Search

Wearable data has never been as affordable and easy to access as it is right now. With the increasing popularity of consumer wearables, like Fitbit and Apple Watch, as well as the growing range of more specialized sensors, such as skin patches and glucose-tracking contacts, collection of longitudinal biological data is now more doable than ever. At the same time, mechanistic, physics-based models are finding new synergy with machine learning techniques, leveraging known physiology to compensate for sparse data. Yet the limited amount of most health-related data remains an enduring problem for any kind of model building. In this review, I outline how these two frontiers of research can come together, with wearables providing massive amounts of data collected from a single individual, and combined mechanistic modeling and machine learning approaches translating this data into better models and incentives for users to share their data.

Summary: Millions of wearable-device users record their heart rate (HR) and activity. We introduce a statistical method to extract and track six key physiological parameters from these data, including an underlying circadian rhythm in HR (CRHR), the direct effects of activity, and the effects of meals, posture, and stress through hormones like cortisol. We test our method on over 130,000 days of real-world data from medical interns on rotating shifts, showing that CRHR dynamics are distinct from those of sleep-wake or physical activity patterns and vary greatly among individuals. Our method also estimates a personalized phase-response curve of CRHR to activity for each individual, representing a passive and personalized determination of how human circadian timekeeping continually changes due to real-world stimuli. We implement our method in the “Social Rhythms” iPhone and Android app, which anonymously collects data from wearable-device users and provides analysis based on our method. Motivation: The exploding popularity of wearable devices, now a multi-billion dollar industry, provides a new opportunity for real-world data collection. Here, we propose a statistical method for analysis of ambulatory wearable-device data that can estimate circadian rhythms. Accounting for circadian rhythms in HR will allow more accurate measurement of other physiological parameters, e.g., basal HR, how activity increases HR, and changes in HR due to infection.

Study ObjectivesA critical barrier to successful treatment of circadian misalignment in shift workers is determining circadian phase in a clinical or field setting. Light and movement data collected passively from wrist actigraphy can generate predictions of circadian phase via mathematical models; however, these models have largely been tested in non-shift working adults. This study tested the feasibility and accuracy of actigraphy in predicting dim light melatonin onset (DLMO) in fixed night shift workers.MethodsA sample of 45 night shift workers wore wrist actigraphs before completing DLMO in the laboratory (17.0 days ± 10.3 SD). DLMO was assessed via 24 hourly saliva samples in dim light (ResultsModel predictions of DLMO showed good concordance with in-lab DLMO, with Lin’s concordance coefficient of 0.70, which was twice as high as agreement using average sleep timing as a proxy of DLMO. The absolute mean error of the predictions was 2.88 h, with 76% and 91% of the predictions falling with 2 and 4 h, respectively.ConclusionThis study is the first to demonstrate the use of wrist actigraphy-based estimates of circadian phase as a clinically useful and valid alternative to in-lab measurement of DLMO in fixed night shift workers. Future research should explore how additional predictors may impact accuracy.

Real world passive measures of circadian timekeeping have the potential to improve the efficacy of medical treatments and improve human performance on a broad scale. Here, we present algorithms for extracting and quantifying the uncertainty of estimates of the underlying circadian rhythm in heart rate (CRHR) from several widely used wearables after removing the masking effects of sleep and activity on heart rate. Analysis of over 130,000 days of data from medical interns on rotating shifts shows how CRHR dynamics are distinct from those of sleep-wake timing. We determine a personalized phase response curve (PRC) of CRHR to activity for each individual, representing the first passive determination of personalized circadian parameters, and validate our methods against previous clinical data. We also validate a simplified algorithm that can be implemented on a smartphone and use this algorithm to test a hypothesis about the role of circadian timekeeping in timing sleep using data from individuals in 5 continents. These results collectively establish CRHR as a practical method to study circadian rhythms in the real world and to understand how circadian disruption affects human health and functioning. Significance Statement We show how the circadian rhythm in heart rate can be extracted from real world data collected by the most popular commercial wearables. Our methods work with many devices and are validated against clinical measurements. Studying data from a large cohort of medical interns working on shifts, we find important dynamics of this circadian rhythm that are independent of the acute effects of activity or sleep-wake timing. These techniques can also determine personalized parameters of circadian timekeeping. Our work allows for large-scale study of human circadian rhythms in the real world.

From smart work scheduling to optimal drug timing, there is enormous potential in translating circadian rhythms research results for precision medicine in the real world. However, the pursuit of such effort requires the ability to accurately estimate circadian phase outside of the laboratory. One approach is to predict circadian phase noninvasively using light and activity measurements and mathematical models of the human circadian clock. Most mathematical models take light as an input and predict the effect of light on the human circadian system. However, consumer-grade wearables that are already owned by millions of individuals record activity instead of light, which prompts an evaluation of the accuracy of predicting circadian phase using motion alone. Here, we evaluate the ability of four different models of the human circadian clock to estimate circadian phase from data acquired by wrist-worn wearable devices. Multiple datasets across populations with varying degrees of circadian disruption were used for generalizability. Though the models we test yield similar predictions, analysis of data from 27 shift workers with high levels of circadian disruption shows that activity, which is recorded in almost every wearable device, is better at predicting circadian phase than measured light levels from wrist-worn devices when processed by mathematical models. In those living under normal living conditions, circadian phase can typically be predicted to within 1 h, even with data from a widely available commercial device (the Apple Watch). These results show that circadian phase can be predicted using existing data passively collected by millions of individuals with comparable accuracy to much more invasive and expensive methods.

