Step by Step - Quality Control in a Wearable Sensor Dataset Collected from People with Parkinson's Disease

Data quality is crucial for producing reliable re-sults from analyses and machine learning models, especially for publicly shared datasets intended for wide use. This paper describes the quality control (QC) processes implemented to create the publicly released “Wearables for gait in Parkinson's Disease and age-matched controls” dataset - a large multi-site, multi-modal dataset of wearable sensor data from individuals with and without Parkinson's Disease. Quality control efforts spanned the entire study, from initial equipment characterization and synchronization, to standardized data collection protocols across sites, to automated and manual checks on the final dataset files. Key processes included stress testing equipment synchronization, defining detailed protocols, conducting site vis-its, providing secondary reviews of manual processing steps, implementing automatic data checks, and visually inspecting data with integrated visualization tools. Out of 560 trials across 70 participants, the QC process identified 115 potential issues. Of these, we corrected 45, identified 31 as non-issues, and logged the remaining issues to provide context to dataset users. Wireless sensor modalities had the most issues related to de synchronization, connectivity, and battery life. We also provide a retrospective list of critical questions and challenges we needed to address to produce our multi-modal dataset as a resource for future researchers. Addressing these questions is crucial for ensuring our publicly shared dataset is well-considered and robust enough to support varied analyses by different users, including machine learning applications. Overall, this rigorous approach facilitated the release of a high-quality, standardized wearables dataset to accelerate Parkinson's Disease research.