited accessibility restrict their widespread application, particularly among disadvantaged populations. Smartphones, with their widespread adoption, provide a more scalable alternative. However, existing smartphone-based sleep monitoring approaches often rely heavily on self-reported data, which can be unreliable due to recall bias and inconsistent reporting. The lack of open-source and validated tools further hinders progress in this field. This study builds upon existing research in digital phenotyping, which aims to capture moment-by-moment behavioral and physiological data using readily available personal devices to understand and predict health outcomes.

This study utilized data collected from a larger study using the open-source mindLAMP smartphone application. Sixty-seven college students with elevated stress levels participated, providing data over 28 days. Participants completed daily and weekly surveys assessing sleep duration, sleep quality, and mental health using validated instruments such as the Pittsburgh Sleep Quality Index (PSQI), Patient Health Questionnaire-9 (PHQ-9), Generalized Anxiety Disorder-7 (GAD-7), and Perceived Stress Scale (PSS). Passively collected smartphone sensor data, including accelerometer and device usage data, were used to estimate sleep duration. The accelerometer data was processed using Otsu's method to identify thresholds distinguishing between activity and inactivity. Periods of missing data were handled through concatenation based on the preceding and succeeding states. A mixed linear regression model, incorporating participants as random effects, was used to analyze the relationship between passive sleep duration estimates, self-reported sleep measures, and PSQI scores. A separate linear predictive model was constructed to assess the ability of a combination of passive and active data to predict PSQI scores, validated using leave-one-out cross-validation (LOOCV).

The study found a strong correlation (r=0.83) between mean passive sleep duration estimates derived from smartphone sensor data and mean self-reported sleep duration from daily surveys, when considering only days with data from both sources. Demographic information showed the participants had a mean age of 20.0 with a standard deviation of 2.0; they were primarily female (65.7%), with a slight majority identifying as white (56.7%). In a mixed linear regression model, passive sleep duration estimates were the most significant predictor of PSQI scores, showing a negative correlation (meaning longer sleep duration estimates correlated with lower PSQI scores, indicating better sleep quality). A simple linear predictive model, incorporating both passive and active data (survey-reported sleep duration and sleep quality), predicted weekly PSQI scores with a mean absolute error of 0.93, suggesting an accuracy within one point on the 0-14 PSQI scale. Importantly, passive sleep duration showed a stronger relationship with weekly PSQI than self-reported sleep duration, highlighting the potential advantages of objective smartphone-based measurements.

The strong correlation between smartphone-derived sleep duration estimates and self-reported sleep duration, along with the accurate prediction of PSQI scores using a combination of passive and active data, strongly supports the feasibility and potential clinical utility of using smartphone data for sleep monitoring. The findings suggest that smartphone-based sleep monitoring offers several advantages over traditional methods, including increased scalability, reduced cost, and potentially improved reliability compared to self-reported data. The automated nature of passive data collection may mitigate issues with recall bias and participant engagement often seen with self-reported measures. This approach offers clinicians a practical tool for initiating conversations about sleep quality with patients. While further validation in diverse populations is needed, these findings advance the use of digital phenotyping for sleep monitoring in mental health.

This study demonstrates the potential of using passively collected smartphone sensor data to accurately estimate sleep duration and predict sleep quality using the PSQI. The strong correlation between passive estimates and self-reported sleep duration, coupled with the high predictive accuracy of the developed model, supports the feasibility and clinical utility of smartphone-based sleep monitoring. This method offers a scalable, cost-effective, and potentially more reliable alternative to traditional approaches. Future research should focus on validating this method in larger and more diverse populations, exploring its use in different clinical settings, and investigating the integration of these data into clinical decision-making processes. Further investigation into the influence of inconsistent smartphone usage on data accuracy is also warranted.

The study's findings are limited by its sample size and the specific population studied (college students). The assumption that inactivity equates to sleep might not hold universally, and inconsistent smartphone usage could affect the accuracy of sleep duration estimations. The correlation between passive sleep duration and PSQI is not a direct measure of sleep quality; it correlates with a composite measure that also includes sleep latency, sleep disturbances, and daytime dysfunction. Additionally, this study did not compare the smartphone-based sleep estimates to a gold standard measure such as polysomnography or actigraphy. Further validation in diverse populations is necessary to generalize these results.