Self-supervised learning of accelerometer data provides new insights for sleep and its association with mortality.

Sleep is essential to life. Accurate measurement and classification of sleep/wake and sleep stages is important in clinical studies for sleep disorder diagnoses and in the interpretation of data from consumer devices for monitoring physical and mental well-being. Existing non-polysomnography sleep classification techniques mainly rely on heuristic methods developed in relatively small cohorts. Thus, we aimed to establish the accuracy of wrist-worn accelerometers for sleep stage classification and subsequently describe the association between sleep duration and efficiency (proportion of total time asleep when in bed) with mortality outcomes. We developed a self-supervised deep neural network for sleep stage classification using concurrent laboratory-based polysomnography and accelerometry. After exclusion, 1448 participant nights of data were used for training. The difference between polysomnography and the model classifications on the external validation was 34.7 min (95% limits of agreement (LoA): -37.8-107.2 min) for total sleep duration, 2.6 min for REM duration (95% LoA: -68.4-73.4 min) and 32.1 min (95% LoA: -54.4-118.5 min) for NREM duration. The sleep classifier was deployed in the UK Biobank with 100,000 participants to study the association of sleep duration and sleep efficiency with all-cause mortality. Among 66,214 UK Biobank participants, 1642 mortality events were observed. Short sleepers (<6 h) had a higher risk of mortality compared to participants with normal sleep duration of 6-7.9 h, regardless of whether they had low sleep efficiency (Hazard ratios (HRs): 1.58; 95% confidence intervals (CIs): 1.19-2.11) or high sleep efficiency (HRs: 1.45; 95% CIs: 1.16-1.81). Deep-learning-based sleep classification using accelerometers has a fair to moderate agreement with polysomnography. Our findings suggest that having short overnight sleep confers mortality risk irrespective of sleep continuity.

Self-Supervised Machine Learning to Characterise Step Counts from Wrist-Worn Accelerometers in the UK Biobank.

PURPOSE: Step count is an intuitive measure of physical activity frequently quantified in health-related studies; however, accurate step counting is difficult in the free-living environment, with error routinely above 20% in wrist-worn devices against camera-annotated ground truth. This study aims to describe the development and validation of step count derived from a wrist-worn accelerometer and assess its association with cardiovascular and all-cause mortality in a large prospective cohort. METHODS: We developed and externally validated a self-supervised machine learning step detection model, trained on an open-source and step-annotated free-living dataset. 39 individuals will free-living ground-truth annotated step counts were used for model development. An open-source dataset with 30 individuals was used for external validation. Epidemiological analysis was performed using 75,263 UK Biobank participants without prevalent cardiovascular disease (CVD) or cancer. Cox regression was used to test the association of daily step count with fatal CVD and all-cause mortality after adjustment for potential confounders. RESULTS: The algorithm substantially outperformed reference models (free-living mean absolute percent error of 12.5%, versus 65-231%). Our data indicate an inverse dose-response association, where taking 6,430-8,277 daily steps was associated with 37% [25-48%] and 28% [20-35%] lower risk of fatal CVD and all-cause mortality up to seven years later, compared to those taking fewer steps each day. CONCLUSIONS: We have developed an open and transparent method that markedly improves the measurement of steps in large-scale wrist-worn accelerometer datasets. The application of this method demonstrated expected associations with CVD and all-cause mortality, indicating excellent face validity. This reinforces public health messaging for increasing physical activity and can help lay the groundwork for the inclusion of target step counts in future public health guidelines.

Self-supervised learning of accelerometer data provides new insights for sleep and its association with mortality.

