Generalizability of A Neural Network Model for Circadian Phase Prediction in Real-World Conditions

Estimating dim light melatonin onset time in children using delta changes in melatonin

We aimed to establish a method for estimating dim light melatonin onset (DLMO) using mathematical slopes calculated from melatonin concentrations at three sampling points before and after sleep in children. The saliva of 30 children (mean age ± SD: 10.2 ± 1.3 years old) was collected under dim-light conditions up to six times every hour starting at 17:30 (t17), namely, 18:30 (t18), 19:30 (t19), 20:30 (t20), 21:30 (t21), 22:30 (t22), and 23:30 (t23), in the evening, and at 6:00 (t30) the following morning. We calculated SLOPE on (mathematical slope between melatonin concentrations at t18 and t20, t21 or t22), SLOPE off (the slope between t20, t21 or t22, and t30), and Δ S L O P E , which is generated by subtracting SLOPE on from SLOPE off . DLMO was estimated by multiple regression analysis with the leave-one-out cross-validation (LOOCV) method using SLOPE on and SLOPE off , and Δ S L O P E . The intraclass correlation coefficient (ICC) between the estimated and measured DLMOs was used as the index of estimation accuracy. DLMOs estimated using multiple regression equations with SLOPE on and SLOPE off yielded significant ICCs for the measured DLMOs, with the largest ICC at t20 (ICC = 0.634). Additionally, the ICC between the estimated and measured DLMOs using the equation with Δ S L O P E was significant, with a larger ICC at t20 (ICC = 0.726) than that of the equation with SLOPE on and SLOPE off . Our results showed that DLMO could be estimated with a certain level of accuracy from salivary melatonin levels at three time points before and after sleep in children.

Supplementary Information

The online version contains supplementary material available at 10.1007/s41105-023-00493-x.

CARE as a wearable derived feature linking circadian amplitude to human cognitive functions

Circadian rhythms are crucial for regulating physiological and behavioral processes. Pineal hormone melatonin is often used to measure circadian amplitude but its collection is costly and time-consuming. Wearable activity data are promising alternative, but the most commonly used measure, relative amplitude, is subject to behavioral masking. In this study, we firstly derive a feature named circadian activity rhythm energy (CARE) to better characterize circadian amplitude and validate CARE by correlating it with melatonin amplitude (Pearson's r = 0.46, P = 0.007) among 33 healthy participants. Then we investigate its association with cognitive functions in an adolescent dataset (Chinese SCHEDULE-A, n = 1703) and an adult dataset (UK Biobank, n = 92,202), and find that CARE is significantly associated with Global Executive Composite (β = 30.86, P = 0.016) in adolescents, and reasoning ability, short-term memory, and prospective memory (OR = 0.01, 3.42, and 11.47 respectively, all P < 0.001) in adults. Finally, we identify one genetic locus with 126 CARE-associated SNPs using the genome-wide association study, of which 109 variants are used as instrumental variables in the Mendelian Randomization analysis, and the results show a significant causal effect of CARE on reasoning ability, short-term memory, and prospective memory (β = -59.91, 7.94, and 16.85 respectively, all P < 0.0001). The present study suggests that CARE is an effective wearable-based metric of circadian amplitude with a strong genetic basis and clinical significance, and its adoption can facilitate future circadian studies and potential intervention strategies to improve circadian rhythms and cognitive functions.

Later energy intake relative to mathematically modeled circadian time is associated with higher percentage body fat

OBJECTIVE

Later circadian timing of energy intake is associated with higher body fat percentage. Current methods for obtaining accurate circadian timing are labor- and cost-intensive, limiting practical application of this relationship. This study investigated whether the timing of energy intake relative to a mathematically modeled circadian time, derived from easily collected ambulatory data, would differ between participants with a lean or overweight/obesity body fat percentage.

METHODS

Participants (N = 87) wore a light- and activity-measuring device (actigraph) throughout a cross-sectional 30-day study. For 7 consecutive days within these 30 days, participants used a time-stamped-picture phone application to record energy intake. Body fat percentage was recorded. Circadian time was defined using melatonin onset from in-laboratory collected repeat saliva sampling or using light and activity or activity data alone entered into a mathematical model.

RESULTS

Participants with overweight/obesity body fat percentages ate 50% of their daily calories significantly closer to model-predicted melatonin onset from light and activity data (0.61 hours closer) or activity data alone (0.86 hours closer; both log-rank p < 0.05).

CONCLUSIONS

Use of mathematically modeled circadian timing resulted in similar relationships between the timing of energy intake and body composition as that observed using in-laboratory collected metrics. These findings may facilitate use of circadian timing in time-based interventions.