Wearable, multisensor, consumer devices that estimate sleep are now commonplace, but the algorithms used by these devices to score sleep are not open source, and the raw sensor data is rarely accessible for external use. As a result, these devices are limited in their usefulness for clinical and research applications, despite holding much promise. We used a mobile application of our own creation to collect raw acceleration data and heart rate from the Apple Watch worn by participants undergoing polysomnography, as well as during the ambulatory period preceding in lab testing. Using this data, we compared the contributions of multiple features (motion, local standard deviation in heart rate, and “clock proxy”) to performance across several classifiers. Best performance was achieved using neural nets, though the differences across classifiers were generally small. For sleep-wake classification, our method scored 90% of epochs correctly, with 59.6% of true wake epochs (specificity) and 93% of true sleep epochs (sensitivity) scored correctly. Accuracy for differentiating wake, NREM sleep, and REM sleep was approximately 72% when all features were used. We generalized our results by testing the models trained on Apple Watch data using data from the Multi-ethnic Study of Atherosclerosis (MESA), and found that we were able to predict sleep with performance comparable to testing on our own dataset. This study demonstrates, for the first time, the ability to analyze raw acceleration and heart rate data from a ubiquitous wearable device with accepted, disclosed mathematical methods to improve accuracy of sleep and sleep stage prediction.

The retina drives various non-image-forming photoresponses, including circadian photoentrainment and pupil constriction. Previous investigators showed that in humans, photic suppression of the clock-controlled hormone melatonin is most sensitive to 460-nm blue light, with a threshold of ~12 log photons cm–2 s–1. This threshold is surprising because non-image-forming vision is mediated by intrinsically photosensitive retinal ganglion cells, which receive rod-driven synaptic input and can respond to light levels as low as ~7 log photons cm–2 s–1. Using a protocol that enhances data precision, we have found the threshold for human melatonin suppression to be ~10 log photons cm–2 s–1 at 460 nm. This finding has far-reaching implications since there is mounting evidence that nocturnal activation of the circadian system can be harmful.

Summary Daily rhythms in human physiology and behavior are driven by the interplay of circadian rhythms, environmental cycles, and social schedules. Much research has focused on the mechanism and function of circadian rhythms in constant conditions or in idealized light-dark environments. There have been comparatively few studies into how social pressures, such as work and school schedules, affect human activity rhythms day to day and season to season. To address this issue, we analyzed activity on Twitter in >1,500 US counties throughout the 2012–2013 calendar years in 15-min intervals using geographically tagged tweets representing ≈0.1% of the total population each day. We find that sustained periods of low Twitter activity are correlated with sufficient sleep as measured by conventional surveys. We show that this nighttime lull in Twitter activity is shifted to later times on weekends relative to weekdays, a phenomenon we term “Twitter social jet lag.” The magnitude of this social jet lag varies seasonally and geographically—with the West Coast experiencing less Twitter social jet lag compared to the Central and Eastern US—and is correlated with average commuting schedules and disease risk factors such as obesity. Most counties experience the largest amount of Twitter social jet lag in February and the lowest in June or July. We present evidence that these shifts in weekday activity coincide with relaxed social pressures due to local K-12 school holidays and that the direct seasonal effect of altered day length is comparatively weaker.

Using data collected through smartphones, we assess the effects of age, sex, lighting, and home country on sleep., The influence of the circadian clock on sleep scheduling has been studied extensively in the laboratory; however, the effects of society on sleep remain largely unquantified. We show how a smartphone app that we have developed, ENTRAIN, accurately collects data on sleep habits around the world. Through mathematical modeling and statistics, we find that social pressures weaken and/or conceal biological drives in the evening, leading individuals to delay their bedtime and shorten their sleep. A country’s average bedtime, but not average wake time, predicts sleep duration. We further show that mathematical models based on controlled laboratory experiments predict qualitative trends in sunrise, sunset, and light level; however, these effects are attenuated in the real world around bedtime. Additionally, we find that women schedule more sleep than men and that users reporting that they are typically exposed to outdoor light go to sleep earlier and sleep more than those reporting indoor light. Finally, we find that age is the primary determinant of sleep timing, and that age plays an important role in the variability of population-level sleep habits. This work better defines and personalizes “normal” sleep, produces hypotheses for future testing in the laboratory, and suggests important ways to counteract the global sleep crisis.

Intrinsically photosensitive retinal ganglion cells (ipRGCs) mediate both image-forming vision and non-image-forming visual responses such as pupillary constriction and circadian photoentrainment. Five types of ipRGCs, named M1–M5, have been discovered in rodents. To further investigate their photoresponse properties, we made multielectrode array spike recordings from rat ipRGCs, classified them into M1, M2/M4, and M3/M5 clusters, and measured their intrinsic, melanopsin-based responses to single and flickering light pulses. Results showed that ipRGC spiking can track flickers up to ∼0.2 Hz in frequency and that flicker intervals between 5 and 14 s evoke the most spikes. We also learned that melanopsin’s integration time is intensity and cluster dependent. Using these data, we constructed a mathematical model for each cluster’s intrinsic photoresponse. We found that the data for the M1 cluster are best fit by a model that assumes a large photoresponse, causing the cell to enter depolarization block. Our models also led us to hypothesize that the M2/M4 and M3/M5 clusters experience comparable photoexcitation but that the M3/M5 cascade decays significantly faster than the M2/M4 cascade, resulting in different response waveforms between these clusters. These mathematical models will help predict how each ipRGC cluster might respond to stimuli of any waveform and could inform the invention of lighting technologies that promote health through melanopsin stimulation.