Background: Sleep is essential to life. Accurate measurement and classification of sleep/wake and sleep stages is important in clinical studies for sleep disorder diagnoses and in the interpretation of data from consumer devices for monitoring physical and mental well-being. Existing non-polysomnography sleep classification techniques mainly rely on heuristic methods developed in relatively small cohorts. Thus, we aimed to establish the accuracy of wrist-worn accelerometers for sleep stage classification and subsequently describe the association between sleep duration and efficiency (proportion of total time asleep when in bed) with mortality outcomes. Methods: We developed and validated a self-supervised deep neural network for sleep stage classification using concurrent laboratory-based polysomnography and accelerometry data from three countries (Australia, the UK, and the USA). The model was validated within-cohort using subject-wise five-fold cross-validation for sleep-wake classification and in a three-class setting for sleep stage classification wake, rapid-eye-movement sleep (REM), non-rapid-eye-movement sleep (NREM) and by external validation. We assessed the face validity of our model for population inference by applying the model to the UK Biobank with 100,000 participants, each of whom wore a wristband for up to seven days. The derived sleep parameters were used in a Cox regression model to study the association of sleep duration and sleep efficiency with all-cause mortality. Findings: After exclusion, 1,448 participant nights of data were used to train the sleep classifier. The difference between polysomnography and the model classifications on the external validation was 34.7 minutes (95% limits of agreement (LoA): -37.8 to 107.2 minutes) for total sleep duration, 2.6 minutes for REM duration (95% LoA: -68.4 to 73.4 minutes) and 32.1 minutes (95% LoA: -54.4 to 118.5 minutes) for NREM duration. The derived sleep architecture estimate in the UK Biobank sample showed good face validity. Among 66,214 UK Biobank participants, 1,642 mortality events were observed. Short sleepers (<6 hours) had a higher risk of mortality compared to participants with normal sleep duration (6 to 7.9 hours), regardless of whether they had low sleep efficiency (Hazard ratios (HRs): 1.69; 95% confidence intervals (CIs): 1.28 to 2.24 ) or high sleep efficiency (HRs: 1.42; 95% CIs: 1.14 to 1.77). Interpretation: Deep-learning-based sleep classification using accelerometers has a fair to moderate agreement with polysomnography. Our findings suggest that having short overnight sleep confers mortality risk irrespective of sleep continuity.

Development and Validation of a Machine Learning Wrist-worn Step Detection Algorithm with Deployment in the UK Biobank.

Background: Step count is an intuitive measure of physical activity frequently quantified in a range of health-related studies; however, accurate quantification of step count can be difficult in the free-living environment, with step counting error routinely above 20% in both consumer and research-grade wrist-worn devices. This study aims to describe the development and validation of step count derived from a wrist-worn accelerometer and to assess its association with cardiovascular and all-cause mortality in a large prospective cohort study. Methods: We developed and externally validated a hybrid step detection model that involves self-supervised machine learning, trained on a new ground truth annotated, free-living step count dataset (OxWalk, n=39, aged 19-81) and tested against other open-source step counting algorithms. This model was applied to ascertain daily step counts from raw wrist-worn accelerometer data of 75,493 UK Biobank participants without a prior history of cardiovascular disease (CVD) or cancer. Cox regression was used to obtain hazard ratios and 95% confidence intervals for the association of daily step count with fatal CVD and all-cause mortality after adjustment for potential confounders. Findings: The novel step algorithm demonstrated a mean absolute percent error of 12.5% in free-living validation, detecting 98.7% of true steps and substantially outperforming other recent wrist-worn, open-source algorithms. Our data are indicative of an inverse dose-response association, where, for example, taking 6,596 to 8,474 steps per day was associated with a 39% [24-52%] and 27% [16-36%] lower risk of fatal CVD and all-cause mortality, respectively, compared to those taking fewer steps each day. Interpretation: An accurate measure of step count was ascertained using a machine learning pipeline that demonstrates state-of-the-art accuracy in internal and external validation. The expected associations with CVD and all-cause mortality indicate excellent face validity. This algorithm can be used widely for other studies that have utilised wrist-worn accelerometers and an open-source pipeline is provided to facilitate implementation.

Amount and intensity of physical activity and risk of incident cancer in the UK Biobank.