The circadian stimulus-oscillator model: Improvements to Kronauer’s model of the human circadian pacemaker

Modeling how patterns of light and dark affect circadian phase is important clinically and organizationally (e.g., the military) because circadian disruption can compromise health and performance. Limit-cycle oscillator models in various forms have been used to characterize phase changes to a limited set of light interventions. We approached the analysis of the van der Pol oscillator-based model proposed by Kronauer and colleagues in 1999 and 2000 (Kronauer99) using a well-established framework from experimental psychology whereby the stimulus (S) acts on the organism (O) to produce a response (R). Within that framework, using four independent data sets utilizing calibrated personal light measurements, we conducted a serial analysis of the factors in the Kronauer99 model that could affect prediction accuracy characterized by changes in dim-light melatonin onset. Prediction uncertainty was slightly greater than 1 h for the new data sets using the original Kronauer99 model. The revised model described here reduced prediction uncertainty for these same data sets by roughly half.

Distinct Circadian Assessments From Wearable Data Reveal Social Distancing Promoted Internal Desynchrony Between Circadian Markers

Mobile measures of human circadian rhythms (CR) are needed in the age of chronotherapy. Two wearable measures of CR have recently been validated: one that uses heart rate to extract circadian rhythms that originate in the sinoatrial node of the heart, and another that uses activity to predict the laboratory gold standard and central circadian pacemaker marker, dim light melatonin onset (DLMO). We first find that the heart rate markers of normal real-world individuals align with laboratory DLMO measurements when we account for heart rate phase error. Next, we expand upon previous work that has examined sleep patterns or chronotypes during the COVID-19 lockdown by studying the effects of social distancing on circadian rhythms. In particular, using data collected from the Social Rhythms app, a mobile application where individuals upload their wearable data and receive reports on their circadian rhythms, we compared the two circadian phase estimates before and after social distancing. Interestingly, we found that the lockdown had different effects on the two ambulatory measurements. Before the lockdown, the two measures aligned, as predicted by laboratory data. After the lockdown, when circadian timekeeping signals were blunted, these measures diverged in 70% of subjects (with circadian rhythms in heart rate, or CRHR, becoming delayed). Thus, while either approach can measure circadian rhythms, both are needed to understand internal desynchrony. We also argue that interventions may be needed in future lockdowns to better align separate circadian rhythms in the body.

Prediction of shiftworker alertness, sleep, and circadian phase using a model of arousal dynamics constrained by shift schedules and light exposure

STUDY OBJECTIVES

The study aimed to, for the first time, (1) compare sleep, circadian phase, and alertness of intensive care unit (ICU) nurses working rotating shifts with those predicted by a model of arousal dynamics; and (2) investigate how different environmental constraints affect predictions and agreement with data.

METHODS

The model was used to simulate individual sleep-wake cycles, urinary 6-sulphatoxymelatonin (aMT6s) profiles, subjective sleepiness on the Karolinska Sleepiness Scale (KSS), and performance on a Psychomotor Vigilance Task (PVT) of 21 ICU nurses working day, evening, and night shifts. Combinations of individual shift schedules, forced wake time before/after work and lighting, were used as inputs to the model. Predictions were compared to empirical data. Simulations with self-reported sleep as an input were performed for comparison.

RESULTS

All input constraints produced similar prediction for KSS, with 56%-60% of KSS scores predicted within ±1 on a day and 48%-52% on a night shift. Accurate prediction of an individual's circadian phase required individualized light input. Combinations including light information predicted aMT6s acrophase within ±1 h of the study data for 65% and 35%-47% of nurses on diurnal and nocturnal schedules. Minute-by-minute sleep-wake state overlap between the model and the data was between 81 ± 6% and 87 ± 5% depending on choice of input constraint.

CONCLUSIONS

The use of individualized environmental constraints in the model of arousal dynamics allowed for accurate prediction of alertness, circadian phase, and sleep for more than half of the nurses. Individual differences in physiological parameters will need to be accounted for in the future to further improve predictions.

