On Placement, Location and Orientation of Wrist-Worn Tri-Axial Accelerometers during Free-Living Measurements

Unleashing the full potential of digital outcome measures in clinical trials: eight questions that need attention

The use of digital health technologies to measure outcomes in clinical trials opens new opportunities as well as methodological challenges. Digital outcome measures may provide more sensitive and higher-frequency measurements but pose vital statistical challenges around how such outcomes should be defined and validated and how trials incorporating digital outcome measures should be designed and analysed. This article presents eight methodological questions, exploring issues such as the length of measurement period, choice of summary statistic and definition and handling of missing data as well as the potential for new estimands and new analyses to leverage the time series data from digital devices. The impact of key issues highlighted by the eight questions on a primary analysis of a trial are illustrated through a simulation study based on the 2019 Bellerophon INOPulse trial which had time spent in MVPA as a digital outcome measure. These eight questions present broad areas where methodological guidance is needed to enable wider uptake of digital outcome measures in trials.

Accelerometer-measured physical activity and sedentary behavior in nonagenarians: Associations with self-reported physical activity, anthropometric, sociodemographic, health and cognitive characteristics

BACKGROUND

Research on device-based physical activity in the oldest-old adults is scarce. We examined accelerometer-measured physical activity and sedentary behavior in nonagenarians. We also investigated how the accelerometer characteristics associate with nonagenarians' self-reported physical activity, anthropometric, sociodemographic, health and cognitive characteristics.

METHODS

Nonagenarians from a population-based cohort study (N = 38, mean age 91.2) used accelerometers during the waking hours for seven days. They also participated in a health survey and cognitive telephone interview. The Wald test and Pearson and polyserial correlations were used to analyze the data.

RESULTS

The participants' average day consisted of 2931 steps, 11 minutes of moderate-to-vigorous physical activity and 13.6 hours of sedentary time. Physical activity bouts less than 3 minutes per day and sedentary time bouts of 20-60 minutes per day were the most common. No sex differences were found. Many accelerometer-measured and self-reported physical activity characteristics correlated positively (correlations ≥0.34, p-values <0.05). The low levels of many accelerometer-measured physical activity characteristics associated with low education (correlations ≥0.25, p-values <0.05), dizziness (correlations ≤-0.42, p-values <0.01) and fear of falling (correlations ≤-0.45, p-values <0.01). Fear of falling was also associated with accelerometer-measured sedentary behavior characteristics (correlations -0.42 or ≥0.43).

CONCLUSIONS

Nonagenarians were mostly sedentary and low in physical activity, but individual variability existed. Accelerometer-measured and self-reported physical activity had a good consistency. Education, dizziness and fear of falling were consistently related to accelerometer-measured characteristics in nonagenarians.

Can Gait Characteristics Be Represented by Physical Activity Measured with Wrist-Worn Accelerometers?

Wearable accelerometers allow for continuous monitoring of function and behaviors in the participant's naturalistic environment. Devices are typically worn in different body locations depending on the concept of interest and endpoint under investigation. The lumbar and wrist are commonly used locations: devices placed at the lumbar region enable the derivation of spatio-temporal characteristics of gait, while wrist-worn devices provide measurements of overall physical activity (PA). Deploying multiple devices in clinical trial settings leads to higher patient burden negatively impacting compliance and data quality and increases the operational complexity of the trial. In this work, we evaluated the joint information shared by features derived from the lumbar and wrist devices to assess whether gait characteristics can be adequately represented by PA measured with wrist-worn devices. Data collected at the Pfizer Innovation Research (PfIRe) Lab were used as a real data example, which had around 7 days of continuous at-home data from wrist- and lumbar-worn devices (GENEActiv) obtained from a group of healthy participants. The relationship between wrist- and lumbar-derived features was estimated using multiple statistical methods, including penalized regression, principal component regression, partial least square regression, and joint and individual variation explained (JIVE). By considering multilevel models, both between- and within-subject effects were taken into account. This work demonstrated that selected gait features, which are typically measured with lumbar-worn devices, can be represented by PA features measured with wrist-worn devices, which provides preliminary evidence to reduce the number of devices needed in clinical trials and to increase patients' comfort. Moreover, the statistical methods used in this work provided an analytic framework to compare repeated measures collected from multiple data modalities.

Accuracy and Reliability of a Suite of Digital Measures of Walking Generated Using a Wrist-Worn Sensor in Healthy Individuals: Performance Characterization Study

BACKGROUND

Mobility is a meaningful aspect of an individual's health whose quantification can provide clinical insights. Wearable sensor technology can quantify walking behaviors (a key aspect of mobility) through continuous passive monitoring.

