Automated and Continuous Chronotyping from a Calendar using Machine Learning

Abstract

Objectives: Chronotypes -- comparisons of individuals' circadian phase relative to others -- can contextualize mental health risk assessments, and support detection of social jet lag, which can hamper mental health and cognition. Existing ways of determining chronotypes, such as Dim Light Melatonin Onset (DLMO) or the Morningness-Eveningness Questionnaire (MEQ), are limited by being discrete in time and time-intensive to update, rarely capturing real-world variability over time. Chronotyping users based on living schedules, as in daily planner apps, might augment existing methods by assessing chronotype and social jet lag continuously and at scale. Developing this functionality would require a novel tool to translate between digital schedules and chronotypes. Here we use a supervised binary classifier to assess the feasibility of this approach. Methods: In this study, 1,460 registered users from the Owaves app opted in to filled out the MEQ survey. Of those, 142 met the eligibility criteria for data analysis. We used multimodal app data to assess the classification of individuals identified as morning and evening types from MEQ data, basing the classifier on app time series data. This includes daily timing for 8 main lifestyle activity categories (exercise, sleep, social interactions, meal times, relaxation, work, play, and miscellaneous) as defined in the app. Results: The novel chronotyping tool was able to predict the morningness and eveningness of its users with an ROC AUC of 0.70. Conclusion: Our findings support the feasibility of chronotype classification from multimodal, real-world app data. We highlight challenges to applying binary labels to complex, multimodal behaviors. Our findings suggest a potential for real-time monitoring to support future, prospective mental health research.