Importance: The influence of total daily and light intensity activity on cancer risk remains unclear, as most existing knowledge is drawn from studies relying on self-reported leisure-time activities of moderate-vigorous intensity. Objective: To investigate associations between total daily activity, including step counts, and activity intensity on incident cancer risk. Design Setting and Participants: Prospective analysis of cancer-free UK Biobank participants who wore accelerometers for 7-days (between 2013-2015), followed for cancer incidence through national registries (mean follow-up 5.8 years (SD=1.3)). Exposures: Time-series machine learning models derived daily total activity (average acceleration), behaviour time, step counts, and peak 30-minute cadence from wrist-based accelerometer data. Main Outcomes and Measures: A composite cancer outcome of 13 cancers previously associated with low physical activity (bladder, breast, colon, endometrial, oesophageal adenocarcinoma, gastric cardia, head and neck, kidney, liver, lung, myeloid leukaemia, myeloma, and rectum) based on previous studies of self-reported activity. Cox proportional hazards regression models estimated hazard ratios (HR) and 95% confidence intervals (CI), adjusted for age, sex, ethnicity, smoking, alcohol, education, Townsend Deprivation Index, and reproductive factors. Associations of reducing sedentary time in favour of increased light and moderate-vigorous activity were examined using compositional data analyses. Results: Among 86 556 participants (mean age 62.0 years (SD=7.9) at accelerometer assessment), 2 669 cancers occurred. Higher total physical activity was associated with a lower overall cancer risk (HR1SD=0.85, [95%CI 0.81-0.89]). On average, reallocating one hour/day from sedentary behaviour to moderate-vigorous physical activity was associated with a lower risk (HR=0.92, [0.89-0.95]), as was reallocating one hour/day to light-intensity physical activity (HR=0.94, [0.92-0.96]). Compared to individuals taking 5 000 daily steps, those who took 9 000 steps had an 18% lower risk of physical-activity-related cancer (HR=0.82, [0.74-0.90]). We found no significant association with peak 30-minute cadence after adjusting for total steps. Conclusion and Relevance: Higher total daily physical activity and less sedentary time, in favour of both light and moderate-vigorous intensity activity, were associated with a lower risk of certain cancers. For less active adults, increasing step counts by 4 000 daily steps may be a practical public health intervention for lowering the risk of some cancers.

Your best day: An interactive app to translate how time reallocations within a 24-hour day are associated with health measures.

Reallocations of time between daily activities such as sleep, sedentary behavior and physical activity are differentially associated with markers of physical, mental and social health. An individual's most desirable allocation of time may differ depending on which outcomes they value most, with these outcomes potentially competing with each other for reallocations. We aimed to develop an interactive app that translates how self-selected time reallocations are associated with multiple health measures. We used data from the Australian Child Health CheckPoint study (n = 1685, 48% female, 11-12 y), with time spent in daily activities derived from a validated 24-h recall instrument, %body fat from bioelectric impedance, psychosocial health from the Pediatric Quality of Life Inventory and academic performance (writing) from national standardized tests. We created a user-interface to the compositional isotemporal substitution model with interactive sliders that can be manipulated to self-select time reallocations between activities. The time-use composition was significantly associated with body fat percentage (F = 2.66, P < .001), psychosocial health (F = 4.02, P < .001), and academic performance (F = 2.76, P < .001). Dragging the sliders on the app shows how self-selected time reallocations are associated with the health measures. For example, reallocating 60 minutes from screen time to physical activity was associated with -0.8 [95% CI -1.0 to -0.5] %body fat, +1.9 [1.4 to 2.5] psychosocial score and +4.5 [1.8 to 7.2] academic performance. Our app allows the health associations of time reallocations to be compared against each other. Interactive interfaces provide flexibility in selecting which time reallocations to investigate, and may transform how research findings are disseminated.

GRANADA consensus on analytical approaches to assess associations with accelerometer-determined physical behaviours (physical activity, sedentary behaviour and sleep) in epidemiological studies.

