Identification of physical activity and sedentary behaviour dimensions that predict mortality risk in older adults: Development of a machine learning model in the Whitehall II accelerometer sub-study and external validation in the CoLaus study

Diet, physical activity, and sleep in relation to postprandial glucose responses under free-living conditions: an intensive longitudinal observational study

BACKGROUND

It remains unclear what lifestyle behaviors are optimal for controlling postprandial glucose responses under real-world circumstances in persons without diabetes. We aimed to assess associations of diet, physical activity, and sleep with postprandial glucose responses in Asian adults without diabetes under free-living conditions.

METHODS

We conducted an observational study collecting intensive longitudinal data using smartphone-based ecological momentary assessments, accelerometers, and continuous glucose monitors over nine free-living days in Singaporean men and women aged 21-69 years without diabetes. The outcome was the 2-h postprandial glucose incremental area under the curve (mmol/l*min). Associations were estimated using linear mixed-effect models.

RESULTS

The analyses included 11,333 meals in 789 participants. Greater variations in glucose and lifestyle measures were observed within individuals than between individuals. Higher consumption of carbohydrate-rich and deep-fried foods and lower consumption of protein-rich foods were significantly associated with higher postprandial glucose levels (incremental area under the curve). The strongest association was observed for including refined grains (46.2 [95% CI: 40.3, 52.1]) in meals. Longer postprandial light-intensity physical activity (-24.7 [(-39.5, -9.9] per h) and moderate-to-vigorous-intensity physical activity (-58.0 [-73.8, -42.3]) were associated with substantially lower postprandial glucose levels. Longer daily light-intensity physical activity (-7.5 [-10.7, -4.2]) and sleep duration (-2.7 [-4.4, -1.0]) were also associated with lower postprandial glucose levels. Furthermore, postprandial glucose levels were the lowest in the morning and the highest in the afternoon. The results were largely consistent for males and females and for participants with and without prediabetes.

CONCLUSIONS

Consuming less refined grains and more protein-rich foods, getting more physical activity (particularly during the postprandial period), and having a longer sleep duration were associated with lower postprandial glucose levels in Asian adults without diabetes. Our findings support multi-component lifestyle modifications for postprandial glucose control and highlight the importance of the timing of eating and physical activity.

Reliable measures of rest-activity rhythm fragmentation: how many days are needed?

BACKGROUND

A more fragmented, less stable rest-activity rhythm (RAR) is emerging as a risk factor for health. Accelerometer devices are increasingly used to measure RAR fragmentation using metrics such as inter-daily stability (IS), intradaily variability (IV), transition probabilities (TP), self-similarity parameter (α), and activity balance index (ABI). These metrics were proposed in the context of long period of wear but, in real life, non-wear might introduce measurement bias. This study aims to determine the minimum number of valid days to obtain reliable fragmentation metrics.

METHODS

Wrist-worn accelerometer data were drawn from the Whitehall accelerometer sub-study (age: 60 to 83 years) to simulate different non-wear patterns. Pseudo-simulated data with different numbers of valid days (one to seven), defined as < 1/3 of non-wear during both day and night periods, and with omission or imputation of non-wear periods were compared against complete data using intraclass correlation coefficient (ICC) and mean absolute percent error (MAPE).

RESULTS

Five days with valid data (97.8% of participants) and omission of non-wear periods allowed an ICC ≥ 0.75 and MAPE ≤ 15%, acceptable cut points for reliability, for IS and ABI; this number was lower for TPs (two-three days), α and IV (four days). Overall, imputation of data did not provide better estimates. Findings were consistent across age and sex groups.

CONCLUSIONS

The number of days of wrist accelerometer data with at least 2/3 of wear time for both day and night periods varies from two (TPs) to five (IS, ABI) days for reliable RAR measures among older adults.

Behavioural digital biomarkers enable real-time monitoring of patient-reported outcomes: a substudy of the multicentre, prospective observational SafeHeart study

AIMS

Patient-reported outcome measures (PROMs) serve multiple purposes, including shared decision-making and patient communication, treatment monitoring, and health technology assessment. Patient monitoring using PROMs is constrained by recall and non-response bias, respondent burden, and missing data. We evaluated the potential of behavioural digital biomarkers obtained from a wearable accelerometer to achieve personalized predictions of PROMs.