A classification approach to estimating human circadian phase under circadian alignment from actigraphy and photometry data

The time of dim light melatonin onset (DLMO) is the gold standard for circadian phase assessment in humans, but collection of samples for DLMO is time and resource-intensive. Numerous studies have attempted to estimate circadian phase from actigraphy data, but most of these studies have involved individuals on controlled and stable sleep-wake schedules, with mean errors reported between 0.5 and 1 hour. We found that such algorithms are less successful in estimating DLMO in a population of college students with more irregular schedules: Mean errors in estimating the time of DLMO are approximately 1.5-1.6 hours. We reframed the problem as a classification problem and estimated whether an individual's current phase was before or after DLMO. Using a neural network, we found high classification accuracy of about 90%, which decreased the mean error in DLMO estimation-identifying the time at which the switch in classification occurs-to approximately 1.3 hours. To test whether this classification approach was valid when activity and circadian rhythms are decoupled, we applied the same neural network to data from inpatient forced desynchrony studies in which participants are scheduled to sleep and wake at all circadian phases (rather than their habitual schedules). In participants on forced desynchrony protocols, overall classification accuracy dropped to 55%-65% with a range of 20%-80% for a given day; this accuracy was highly dependent upon the phase angle (ie, time) between DLMO and sleep onset, with the highest accuracy at phase angles associated with nighttime sleep. Circadian patterns in activity, therefore, should be included when developing and testing actigraphy-based approaches to circadian phase estimation. Our novel algorithm may be a promising approach for estimating the onset of melatonin in some conditions and could be generalized to other hormones.

Identification of a Preliminary Plasma Metabolome-based Biomarker for Circadian Phase in Humans

Measuring individual circadian phase is important to diagnose and treat circadian rhythm sleep-wake disorders and circadian misalignment, inform chronotherapy, and advance circadian science. Initial findings using blood transcriptomics to predict the circadian phase marker dim-light melatonin onset (DLMO) show promise. Alternatively, there are limited attempts using metabolomics to predict DLMO and no known omics-based biomarkers predict dim-light melatonin offset (DLMOff). We analyzed the human plasma metabolome during adequate and insufficient sleep to predict DLMO and DLMOff using one blood sample. Sixteen (8 male/8 female) healthy participants aged 22.4 ± 4.8 years (mean ± SD) completed an in-laboratory study with 3 baseline days (9 h sleep opportunity/night), followed by a randomized cross-over protocol with 9-h adequate sleep and 5-h insufficient sleep conditions, each lasting 5 days. Blood was collected hourly during the final 24 h of each condition to independently determine DLMO and DLMOff. Blood samples collected every 4 h were analyzed by untargeted metabolomics and were randomly split into training (68%) and test (32%) sets for biomarker analyses. DLMO and DLMOff biomarker models were developed using partial least squares regression in the training set followed by performance assessments using the test set. At baseline, the DLMOff model showed the highest performance (0.91 R2 and 1.1 ± 1.1 h median absolute error ± interquartile range [MdAE ± IQR]), with significantly (p < 0.01) lower prediction error versus the DLMO model. When all conditions (baseline, 9 h, and 5 h) were included in performance analyses, the DLMO (0.60 R2; 2.2 ± 2.8 h MdAE; 44% of the samples with an error under 2 h) and DLMOff (0.62 R2; 1.8 ± 2.6 h MdAE; 51% of the samples with an error under 2 h) models were not statistically different. These findings show promise for metabolomics-based biomarkers of circadian phase and highlight the need to test biomarkers that predict multiple circadian phase markers under different physiological conditions.

Workshop report. Circadian rhythm sleep-wake disorders: gaps and opportunities

This White Paper presents the results from a workshop cosponsored by the Sleep Research Society (SRS) and the Society for Research on Biological Rhythms (SRBR) whose goals were to bring together sleep clinicians and sleep and circadian rhythm researchers to identify existing gaps in diagnosis and treatment and areas of high-priority research in circadian rhythm sleep-wake disorders (CRSWD). CRSWD are a distinct class of sleep disorders caused by alterations of the circadian time-keeping system, its entrainment mechanisms, or a misalignment of the endogenous circadian rhythm and the external environment. In these disorders, the timing of the primary sleep episode is either earlier or later than desired, irregular from day-to-day, and/or sleep occurs at the wrong circadian time. While there are incomplete and insufficient prevalence data, CRSWD likely affect at least 800,000 and perhaps as many as 3 million individuals in the United States, and if Shift Work Disorder and Jet Lag are included, then many millions more are impacted. The SRS Advocacy Taskforce has identified CRSWD as a class of sleep disorders for which additional high-quality research could have a significant impact to improve patient care. Participants were selected for their expertise and were assigned to one of three working groups: Phase Disorders, Entrainment Disorders, and Other. Each working group presented a summary of the current state of the science for their specific CRSWD area, followed by discussion from all participants. The outcome of those presentations and discussions are presented here.

