Long-term mobility monitoring of older adults using accelerometers in a clinical environment

Objective: To assess the accuracy of accelerometer-based mobility monitoring during extended measurements on older adults in a clinical setting and to evaluate two different approaches to thresholding. Design: The monitoring device consisted of two Analog Devices ADXL202 accelerometers, an ambulatory data-logger and associated cabling. The monitoring system used custom-designed analysis software to detect activities of daily living, namely duration of sitting, standing, lying and moving during the period monitored. An investigator shadowed the subjects throughout the recording period. Subjects and setting: This study monitored five older adults, with varying degrees of mobility, resident in a rehabilitation clinic, over four days. Interventions: The accelerometer data were analysed using a MATLAB(R) program that allowed trunk and thigh threshold angles to be set to distinguish between sitting, standing, lying and moving. Two different approaches to setting these thresholds were investigated: (1) using a midpoint tolerance value of 458 and (2) using a 'best estimate' tolerance value. The analysis program generates a summary of activities, which is then compared line-by-line with the manual summary created by the observer. The result was a hit/miss ratio representative of the system's accuracy. Results: The detection accuracies for sitting and lying using a mid-point tolerance value were poor, with an average detection accuracy of 75% obtained. The 'best estimate' approach improved the detection accuracies for sitting and lying by approximately 18% to an average value of 93%. Conclusion: In a population of older adults, the static activities of sitting, standing and lying and dynamic activities can be distinguished using the technique and threshold values outlined here to a degree of accuracy of 92% and higher.