Publication

Predicting the timing of Dim Light Melatonin Onset in Real-world Conditions using a Mathematical Model

June 1, 2017

Groups

Phillips AJ, McHill AM, Chen D, Beckett S, Barger LK, O’Brien CS, Sano A, Taylor S, Lockley SW, Czeisler CA, Klerman E.B. "Predicting the timing of Dim Light Melatonin Onset in Real-world Conditions using a Mathematical Model," Sleep2017, June 2017.

Abstract

Introduction:

Circadian rhythms modulate the timing and quality of sleep. Methods for measuring and predicting the timing of an individual’s circadian rhythms are therefore valuable to clinical assessment of sleep pathology or operational assessment of sleep scheduling. Mathematical models have been developed to predict the effects of light/dark patterns on the timing of human circadian rhythms. To date, however, these models have only been validated in clinical or inpatient laboratory settings, and typically at the group-average level rather than the individual level. Here, we compare individual-level predictions of the leading model to real-world assessments of circadian timing in a college student population.

Methods:

Light data were recorded around-the-clock from 256 college students for 1–4 weeks using wrist-worn actigraphy, which were first tested for accuracy against a calibrated light meter. Each individual also completed a single overnight inpatient laboratory visit to measure the timing of salivary dim light melatonin onset (DLMO). For each individual, light data, binned to maximum values in 1-h bins, were input to the St. Hilaire 2007 model of the human circadian pacemaker and its sensitivity to light. Model predictions of the average timing of DLMO were compared to the experimental data.

Results:

Preliminary results are presented for a subset of ~80 participants, with average DLMO of 22:40 ± 1:50. In approximately 20% of participants, the model was unable to predict circadian phase due to inability to entrain to the measured light pattern (despite a sleep/wake pattern that appeared entrained and normal DLMO time). Among the entrained cases, the model predicted DLMO to within ±1 hour of the true value in 40% of individuals and to within ±2 hours in 80% of individuals.

Conclusion:

Our findings suggest that if the goal is predicting DLMO from light data, actiwatches can be used with models to predict DLMO within ~1–2 hours in most individuals. Additional data or model refinement may be needed to improve accuracy.

Related Content