Validating the accuracy of an activity monitor in a visually impaired older population

Purpose: Activity monitors have been used to objectively measure physical activity and its association with visual impairment in older adults. However, there is limited understanding of the accuracy of activity monitors in people with low vision. This study investigated the accuracy of an activity monitor compared with manual step counting in a low vision population and sought to find the most accurate placement location for the device. Methods: We recruited 32 individuals aged 50 years and older with low vision. ActiGraph activity monitors were secured bilaterally on the wrists, ankles, and hips of each participant, who then walked a flat, linear course in their home at a comfortable pace for 4 min, using any necessary assistive device such as a long cane, support cane, or guide dog. Steps were counted using a hand-held tally counter. ActiGraph-measured step data from the 4-min period were downloaded using the standard and low frequency filters at 1 epoch s ⁻¹ through ActiLife. Results: Of the 32 participants, 20 (63%) were female, median visual acuity was 1.48 logMAR (6/180 Snellen), average age was 73 (standard deviation, S.D., 9) years, average body mass index was 28.9 (S.D. 7.0) kg m ⁻² , and 47% of participants used an assistive device. Average distance for the test course was 10.9 (S.D. 3.4) m and participants completed an average of 368 (S.D. 68) steps during the 4 min. The number of steps recorded by the two, bilaterally-worn devices at each location were averaged. Ankle, hip, and wrist activity monitors detected 85% (interquartile range, IQR 76–94%), 56% (IQR 39–85%), and 56% (IQR 43–69%), respectively, of directly-observed steps when using the standard ActiGraph filter. Detected steps more closely matched directly observed steps for all placement sites when the low-frequency ActiGraph filter was applied: 101% (IQR 99–104%) at the ankle, 94% (IQR 85–101%) at the hip, and 83% (IQR 72–94%) at the wrist. Bland-Altman plots showed greater levels of agreement between ActiGraph-recorded and directly-observed steps at faster walking speeds. Conclusions: Our results demonstrate that the most accurate location of activity monitor placement is the ankle and that when using the low-frequency filter the level of agreement becomes more acceptable on the wrist and hip, in this population. Use of the low activity filter can help minimise inaccurate calculation of steps in people with low vision, particularly those who walk slowly.