Project OverviewMy PhD focuses on evaluating the validity and reliability of wrist-worn accelerometers for measuring 24-hour movement behaviours such as physical activity intensity, activity types, and sleep-wake patterns. These sensors are widely used in research, clinical trials, and public health surveillance. However, there is considerable variation in how validation studies are conducted and reported, which limits comparability and uptake in real-world applications. My research aims to improve transparency, methodological rigor, and reporting standards in this field. The PhD includes a scoping review, a metrological study on sensor signal quality, an upcoming project on behaviour classification schema, and an upcoming validation study comparing multiple wrist-worn devices against reference standards.First StudyWrist-worn accelerometers have become the preferred tool for continuous, objective, and low-burden monitoring of 24-hour movement behaviours. This project systematically mapped the landscape of criterion validation studies for single-sensor wrist-worn accelerometry. It focused on models predicting physical activity intensity, energy expenditure, activity type, and sleep–wake states. I reviewed 137 studies and 272 unique validation models, extracting data on study design, sample characteristics, data processing methods, validation metrics, and risk of bias. The findings highlighted key reporting gaps (e.g., incomplete device and sensor descriptions, unclear training-testing splits) and identified best practices. This work combined a scoping synthesis and meta-analytical summary of the performance of accelerometer-based prediction models benchmarked against gold-standard reference. This review paper also proposes a reporting framework to support transparent and reproducible validation studies in wearable sensor research.Second StudyTo support trustworthy validation studies, this project assessed the performance of six common research-grade wrist-worn accelerometers (ActiGraph LEAP, GT9X, GT3X; Axivity AX3; ActivInsights GENEActiv; SENs Motion) under controlled static and dynamic conditions benchmarked against an industry-grade NIST-traceable reference accelerometer (HBK G-Link-200). Metrics assessed were according to ISO standards and included: sensor drift, signal-to-noise ratio, accuracy, repeatability, non-linearity, and composite error. I also investigated the influence of calibration and digital filtering on the data and evaluated four processing pipelines (ENMO, MIMs-unit, ActiGraph counts, and my own) on its effect on harmonising data across devices to promote cross-study comparability. A reproducible R-based pipeline was developed to analyse the data. Results inform minimum technical requirements and signal processing procedures (e.g. filtering, calibration) needed to ensure reliable input for downstream validation or free-living studies.Third StudyThis upcoming project (Oct–Dec 2025) will be conducted as a secondment at ActivInsights Ltd., focusing on the development of a semantic schema for bout-level outputs from wrist-worn accelerometers. While wearable devices generate valuable data on physical activity, sedentary behaviour, and sleep, current outputs are often proprietary and lack interoperability across platforms. The project builds on the COELITION (COEL) framework, which offers a machine-readable specification for behavioural event data. However, COEL v1.0 does not yet support bouts—sustained periods of behaviour that are central to movement behaviour research. This project aims to address that gap by designing an ontology and behaviour classification schema to represent bout-level data in a standardised, privacy-preserving, and interoperable format for FAIR data exchange.Fourth StudyAccurate measurement of physical activity and sedentary time is essential for health research, clinical monitoring, and public policy. While wearable devices are widely used for this purpose, there is limited evidence on their validity and consistency when compared to gold-standard physiological measures. Differences in device hardware, algorithms, and data processing pipelines further limit comparability and generalisability across studies. This future study will evaluate the validity and reliability of wearable devices for capturing levels of physical activity intensity and sedentary behaviour. The project will examine how well these devices perform when benchmarked against indirect calorimetry using a combination of structured and unstructured laboratory activity protocols. |