Unobtrusive inference of circadian rhythms from smartphone data

Circadian rhythms are an integral feature of human cognition and psychopathology but are difficult to measure outside laboratory conditions and over long timescales. Smartphones and other wearables are ubiquitous in daily life and therefore uniquely positioned to measure these rhythms non-invasively and continuously via digital phenotyping. Here, we propose a digital phenotyping framework to quantify circadian rhythms in passive data collected unobtrusively from such devices. First, we use our approach to map the sleep-wake cycle in a clinical, female outpatient sample by predicting sleep duration measured by a dedicated wearable device, the Oura Ring, from smartphone typing dynamics collected through the BiAffect platform. Next, we show that our method captures the desynchronisation between the endogenous circadian rhythm and external clock during international travel by applying it to a year's worth of typing data. Our framework provides a starting point for the analysis of circadian variations in behavioural digital phenotyping data, complementing physiological measurements from wearables.