OBJECTIVE

Our objective was to characterize the analytical performance (accuracy and reliability) of a suite of digital measures of walking behaviors as critical aspects in the practical implementation of digital measures into clinical studies.

METHODS

We collected data from a wrist-worn device (the Verily Study Watch) worn for multiple days by a cohort of volunteer participants without a history of gait or walking impairment in a real-world setting. On the basis of step measurements computed in 10-second epochs from sensor data, we generated individual daily aggregates (participant-days) to derive a suite of measures of walking: step count, walking bout duration, number of total walking bouts, number of long walking bouts, number of short walking bouts, peak 30-minute walking cadence, and peak 30-minute walking pace. To characterize the accuracy of the measures, we examined agreement with truth labels generated by a concurrent, ankle-worn, reference device (Modus StepWatch 4) with known low error, calculating the following metrics: intraclass correlation coefficient (ICC), Pearson r coefficient, mean error, and mean absolute error. To characterize the reliability, we developed a novel approach to identify the time to reach a reliable readout (time to reliability) for each measure. This was accomplished by computing mean values over aggregation scopes ranging from 1 to 30 days and analyzing test-retest reliability based on ICCs between adjacent (nonoverlapping) time windows for each measure.

RESULTS

In the accuracy characterization, we collected data for a total of 162 participant-days from a testing cohort (n=35 participants; median observation time 5 days). Agreement with the reference device-based readouts in the testing subcohort (n=35) for the 8 measurements under evaluation, as reflected by ICCs, ranged between 0.7 and 0.9; Pearson r values were all greater than 0.75, and all reached statistical significance (P<.001). For the time-to-reliability characterization, we collected data for a total of 15,120 participant-days (overall cohort N=234; median observation time 119 days). All digital measures achieved an ICC between adjacent readouts of >0.75 by 16 days of wear time.

CONCLUSIONS

We characterized the accuracy and reliability of a suite of digital measures that provides comprehensive information about walking behaviors in real-world settings. These results, which report the level of agreement with high-accuracy reference labels and the time duration required to establish reliable measure readouts, can guide the practical implementation of these measures into clinical studies. Well-characterized tools to quantify walking behaviors in research contexts can provide valuable clinical information about general population cohorts and patients with specific conditions.

A "one-size-fits-most" walking recognition method for smartphones, smartwatches, and wearable accelerometers

The ubiquity of personal digital devices offers unprecedented opportunities to study human behavior. Current state-of-the-art methods quantify physical activity using "activity counts," a measure which overlooks specific types of physical activities. We propose a walking recognition method for sub-second tri-axial accelerometer data, in which activity classification is based on the inherent features of walking: intensity, periodicity, and duration. We validate our method against 20 publicly available, annotated datasets on walking activity data collected at various body locations (thigh, waist, chest, arm, wrist). We demonstrate that our method can estimate walking periods with high sensitivity and specificity: average sensitivity ranged between 0.92 and 0.97 across various body locations, and average specificity for common daily activities was typically above 0.95. We also assess the method's algorithmic fairness to demographic and anthropometric variables and measurement contexts (body location, environment). Finally, we release our method as open-source software in Python and MATLAB.

Sleep Measurement Using Wrist-Worn Accelerometer Data Compared with Polysomnography

This study determined if using alternative sleep onset (SO) definitions impacted accelerometer-derived sleep estimates compared with polysomnography (PSG). Nineteen participants (48%F) completed a 48 h visit in a home simulation laboratory. Sleep characteristics were calculated from the second night by PSG and a wrist-worn ActiGraph GT3X+ (AG). Criterion sleep measures included PSG-derived Total Sleep Time (TST), Sleep Onset Latency (SOL), Wake After Sleep Onset (WASO), Sleep Efficiency (SE), and Efficiency Once Asleep (SE_ASLEEP). Analogous variables were derived from temporally aligned AG data using the Cole-Kripke algorithm. For PSG, SO was defined as the first score of 'sleep'. For AG, SO was defined three ways: 1-, 5-, and 10-consecutive minutes of 'sleep'. Agreement statistics and linear mixed effects regression models were used to analyze 'Device' and 'Sleep Onset Rule' main effects and interactions. Sleep-wake agreement and sensitivity for all AG methods were high (89.0-89.5% and 97.2%, respectively); specificity was low (23.6-25.1%). There were no significant interactions or main effects of 'Sleep Onset Rule' for any variable. The AG underestimated SOL (19.7 min) and WASO (6.5 min), and overestimated TST (26.2 min), SE (6.5%), and SE_ASLEEP (1.9%). Future research should focus on developing sleep-wake detection algorithms and incorporating biometric signals (e.g., heart rate).