The inter-relationship between physical activity, sedentary behaviour and sleep (collectively defined as physical behaviours) is of interest to researchers from different fields. Each of these physical behaviours has been investigated in epidemiological studies, yet their codependency and interactions need to be further explored and accounted for in data analysis. Modern accelerometers capture continuous movement through the day, which presents the challenge of how to best use the richness of these data. In recent years, analytical approaches first applied in other scientific fields have been applied to physical behaviour epidemiology (eg, isotemporal substitution models, compositional data analysis, multivariate pattern analysis, functional data analysis and machine learning). A comprehensive description, discussion, and consensus on the strengths and limitations of these analytical approaches will help researchers decide which approach to use in different situations. In this context, a scientific workshop and meeting were held in Granada to discuss: (1) analytical approaches currently used in the scientific literature on physical behaviour, highlighting strengths and limitations, providing practical recommendations on their use and including a decision tree for assisting researchers' decision-making; and (2) current gaps and future research directions around the analysis and use of accelerometer data. Advances in analytical approaches to accelerometer-determined physical behaviours in epidemiological studies are expected to influence the interpretation of current and future evidence, and ultimately impact on future physical behaviour guidelines.

Reallocation of time between device-measured movement behaviours and risk of incident cardiovascular disease.

OBJECTIVE: To improve classification of movement behaviours in free-living accelerometer data using machine-learning methods, and to investigate the association between machine-learned movement behaviours and risk of incident cardiovascular disease (CVD) in adults. METHODS: Using free-living data from 152 participants, we developed a machine-learning model to classify movement behaviours (moderate-to-vigorous physical activity behaviours (MVPA), light physical activity behaviours, sedentary behaviour, sleep) in wrist-worn accelerometer data. Participants in UK Biobank, a prospective cohort, were asked to wear an accelerometer for 7 days, and we applied our machine-learning model to classify their movement behaviours. Using compositional data analysis Cox regression, we investigated how reallocating time between movement behaviours was associated with CVD incidence. RESULTS: In leave-one-participant-out analysis, our machine-learning method classified free-living movement behaviours with mean accuracy 88% (95% CI 87% to 89%) and Cohen's kappa 0.80 (95% CI 0.79 to 0.82). Among 87 498 UK Biobank participants, there were 4105 incident CVD events. Reallocating time from any behaviour to MVPA, or reallocating time from sedentary behaviour to any behaviour, was associated with lower CVD risk. For an average individual, reallocating 20 min/day to MVPA from all other behaviours proportionally was associated with 9% (95% CI 7% to 10%) lower risk, while reallocating 1 hour/day to sedentary behaviour from all other behaviours proportionally was associated with 5% (95% CI 3% to 7%) higher risk. CONCLUSION: Machine-learning methods classified movement behaviours accurately in free-living accelerometer data. Reallocating time from other behaviours to MVPA, and from sedentary behaviour to other behaviours, was associated with lower risk of incident CVD, and should be promoted by interventions and guidelines.

Correction: Accelerometer measured physical activity and the incidence of cardiovascular disease: Evidence from the UK Biobank cohort study.

[This corrects the article DOI: 10.1371/journal.pmed.1003487.].

Accelerometer-measured physical activity and functional behaviours among people on dialysis.

BACKGROUND: The feasibility of wrist-worn accelerometers, and the patterns and determinants of physical activity, among people on dialysis are uncertain. METHODS: People on maintenance dialysis were fitted with a wrist-worn AxivityAX3 accelerometer. Subsets also wore a 14-day electrocardiograph patch (Zio®PatchXT) and wearable cameras. Age-, sex- and season-matched UK Biobank control groups were derived for comparison. RESULTS: Median (interquartile range) accelerometer wear time for the 101 recruits was 12.5 (10.4-13.5) days, of which 73 participants (mean age 66.5 years) had excellent wear on both dialysis and non-dialysis days. Mean (standard error) overall physical activity levels were 15.5 (0.7) milligravity units (mg), 14.8 (0.7) mg on dialysis days versus 16.2 (0.8) mg on non-dialysis days. This compared with 28.1 (0.5) mg for apparently healthy controls, 23.4 (0.4) mg for controls with prior cardiovascular disease (CVD) and/or diabetes mellitus and 22.9 (0.6) mg for heart failure controls. Each day, we estimated that those on dialysis spent an average of about 1 hour (h/day) walking, 0.6 h/day engaging in moderate-intensity activity, 0.7 h/day on light tasks, 13.2 h/day sedentary and 8.6 h/day asleep. Older age and self-reported leg weakness were associated with decreased levels of physical activity, but the presence of prior CVD, arrhythmias and listing for transplantation were not. CONCLUSIONS: Wrist-worn accelerometers are an acceptable and reliable method to measure physical activity in people on dialysis and may also be used to estimate functional behaviours. Among people on dialysis, who are broadly half as active as general population controls, age and leg weakness appear to be more important determinants of low activity levels than CVD.