Circadian Phase Prediction From Non-Intrusive and Ambulatory Physiological Data

Chronotherapy aims to treat patients according to their endogenous biological rhythms and requires, therefore, knowing their circadian phase. Circadian phase is partially determined by genetics and, under natural conditions, is normally entrained by environmental signals (zeitgebers), predominantly by light. Physiological data such as melatonin concentration and core body temperature (CBT) have been used to estimate circadian phase. However, due to their expensive and intrusive obtention, other physiological variables that also present circadian rhythmicity, such as heart rate variability, skin temperature, activity, and body position, have recently been proposed in several studies to estimate circadian phase. This study aims to predict circadian phase using minimally intrusive ambulatory physiological data modeled with machine learning techniques. Two approaches were considered; first, time-series were used to train artificial neural networks (ANNs) that predict CBT and melatonin dynamics and, second, a novel approach that uses scalar variables to build regression models that predict the time of the minimum CBT and the dim light melatonin onset (DLMO). ANNs require less than 48 hours of minimally intrusive data collection to predict circadian phase with an accuracy of less than one hour. On the other hand, regression models that use only three variables (body mass index, activity, and heart rate) are simpler and show higher accuracy with less than one minute of error, although they require longer times of data collection. This is a promising approach that should be validated in further studies considering a broader population and a wider range of conditions, including circadian misalignment.

Fatigue risk management based on self-reported fatigue: Expanding a biomathematical model of fatigue-related performance deficits to also predict subjective sleepiness

Biomathematical models of fatigue can be used to predict neurobehavioral deficits during sleep/wake or work/rest schedules. Current models make predictions for objective performance deficits and/or subjective sleepiness, but known differences in the temporal dynamics of objective versus subjective outcomes have not been addressed. We expanded a biomathematical model of fatigue previously developed to predict objective performance deficits as measured on the Psychomotor Vigilance Test (PVT) to also predict subjective sleepiness as self-reported on the Karolinska Sleepiness Scale (KSS). Four model parameters were re-estimated to capture the distinct dynamics of the KSS and account for the scale difference between KSS and PVT. Two separate ensembles of datasets - drawn from laboratory studies of sleep deprivation, sleep restriction, simulated night work, napping, and recovery sleep - were used for calibration and subsequent validation of the model for subjective sleepiness. The expanded model was found to exhibit high prediction accuracy for subjective sleepiness, while retaining high prediction accuracy for objective performance deficits. Application of the validated model to an example scenario based on cargo aviation operations revealed divergence between predictions for objective and subjective outcomes, with subjective sleepiness substantially underestimating accumulating objective impairment, which has important real-world implications. In safety-sensitive operations such as commercial aviation, where self-ratings of sleepiness are used as part of fatigue risk management, the systematic differences in the temporal dynamics of objective versus subjective measures of functional impairment point to a potentially significant risk evaluation sensitivity gap. The expanded biomathematical model of fatigue presented here provides a useful quantitative tool to bridge this previously unrecognized gap.

Role of AI physical education based on application of functional sports training

The physical health test of college students is an important part of the school physical education work and an important part of the school education evaluation system. It is an educational method that promotes the healthy development of students’ physical fitness and encourages students to actively take physical exercises. It is an individual evaluation standard for students’ physical fitness. It is also one of the necessary conditions for students to graduate. In order to improve the physique and health of college students, this article first introduces functional exercise tests to comprehensively measure the exercise capacity of the main muscle groups and joints of the human body, and integrate flexibility and strength qualities. Secondly, this article quantitatively studies the interaction law between the natural light environment comfort of sports training facilities and architectural design elements, and adopts appropriate dynamic optimization methods to improve the light environment quality of the sports space, thereby enhancing the visual comfort of the sports crowd in the stadium. Finally, the artificial intelligence technology is introduced, through the design of artificial intelligence system, intelligent data collection, and analysis. From the perspective of physical education, the functional exercise test based on artificial intelligence conforms to the essential meaning of the physical fitness test and helps to enhance the awareness of college students’ physical exercise. And the intelligent remote multimedia physical education system based on artificial intelligence makes the physical education process flexible, free from time and place restrictions, and can adopt different teaching strategies according to the different situations of students to implement personalized teaching.

An Optimal Time for Treatment-Predicting Circadian Time by Machine Learning and Mathematical Modelling

Tailoring medical interventions to a particular patient and pathology has been termed personalized medicine. The outcome of cancer treatments is improved when the intervention is timed in accordance with the patient's internal time. Yet, one challenge of personalized medicine is how to consider the biological time of the patient. Prerequisite for this so-called chronotherapy is an accurate characterization of the internal circadian time of the patient. As an alternative to time-consuming measurements in a sleep-laboratory, recent studies in chronobiology predict circadian time by applying machine learning approaches and mathematical modelling to easier accessible observables such as gene expression. Embedding these results into the mathematical dynamics between clock and cancer in mammals, we review the precision of predictions and the potential usage with respect to cancer treatment and discuss whether the patient's internal time and circadian observables, may provide an additional indication for individualized treatment timing. Besides the health improvement, timing treatment may imply financial advantages, by ameliorating side effects of treatments, thus reducing costs. Summarizing the advances of recent years, this review brings together the current clinical standard for measuring biological time, the general assessment of circadian rhythmicity, the usage of rhythmic variables to predict biological time and models of circadian rhythmicity.

