Validation of the K-Statistic for Accelerometer Data Quality

• Activity Number: 345
• Type: Contributed
• Date/Time: Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
• Sponsor: Section on Quality and Productivity
• Author(s): James Slaven*+ and Michael Andrew and John Violanti and Cecil Burchfiel and Bryan Vila
• Companies: National Institute for Occupational Safety and Health and National Institute for Occupational Safety and Health and SUNY University at Buffalo and National Institute for Occupational Safety and Health and Washington State University
• Keywords: actigraphy ; quality ; sleep ; statistics

Abstract

Accelerometer data quality can be inadequate due to data corruption or noncompliance of study protocols. We discuss a statistical test, the K-statistic, to determine the quality of accelerometer data. This method was evaluated using 105 subjects who wore accelerometers to assess sleep quality. This method uses the overall average distance between consecutive time points and the overall average mean amplitude of consecutive time points. Using leave-one-out cross-validation and calibration-testing methods of discrimination statistics, the error rates for this method were 0.0333 and 0.0167, respectively. An additional group of 37 subjects, whose data were collected at a later time, were used to validate the method. Discrimination analysis was performed using the original dataset as a calibration set and the new dataset as the test set, giving an error rate of 0.0263.