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

Abstract

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.

Automated Gait Detection in Older Adults during Daily-Living using Self-Supervised Learning of Wrist- Worn Accelerometer Data: Development and Validation of ElderNet

“…ElderNet demonstrated superior performance compared to the two state-of-the-art models. It achieved the highest accuracy, signi cantly surpassing the OxWalk model 39 . Moreover, its F1 score was higher than both OxWalk and the U-Net 24 models.…”

“…To maintain uniformity in comparison with state-of-the-art algorithms, we standardized the acceleration data across the various cohorts by resampling to a 30 Hz resolution and dividing the signals into 10second non-overlap windows, following a methodology similar to the UK Biobank study 38,39 . We considered the window as a gait window only when half or more of it was labeled as gait.…”

Comparative Analysis of ActiGraph Step Counting Methods in Adults: A Systematic Literature Review and Meta-Analysis

“…Moreover, applying the existing ActiLife step counting algorithm (developed for a single-axis 7164 worn on the hip) to accelerometer data collected with a triaxial GT3X worn on the wrist lacks logical validity. However, recent developments in step counting algorithms explicitly developed for wrist accelerometer data, like the new ActiGraph algorithms (e.g., MAVM and UWFv1) (59), machine learning–based step detection (60), and the open-source Verisense step algorithm (61) show promise for improved step count accuracy on the wrist. Future research should seek to validate wrist-specific step algorithms and use raw accelerometer data.…”

Triaxial accelerometer-measured physical activity and functional behaviours among people with High Grade Glioma: The BrainWear Study

“…Analysis of peak characteristics such as peak height, width and distance between peaks in short segment epochs e.g., 10 seconds can be used to further characterise gait features. Gait speed has well-documented predictive value for health-related outcomes such as mortality, QOL and physical and cognitive functional decline in older people and will be an area of further work to explore the association between gait features and clinical events such as disease progression, hospitalisation and QOL measures in this dataset [ 55 , 61 , 62 ].…”