Data-driven modelling approach to circadian temperature rhythm profiles in free-living conditions

The individual variation in the circadian rhythms at the physiological level is not well understood. Albeit self-reported circadian preference profiles have been consolidated, their premises are grounded on human experience, not on physiology. We used data-driven, unsupervised time series modelling to characterize distinct profiles of the circadian rhythm measured from skin surface temperature in free-living conditions. We demonstrate the existence of three distinct clusters of individuals which differed in their circadian temperature profiles. The cluster with the highest temperature amplitude and the lowest midline estimating statistic of rhythm, or rhythm-adjusted mean, had the most regular and early-timed sleep-wake rhythm, and was the least probable for those with a concurrent delayed sleep phase, or eveningness chronotype. While the clusters associated with the observed sleep and circadian preference patterns, the entirely unsupervised modelling of physiological data provides a novel basis for modelling and understanding the human circadian functions in free-living conditions.