Aspects of the present disclosure provide methods, apparatuses, and systems for accurately determining sleep and wake onset based on a user's personalized physiological parameters for sleep and wake. First, a user is determined to be asleep using population level data. Thereafter, sensor collected data is used to determine the user's distribution of values of a physiological parameter when the user is asleep. This distribution of values is then used, instead of population-level data, to determine the user is asleep in real-time. As a result, the content and interventions are provided to help users get back to sleep. Further, the described techniques allow more accuracy in determining sleep statistics which can guide recommended interventions and therapies.