Activity Recognition in Parkinson's Patients from Motion Data Using a CNN Model Trained by Healthy Subjects

Physical activity recognition in patients with Parkinson's Disease (PwPD) is challenging due to the lack of large-enough and good quality motion data for PwPD. A common approach to this obstacle involves the use of models trained on better quality data from healthy patients. Models can struggle to generalize across these domains due to motor complications affecting the movement patterns in PwPD and differences in sensor axes orientations between data. In this paper, we investigated the generalizability of a deep convolutional neural network (CNN) model trained on a young, healthy population to PD, and the role of data augmentation on alleviating sensor position variability. We used two publicly available healthy datasets - PAMAP2 and MHEALTH. Both datasets had sensor placements on the chest, wrist, and ankle with 9 and 10 subjects, respectively. A private PD dataset was utilized as well. The proposed CNN model was trained on PAMAP2 in k-fold cross-validation based on the number of subjects, with and without data augmentation, and tested directly on MHEALTH and PD data. Without data augmentation, the trained model resulted in 48.16% accuracy on MHEALTH and 0% on the PD data when directly applied with no model adaptation techniques. With data augmentation, the accuracies improved to 87.43% and 44.78%, respectively, indicating that the method compensated for the potential sensor placement variations between data. Clinical Relevance- Wearable sensors and machine learning can provide important information about the activity level of PwPD. This information can be used by the treating physician to make appropriate clinical interventions such as rehabilitation to improve quality of life.

Re-Orienting Smartphone-Collected Car Motion Data Using Least-Squares Estimation and Machine Learning

Smartphone sensors can collect data in many different contexts. They make it feasible to obtain large amounts of data at little or no cost because most people own mobile phones. In this work, we focus on collecting motion data in the car using a smartphone. Motion sensors, such as accelerometers and gyroscopes, can help obtain information about the vehicle’s dynamics. However, the different positioning of the smartphone in the car leads to difficulty interpreting the sensed data due to an unknown orientation, making the collection useless. Thus, we propose an approach to automatically re-orient smartphone data collected in the car to a standardized orientation (i.e., with zero yaw, roll, and pitch angles with respect to the vehicle). We use a combination of a least-square plane approximation and a Machine Learning model to infer the relative orientation angles. Then we populate rotation matrices and perform the data rotation. We trained the model by collecting data using a vehicle physics simulator.

Considerations to address missing data when deriving clinical trial endpoints from digital health technologies

Digital health technologies (DHTs) enable us to measure human physiology and behavior remotely, objectively and continuously. With the accelerated adoption of DHTs in clinical trials, there is an unmet need to identify statistical approaches to address missing data to ensure that the derived endpoints are valid, accurate, and reliable. It is not obvious how commonly used statistical methods to handle missing data in clinical trials can be directly applied to the complex data collected by DHTs. Meanwhile, current approaches used to address missing data from DHTs are of limited sophistication and focus on the exclusion of data where the quantity of missing data exceeds a given threshold. High-frequency time series data collected by DHTs are often summarized to derive epoch-level data, which are then processed to compute daily summary measures. In this article, we discuss characteristics of missing data collected by DHT, review emerging statistical approaches for addressing missingness in epoch-level data including within-patient imputations across common time periods, functional data analysis, and deep learning methods, as well as imputation approaches and robust modeling appropriate for handling missing data in daily summary measures. We discuss strategies for minimizing missing data by optimizing DHT deployment and by including the patients' perspective in the study design. We believe that these approaches provide more insight into preventing missing data when deriving digital endpoints. We hope this article can serve as a starting point for further discussion among clinical trial stakeholders.

A systematic review of smartphone-based human activity recognition methods for health research

Smartphones are now nearly ubiquitous; their numerous built-in sensors enable continuous measurement of activities of daily living, making them especially well-suited for health research. Researchers have proposed various human activity recognition (HAR) systems aimed at translating measurements from smartphones into various types of physical activity. In this review, we summarized the existing approaches to smartphone-based HAR. For this purpose, we systematically searched Scopus, PubMed, and Web of Science for peer-reviewed articles published up to December 2020 on the use of smartphones for HAR. We extracted information on smartphone body location, sensors, and physical activity types studied and the data transformation techniques and classification schemes used for activity recognition. Consequently, we identified 108 articles and described the various approaches used for data acquisition, data preprocessing, feature extraction, and activity classification, identifying the most common practices, and their alternatives. We conclude that smartphones are well-suited for HAR research in the health sciences. For population-level impact, future studies should focus on improving the quality of collected data, address missing data, incorporate more diverse participants and activities, relax requirements about phone placement, provide more complete documentation on study participants, and share the source code of the implemented methods and algorithms.

