Title: Potential batch effects and biases in the UK Biobank accelerometer data
Authors: John Muschelli - Johns Hopkins University (United States) [presenting]
Abstract: Data from wearable technology, especially wrist-worn accelerometers is becoming more available. Over 100,000 people in a subset of the UK Biobank data set wore accelerometers for approximately 7 days in the study. We explore the potential biases in the data, including differences in the cohort with data and those without. We also discuss potential calibration and data normalization issues that are within the data that are device-dependent. We apply standard batch-effect correction methods, such as Tukeys biweight, to determine how these affect the overall results in a standard mortality analysis.