Accelerometer measured physical activity and the incidence of cardiovascular disease: Evidence from the UK Biobank cohort study.

BACKGROUND: Higher levels of physical activity (PA) are associated with a lower risk of cardiovascular disease (CVD). However, uncertainty exists on whether the inverse relationship between PA and incidence of CVD is greater at the highest levels of PA. Past studies have mostly relied on self-reported evidence from questionnaire-based PA, which is crude and cannot capture all PA undertaken. We investigated the association between accelerometer-measured moderate, vigorous, and total PA and incident CVD. METHODS AND FINDINGS: We obtained accelerometer-measured moderate-intensity and vigorous-intensity physical activities and total volume of PA, over a 7-day period in 2013-2015, for 90,211 participants without prior or concurrent CVD in the UK Biobank cohort. Participants in the lowest category of total PA smoked more, had higher body mass index and C-reactive protein, and were diagnosed with hypertension. PA was associated with 3,617 incident CVD cases during 440,004 person-years of follow-up (median (interquartile range [IQR]): 5.2 (1.2) years) using Cox regression models. We found a linear dose-response relationship for PA, whether measured as moderate-intensity, vigorous-intensity, or as total volume, with risk of incident of CVD. Hazard ratios (HRs) and 95% confidence intervals for increasing quarters of the PA distribution relative to the lowest fourth were for moderate-intensity PA: 0.71 (0.65, 0.77), 0.59 (0.54, 0.65), and 0.46 (0.41, 0.51); for vigorous-intensity PA: 0.70 (0.64, 0.77), 0.54 (0.49,0.59), and 0.41 (0.37,0.46); and for total volume of PA: 0.73 (0.67, 0.79), 0.63 (0.57, 0.69), and 0.47 (0.43, 0.52). We took account of potential confounders but unmeasured confounding remains a possibility, and while removal of early deaths did not affect the estimated HRs, we cannot completely dismiss the likelihood that reverse causality has contributed to the findings. Another possible limitation of this work is the quantification of PA intensity-levels based on methods validated in relatively small studies. CONCLUSIONS: In this study, we found no evidence of a threshold for the inverse association between objectively measured moderate, vigorous, and total PA with CVD. Our findings suggest that PA is not only associated with lower risk for of CVD, but the greatest benefit is seen for those who are active at the highest level.

Compositional Associations of Sleep and Activities within the 24-h Cycle with Cardiometabolic Health Markers in Adults.

PURPOSE: This study aimed to examine how compositions of 24-h time use and time reallocations between movement behaviors are associated with cardiometabolic health in a population-based sample of middle-age Finnish adults. METHODS: Participants were 3443 adults 46 yr of age from the Northern Finland Birth Cohort 1966 study. Participants wore a hip-worn accelerometer for 14 d from which time spent in sedentary behavior (SB), light-intensity physical activity (LPA), and moderate- to vigorous-intensity physical activity (MVPA) were determined. These data were combined with self-reported sleep to obtain the 24-h time-use composition. Cardiometabolic outcomes included adiposity markers, blood lipid levels, and markers of glucose control and insulin sensitivity. Multivariable-adjusted regression analysis, using a compositional data analysis approach based on isometric log-ratio transformation, was used to examine associations between movement behaviors with cardiometabolic outcomes. RESULTS: More daily time in MVPA and LPA, relative to other movement behaviors, was consistently favorably associated with all cardiometabolic outcomes. For example, relative to time spent in other behaviors, 30 min·d-1 more MVPA and LPA were both associated with lower 2-h post-glucose load insulin level (-11.8% and -2.7%, respectively). Relative to other movement behaviors, more daily time in SB was adversely associated with adiposity measures, lipid levels, and markers of insulin sensitivity, and more daily time asleep was adversely associated with adiposity measures, blood lipid, fasting plasma glucose, and 2-h insulin. For example, 60 min·d-1 more SB and sleep relative to the remaining behaviors were both associated with higher 2-h insulin (3.5% and 5.7%, respectively). CONCLUSIONS: Altering daily movement behavior compositions to incorporate more MVPA at the expense of any other movement behavior, or more LPA at the expense of SB or sleep, could help to improve cardiometabolic health in midadulthood.