The Role of Light Sensitivity and Intrinsic Circadian Period in Predicting Individual Circadian Timing

There is large interindividual variability in circadian timing, which is underestimated by mathematical models of the circadian clock. Interindividual differences in timing have traditionally been modeled by changing the intrinsic circadian period, but recent findings reveal an additional potential source of variability: large interindividual differences in light sensitivity. Using an established model of the human circadian clock with real-world light recordings, we investigated whether changes in light sensitivity parameters or intrinsic circadian period could capture variability in circadian timing between and within individuals. Healthy participants (n = 12, aged 18-26 years) underwent continuous light monitoring for 3 weeks (Actiwatch Spectrum). Salivary dim-light melatonin onset (DLMO) was measured each week. Using the recorded light patterns, a sensitivity analysis for predicted DLMO times was performed, varying 3 model parameters within physiological ranges: (1) a parameter determining the steepness of the dose-response curve to light (p), (2) a parameter determining the shape of the phase-response curve to light (K), and (3) the intrinsic circadian period (tau). These parameters were then fitted to obtain optimal predictions of the three DLMO times for each individual. The sensitivity analysis showed that the range of variation in the average predicted DLMO times across participants was 0.65 h for p, 4.28 h for K, and 3.26 h for tau. The default model predicted the DLMO times with a mean absolute error of 1.02 h, whereas fitting all 3 parameters reduced the mean absolute error to 0.28 h. Fitting the parameters independently, we found mean absolute errors of 0.83 h for p, 0.53 h for K, and 0.42 h for tau. Fitting p and K together reduced the mean absolute error to 0.44 h. Light sensitivity parameters captured similar variability in phase compared with intrinsic circadian period, indicating they are viable targets for individualizing circadian phase predictions. Future prospective work is needed that uses measures of light sensitivity to validate this approach.

Integrating wearable data into circadian models

The emergence of wearable health sensors in the last decade has the potential to revolutionize the study of sleep and circadian rhythms. In particular, recent progress has been made in the use of mathematical models in the prediction of a patient's internal circadian state using data measured by wearable devices. This is a vital step in our ability to identify optimal circadian timing for health interventions. We review the available data for fitting circadian phase models with a focus on wearable data sets. Finally, we review the current modeling paradigms and explore avenues for developing personalized parameter sets in limit cycle oscillator models in order to further improve prediction accuracy.

Novel Approaches for Assessing Circadian Rhythmicity in Humans: A Review

The temporal organization of molecular and physiological processes is driven by environmental and behavioral cycles as well as by self-sustained molecular circadian oscillators. Quantification of phase, amplitude, period, and disruption of circadian oscillators is essential for understanding their contribution to sleep-wake disorders, social jet lag, interindividual differences in entrainment, and the development of chrono-therapeutics. Traditionally, assessment of the human circadian system, and the output of the SCN in particular, has required collection of long time series of univariate markers such as melatonin or core body temperature. Data were collected in specialized laboratory protocols designed to control for environmental and behavioral influences on rhythmicity. These protocols are time-consuming, expensive, and not practical for assessing circadian status in patients or in participants in epidemiologic studies. Novel approaches for assessment of circadian parameters of the SCN or peripheral oscillators have been developed. They are based on machine learning or mathematical model-informed analyses of features extracted from 1 or a few samples of high-dimensional data, such as transcriptomes, metabolomes, long-term simultaneous recording of activity, light exposure, skin temperature, and heart rate or in vitro approaches. Here, we review whether these approaches successfully quantify parameters of central and peripheral circadian oscillators as indexed by gold standard markers. Although several approaches perform well under entrained conditions when sleep occurs at night, the methods either perform worse in other conditions such as shift work or they have not been assessed under any conditions other than entrainment and thus we do not yet know how robust they are. Novel approaches for the assessment of circadian parameters hold promise for circadian medicine, chrono-therapeutics, and chrono-epidemiology. There remains a need to validate these approaches against gold standard markers, in individuals of all sexes and ages, in patient populations, and, in particular, under conditions in which behavioral cycles are displaced.