METHODS AND RESULTS

Data from the multicentre, prospective SafeHeart study conducted at Amsterdam University Medical Center in the Netherlands and Copenhagen University Hospital, Rigshospitalet in Copenhagen, Denmark, were used. The study enrolled patients with an implantable cardioverter defibrillator between May 2021 and September 2022 who then wore wearable devices with raw acceleration output to capture digital biomarkers reflecting physical behaviour. To collect PROMs, patients received the Kansas City Cardiomyopathy Questionnaire (KCCQ) and EuroQoL 5-Dimensions 5-Level (EQ5D-5L) questionnaire at two instances: baseline and after six months. Multivariable Tobit regression models were used to explore associations between digital biomarkers and PROMs, specifically whether digital biomarkers could enable PROM prediction. The study population consisted of 303 patients (mean age 62.9 ± 10.9 years, 81.2% male). Digital biomarkers showed significant correlations to patient-reported physical and social limitations, severity and frequency of symptoms, and quality of life. Prospective validation of the Tobit models indicated moderate correlations between the observed and predicted scores for KCCQ [concordance correlation coefficient (CCC) = 0.49, mean difference: 1.07 points] and EQ5D-5L (CCC = 0.38, mean difference: 0.02 points).

CONCLUSION

Wearable digital biomarkers correlate with PROMs, and may be leveraged for real-time prediction. These findings hold promise for monitoring of PROMs through wearable accelerometers.

Data Analytics in Physical Activity Studies With Accelerometers: Scoping Review

BACKGROUND

Monitoring free-living physical activity (PA) through wearable devices enables the real-time assessment of activity features associated with health outcomes and provision of treatment recommendations and adjustments. The conclusions of studies on PA and health depend crucially on reliable statistical analyses of digital data. Data analytics, however, are challenging due to the various metrics adopted for measuring PA, different aims of studies, and complex temporal variations within variables. The application, interpretation, and appropriateness of these analytical tools have yet to be summarized.

OBJECTIVE

This research aimed to review studies that used analytical methods for analyzing PA monitored by accelerometers. Specifically, this review addressed three questions: (1) What metrics are used to describe an individual's free-living daily PA? (2) What are the current analytical tools for analyzing PA data, particularly under the aims of classification, association with health outcomes, and prediction of health events? and (3) What challenges exist in the analyses, and what recommendations for future research are suggested regarding the use of statistical methods in various research tasks?

METHODS

This scoping review was conducted following an existing framework to map research studies by exploring the information about PA. Three databases, PubMed, IEEE Xplore, and the ACM Digital Library, were searched in February 2024 to identify related publications. Eligible articles were classification, association, or prediction studies involving human PA monitored through wearable accelerometers.

RESULTS

After screening 1312 articles, 428 (32.62%) eligible studies were identified and categorized into at least 1 of the following 3 thematic categories: classification (75/428, 17.5%), association (342/428, 79.9%), and prediction (32/428, 7.5%). Most articles (414/428, 96.7%) derived PA variables from 3D acceleration, rather than 1D acceleration. All eligible articles (428/428, 100%) considered PA metrics represented in the time domain, while a small fraction (16/428, 3.7%) also considered PA metrics in the frequency domain. The number of studies evaluating the influence of PA on health conditions has increased greatly. Among the studies in our review, regression-type models were the most prevalent (373/428, 87.1%). The machine learning approach for classification research is also gaining popularity (32/75, 43%). In addition to summary statistics of PA, several recent studies used tools to incorporate PA trajectories and account for temporal patterns, including longitudinal data analysis with repeated PA measurements and functional data analysis with PA as a continuum for time-varying association (68/428, 15.9%).

CONCLUSIONS

Summary metrics can quickly provide descriptions of the strength, frequency, and duration of individuals' overall PA. When the distribution and profile of PA need to be evaluated or detected, considering PA metrics as longitudinal or functional data can provide detailed information and improve the understanding of the role PA plays in health. Depending on the research goal, appropriate analytical tools can ensure the reliability of the scientific findings.