Choice architecture interventions to improve diet and/or dietary behaviour by healthcare staff in high-income countries: a systematic review.

OBJECTIVES: We were commissioned by the behavioural insights team at Public Health England to synthesise the evidence on choice architecture interventions to increase healthy purchasing and/or consumption of food and drink by National Health Service (NHS) staff. DATA SOURCES: MEDLINE, EMBASE, CINAHL, Cochrane Central register of Controlled Trials, PsycINFO, Applied Social Sciences Index and Abstracts and Web of Science were searched from inception until May 2017 and references were screened independently by two reviewers. DESIGN: A systematic review that included randomised experimental or intervention studies, interrupted time series and controlled before and after studies. PARTICIPANTS: Healthcare staff of high-income countries. INTERVENTION: Choice architecture interventions that aimed to improve dietary purchasing and/or consumption (outcomes) of staff. APPRAISAL AND SYNTHESIS: Eligibility assessment, quality appraisal, data abstraction and analysis were completed by two reviewers. Quality appraisal of randomised trials was informed by the Cochrane Handbook, and the Risk of Bias Assessment Tool for Nonrandomized Studies was used for the remainder. Findings were narratively synthesised. RESULTS: Eighteen studies met the inclusion criteria. Five studies included multiple workplaces (including healthcare settings), 13 were conducted in healthcare settings only. Interventions in 10 studies were choice architecture only and 8 studies involved a complex intervention with a choice architecture element. Interventions involving a proximity element (making behavioural options easier or harder to engage with) appear to be frequently effective at changing behaviour. One study presented an effective sizing intervention. Labelling alone was generally not effective at changing purchasing behaviour. Interventions including an availability element were generally reported to be successful at changing behaviour but no included study examined this element alone. There was no strong evidence for the effect of pricing on purchasing or dietary intake. CONCLUSION: Proximity, availability and sizing are choice architecture elements that are likely to be effective for NHS organisations. TRIAL REGISTRATION NUMBER: CRD42017064872.

Choice architecture modifies fruit and vegetable purchasing in a university campus grocery store: time series modelling of a natural experiment.

BACKGROUND: In developed countries, adolescent and young adult diets have been found to be nutritionally poor. The aim of this study was to examine whether a choice architecture intervention, re-arrangement of produce within a grocery store to increase the accessibility of fruit and vegetables, affected purchasing behaviour on a university campus. METHODS: A database of daily sales data from January 2012 to July 2017 was obtained from a campus grocery store. Two changes to the layout were made during this time period. In January 2015, fruit and vegetables were moved from the back of the store, furthest from the entrance, to the aisle closest to the entrance and an entrance-facing display increasing their accessibility. In April 2016, the entrance-facing display of fruit and vegetables was replaced with a chiller cabinet so that fruit and vegetables remained more accessible than during the baseline period, but less accessible than in the period immediately previously. A retrospective interrupted time series analysis using dynamic regression was used to model the data and to examine the effect of the store re-arrangements on purchasing. All analyses were carried out both for sales-by-quantity and for sales-by-money. RESULTS: The first shop re-arrangement which made fruit and vegetables more prominent, increased the percentage of total sales that were fruit and vegetables, when analysed by either items purchased or money spent. The second rearrangement also had a positive effect on the percentage of total sales that were fruit and vegetables compared to baseline, however this was not significant at the 5% level. Over the five year period, the percentage of sales that were fruit and vegetables declined both in terms of items purchased, and money spent. CONCLUSIONS: Increasing accessibility of fruit and vegetables in a grocery store is a feasible way to improve the diet of students in tertiary education. There is evidence of declining fruit and vegetable consumption among the studied population, which should be further investigated.