Estimation of free-living walking cadence from wrist-worn sensor accelerometry data and its association with SF-36 quality of life scores

Objective. We evaluate the stride segmentation performance of the Adaptive Empirical Pattern Transformation (ADEPT) for subsecond-level accelerometry data collected in the free-living environment using a wrist-worn sensor.Approach. We substantially expand the scope of the existing ADEPT pattern-matching algorithm. Methods are applied to subsecond-level accelerometry data collected continuously for 4 weeks in 45 participants, including 30 arthritis and 15 control patients. We estimate the daily walking cadence for each participant and quantify its association with SF-36 quality of life measures.Main results. We provide free, open-source software to segment individual walking strides in subsecond-level accelerometry data. Walking cadence is significantly associated with the role physical score reported via SF-36 after adjusting for age, gender, weight and height.Significance. Methods provide automatic, precise walking stride segmentation, which allows estimation of walking cadence from free-living wrist-worn accelerometry data. Results provide new evidence of associations between free-living walking parameters and health outcomes.

Generation and validation of ActiGraph GT3X+ accelerometer cut-points for assessing physical activity intensity in older adults. The OUTDOOR ACTIVE validation study

The study of physical activity in older adults is becoming more and more relevant. For evaluation of physical activity recommendations, intensity-specific accelerometer cut-points are utilized. However, research on accelerometer cut-points for older adults is still scarce. The aim of the study was to generate placement-specific cut-points of ActiGraph GT3X+ activity counts and raw measures of acceleration to determine physical activity intensity in older adults. A further aim was to compare the validity of the generated cut-points for a range of different physical activities. The study was a single experimental trial using a convenience sample. Study participants were 20 adults aged 59 to 73 years. Accelerometers were worn at six different placements (one on each wrist, one on each ankle, and two at the hip) and breath-by-breath indirect calorimetry was used as the reference for energy. The experiment comprised of two parts; a) The first required participants to walk on a treadmill at incremental speeds (3.0-5.0 km·h-1), and b) Five different everyday activities (reading, cleaning, shopping, cycling, aerobics) were staged in the laboratory setting. Accelerometer cut-points (activity counts, raw data) were derived for each of the investigated placements by linear regression using the treadmill part. Performance of the cut-points was assessed by applying the cut-points to the everyday activities. We provide cut-points for six placements and two accelerometer metrics in the specific age group. However, the derived cut-points did not outperform published ones. More research and innovative approaches are needed for improving internal and external validity of research results across populations and age groups.

Are Reallocations between Sedentary Behaviour and Physical Activity Associated with Better Sleep in Adults Aged 55+ Years? An Isotemporal Substitution Analysis

Physical activity has been proposed as an effective alternative treatment option for the increasing occurrence of sleep problems in older adults. Although higher physical activity levels are associated with better sleep, the association between specific physical activity intensities and sedentary behaviour (SB) with sleep remains unclear. This study examines the associations of statistically modelled time reallocations between sedentary time and different physical activity intensities with sleep outcomes using isotemporal substitution analysis. Device-measured physical activity data and both objective and subjective sleep data were collected from 439 adults aged 55+ years. Replacing 30 min of SB with moderate to vigorous intensity physical activity (MVPA) was significantly associated with an increased number of awakenings. Moreover, a reallocation of 30 min between light physical activity (LPA) and MVPA was significantly associated with increased sleep efficiency. Furthermore, reallocating 30 min of SB to LPA showed a significant association with decreased sleep efficiency. There were no significant associations of time reallocations for wake time after sleep onset, length of awakenings, and sleep quality. These results improve our understanding of the interrelationships between different intensities of movement behaviours and several aspects of sleep in older adults.

The Use of Inertial Measurement Units for the Study of Free Living Environment Activity Assessment: A Literature Review

This article presents an overview of fifty-eight articles dedicated to the evaluation of physical activity in free-living conditions using wearable motion sensors. This review provides a comprehensive summary of the technical aspects linked to sensors (types, number, body positions, and technical characteristics) as well as a deep discussion on the protocols implemented in free-living conditions (environment, duration, instructions, activities, and annotation). Finally, it presents a description and a comparison of the main algorithms and processing tools used for assessing physical activity from raw signals.