Physical activity levels in patients with chronic venous insufficiency

BACKGROUND

Increasing the levels of physical activity (PA) is widely recommended for people with chronic venous insufficiency (CVI). However, studies investigating the patterns of PA and adherence to PA guidelines using objective measures are lacking.

OBJECTIVE

The primary aim was to examine the patterns of PA and adherence to PA guidelines among patients with CVI. A secondary aim was to identify whether adherence to PA recommendations differed according to patients' sociodemographic and clinical characteristics.

METHODS

This cross-sectional study included 96 patients with CVI with Clinical-Etiology-Anatomy-Pathology (CEAP) C3 to C6 (69.1% women 59 ± 11 years; 51.5% C5-C6 on CEAP classification). Objective time spent in PA was measured by a triaxial accelerometer. To examine adherence to PA guidelines, patients were grouped as meeting (or) the recommendations if they had at least 150 min/week of moderate to vigorous PA. Sociodemographic and clinic characteristics were obtained by self-report. Binary logistic regression was employed to examine whether sociodemographic and clinical characteristics were associated with adherence to PA guidelines. T-tests were employed to compare PA levels at different intensities according to patients' age.

RESULTS

Patients spent an average of 311.4 ± 91.5 min/week, 42.1 ± 28.0 min/week, and 19.8 ± 17.8 min/week in low-light PA, high-light PA, and moderate-to-vigorous PA, respectively. The proportion of patients meeting PA recommendations was 36.2%, and older patients had lower odds (OR = 0.94; 95%CI: 0.89 to 0.99). Additional analysis reinforced that by showing lower time in high-light PA (51.2 ± 30.0 min/day vs. 31.9 ± 21.8 min/day; p = .001) and moderate-to-vigorous PA (24.3 ± 15.8 min/day vs. 14.8 ± 18.8 min/day; p = .012) among older patients than their peers younger.

CONCLUSION

Our findings showed that 36,2% of CVI patients met PA recommendations, with lower odds found among older patients. Public health interventions to enhance PA engagement among CVI patients should prioritize those who are older.

Measures of fragmentation of rest activity patterns: mathematical properties and interpretability based on accelerometer real life data

Accelerometers, devices that measure body movements, have become valuable tools for studying the fragmentation of rest-activity patterns, a core circadian rhythm dimension, using metrics such as inter-daily stability (IS), intradaily variability (IV), transition probability (TP), and self-similarity parameter (named α ). However, their use remains mainly empirical. Therefore, we investigated the mathematical properties and interpretability of rest-activity fragmentation metrics by providing mathematical proofs for the ranges of IS and IV, proposing maximum likelihood and Bayesian estimators for TP, introducing the activity balance index (ABI) metric, a transformation of α , and describing distributions of these metrics in real-life setting. Analysis of accelerometer data from 2,859 individuals (age=60-83 years, 21.1% women) from the Whitehall II cohort (UK) shows modest correlations between the metrics, except for ABI and α . Sociodemographic (age, sex, education, employment status) and clinical (body mass index (BMI), and number of morbidities) factors were associated with these metrics, with differences observed according to metrics. For example, a difference of 5 units in BMI was associated with all metrics (differences ranging between -0.261 (95% CI -0.302, -0.220) to 0.228 (0.18, 0.268) for standardised TP rest to activity during the awake period and TP activity to rest during the awake period, respectively). These results reinforce the value of these rest-activity fragmentation metrics in epidemiological and clinical studies to examine their role for health. This paper expands on a set of methods that have previously demonstrated empirical value, improves the theoretical foundation for these methods, and evaluates their empirical use in a large dataset.

Transforming the cardiometabolic disease landscape: Multimodal AI-powered approaches in prevention and management

The rise of artificial intelligence (AI) has revolutionized various scientific fields, particularly in medicine, where it has enabled the modeling of complex relationships from massive datasets. Initially, AI algorithms focused on improved interpretation of diagnostic studies such as chest X-rays and electrocardiograms in addition to predicting patient outcomes and future disease onset. However, AI has evolved with the introduction of transformer models, allowing analysis of the diverse, multimodal data sources existing in medicine today. Multimodal AI holds great promise in more accurate disease risk assessment and stratification as well as optimizing the key driving factors in cardiometabolic disease: blood pressure, sleep, stress, glucose control, weight, nutrition, and physical activity. In this article we outline the current state of medical AI in cardiometabolic disease, highlighting the potential of multimodal AI to augment personalized prevention and treatment strategies in cardiometabolic disease.

