Ambulatory estimation of human circadian phase using models of varying complexity based on non-invasive signal modalities.

In this work, we introduce a number of models for human circadian phase estimation in ambulatory conditions using various sensor modalities. Machine learning techniques have been applied to ambulatory recordings of wrist actigraphy, light exposure, electrocardiograms (ECG), and distal and proximal skin temperature to develop ARMAX models capturing the main signal dependencies on circadian phase and evaluating them versus melatonin onset times. The most accurate models extracted heart rate variability features from an ECG coupled with wrist activity information to produce phase estimations with prediction errors of ~30 minutes. Replacing the ECG features with skin temperature from the upper leg led to a slight degradation, while less accurate results, in the order of 1 hour, were obtained from wrist activity and light measurements. The trade-off between highest precision and least obtrusive configuration is discussed for applications to sleep and mood disorders caused by a misalignment of the internal phase with the external solar and social times.

Human circadian phase estimation from signals collected in ambulatory conditions using an autoregressive model.

Phase estimation of the human circadian rhythm is a topic that has been explored using various modeling approaches. The current models range from physiological to mathematical, all attempting to estimate the circadian phase from different physiological or behavioral signals. Here, we have focused on estimation of the circadian phase from unobtrusively collected signals in ambulatory conditions using a statistically trained autoregressive moving average with exogenous inputs (ARMAX) model. Special attention has been given to the evaluation of heart rate interbeat intervals (RR intervals) as a potential circadian phase predictor. Prediction models were trained using all possible combinations of RR intervals, activity levels, and light exposures, each collected over a period of 24 hours. The signals were measured without any behavioral constraints, aside from the collection of saliva in the evening to determine melatonin concentration, which was measured in dim-light conditions. The model was trained and evaluated using 2 completely independent datasets, with 11 and 19 participants, respectively. The output was compared to the gold standard of circadian phase: dim-light melatonin onset (DLMO). The most accurate model that we found made use of RR intervals and light and was able to yield phase estimates with a prediction error of 2 ± 39 minutes (mean ± SD) from the DLMO reference value.