Harmonizing Raw Accelerometry Methodology in Research

In the rapidly evolving field of physical activity research, the use of raw accelerometry is becoming increasingly prevalent. However, the existing diversity in sensor design, attachment methods, and signal processing techniques presents significant challenges to the comparability of research results. This lack of methodological consistency can hinder valid comparisons of physical activity data across different populations, clinical settings, national samples, and time points. The inherent analytical freedom provided by raw sensor data, while fostering innovation, can paradoxically challenge methodological consistency, and raw accelerometer data may not even be 100% compatible between different brands.

To address these critical issues, a seminal article titled “Challenges and Opportunities for Harmonizing Research Methodology: Raw Accelerometry” provides a comprehensive framework and practical guidance for achieving greater methodological harmonization.

A Collaborative Approach to Standardization: This important work emerged from a two-day workshop held in March 2014, where authors representing diverse expertise – including sensor manufacturing, signal processing, study design, and various research fields – convened to discuss establishing standards for raw accelerometer data methods. The article leverages these discussions to reflect on how harmonization can be achieved and to compile actionable checklists for stakeholders.

Key Stakeholders and Action Points: The paper identifies five key stakeholders crucial for this harmonization process:

1. Manufacturers: Encouraged to provide detailed specifications of their sensors. This includes information such as sensor type, manufacturer, model, possible sampling frequencies, data resolution, dynamic range, and on-board signal processing specifications, along with test reports demonstrating output relationships with reference signals and temperature stability.
2. Method Developers: Should document and motivate each fundamental step of their algorithms for processing raw accelerometer data. They must be open about uncertainties in data description and inference, and all new algorithms deemed “ready for implementation” should be shared with the community to facilitate replication and independent evaluation.
3. Method Users (Application): Need to be transparent about uncertainties in data description and inference. They should report algorithm settings and parameters essential for understanding results and are guided by practical checklists for operational procedures and data management.
4. Publishers: Play an indirect but important role in supporting and endorsing methodological practices.
5. Funders: Also contribute indirectly by allocating resources for method development and evaluation.

The article stresses the importance of dynamic interaction among all method stakeholders to foster a well-informed harmonization process.

Practical Guidance Across the Data Pipeline: The review offers practical checklists and detailed recommendations across the entire raw accelerometry data collection and processing pipeline:

• Hardware: Manufacturers are urged to provide comprehensive sensor specifications, acknowledging the need for balance between accuracy, size, battery life, and operational flexibility. The scientific community, in turn, should inform manufacturers about desired resolution and frequency response.
• Accelerometer Configuration and Study Protocol: Decisions on sample frequency, active channels, anatomical attachment, and measurement duration should be motivated and justified in publications. Minimum standards for operational procedures, such as recording body site, attachment procedure, wear instructions, and participant characteristics, are proposed.
• Data Processing and Inference Methods: The paper advocates for detailed reporting of data management and quality control procedures, including non-wear detection and calibration checks. It emphasizes that every computational step of an algorithm should be motivated and that full algorithms (including optimized parameters) should be published for “ready for application” tools.
• Data Description Before and After Processing: Harmonization here focuses not on standardizing data descriptions, but on completeness of information and explicit expression of known unknowns. A distinction is made between data description before algorithm training, during training (labelling), and for algorithm output, where the vagueness and uncertainty of derived activity types should be reflected.
• Method Evaluation and Ongoing Development: The article highlights the value of combining diverse study designs (robot, laboratory, free-living experiments) for comprehensive method evaluation. It encourages collaboration between research groups and the replication of algorithm performance assessment techniques to ease comparisons of results. Reporting of results from both new and established methods is also recommended to enable comparisons of temporal and spatial differences in physical activity levels.

Conclusion: This pivotal article serves as a crucial guide for researchers and industry alike, demonstrating that methodological harmonization in raw accelerometry is highly desirable and should be the outcome of ongoing scientific debate and the accumulation of empirical evidence. While acknowledging that some methodological inconsistency may persist due to innovation and the need for historical data comparisons, the proposed guidelines aim to discourage unnecessary inconsistencies. It is the first review of its kind to exclusively focus on harmonization across the entire information pipeline, using a multi-stakeholder panel to establish practical guidelines.

Reference for the Article:

Hees, V. T. V., Thaler-Kall, K., Wolf, K. H., Brønd, J. C., Bonomi, A., Schulze, M., … & Horsch, A. (2016). Challenges and opportunities for harmonizing research methodology: Raw accelerometry. Methods of Information in Medicine, 55(5), 482–491. doi:10.3414/ME15-05-0013