Deep learning of movement behavior profiles and their association with markers of cardiometabolic health

BACKGROUND

Traditionally, existing studies assessing the health associations of accelerometer-measured movement behaviors have been performed with few averaged values, mainly representing the duration of physical activities and sedentary behaviors. Such averaged values cannot naturally capture the complex interplay between the duration, timing, and patterns of accumulation of movement behaviors, that altogether may be codependently related to health outcomes in adults. In this study, we introduce a novel approach to visually represent recorded movement behaviors as images using original accelerometer outputs. Subsequently, we utilize these images for cluster analysis employing deep convolutional autoencoders.

METHODS

Our method involves converting minute-by-minute accelerometer outputs (activity counts) into a 2D image format, capturing the entire spectrum of movement behaviors performed by each participant. By utilizing convolutional autoencoders, we enable the learning of these image-based representations. Subsequently, we apply the K-means algorithm to cluster these learned representations. We used data from 1812 adult (20-65 years) participants in the National Health and Nutrition Examination Survey (NHANES, 2003-2006 cycles) study who worn a hip-worn accelerometer for 7 seven consecutive days and provided valid accelerometer data.

RESULTS

Deep convolutional autoencoders were able to learn the image representation, encompassing the entire spectrum of movement behaviors. The images were encoded into 32 latent variables, and cluster analysis based on these learned representations for the movement behavior images resulted in the identification of four distinct movement behavior profiles characterized by varying levels, timing, and patterns of accumulation of movement behaviors. After adjusting for potential covariates, the movement behavior profile characterized as "Early-morning movers" and the profile characterized as "Highest activity" both had lower levels of insulin (P < 0.01 for both), triglycerides (P < 0.05 and P < 0.01, respectively), HOMA-IR (P < 0.01 for both), and plasma glucose (P < 0.05 and P < 0.1, respectively) compared to the "Lowest activity" profile. No significant differences were observed for the "Least sedentary movers" profile compared to the "Lowest activity" profile.

CONCLUSIONS

Deep learning of movement behavior profiles revealed that, in addition to duration and patterns of movement behaviors, the timing of physical activity may also be crucial for gaining additional health benefits.

Measures of fragmentation of rest activity patterns: mathematical properties and interpretability based on accelerometer real life data

Accelerometers, devices that measure body movements, have become valuable tools for studying the fragmentation of rest-activity patterns, a core circadian rhythm dimension, using metrics such as inter-daily stability (IS), intradaily variability (IV), transition probability (TP), and self-similarity parameter (named α ). However, their use remains mainly empirical. Therefore, we investigated the mathematical properties and interpretability of rest-activity fragmentation metrics by providing mathematical proofs for the ranges of IS and IV, proposing maximum likelihood and Bayesian estimators for TP, introducing the activity balance index metric, an adaptation of α , and describing distributions of these metrics in real-life setting. Analysis of accelerometer data from 2,859 individuals (age=60-83 years, 21.1% women) from the Whitehall II cohort (UK) shows modest correlations between the metrics, except for ABI and α . Sociodemographic (age, sex, education, employment status) and clinical (body mass index (BMI), and number of morbidities) factors were associated with these metrics, with differences observed according to metrics. For example, a difference of 5 units in BMI was associated with all metrics (differences ranging between -0.261 (95% CI -0.302, -0.220) to 0.228 (0.18, 0.268) for standardised TP rest to activity during the awake period and TP activity to rest during the awake period, respectively). These results reinforce the value of these rest-activity fragmentation metrics in epidemiological and clinical studies to examine their role for health. This paper expands on a set of methods that have previously demonstrated empirical value, improves the theoretical foundation for these methods, and evaluates their empirical worth in a large dataset.