Introduction: Physical activity (PA) and sleep monitoring provide valuable information about health, especially in the older adults. Population studies often use subjective questionnaires, which have only average accuracy and reliability. Accelerometry provides an objective measurement, but requires automated signal processing and classification to be clinically useful. Current processing measures do not take into account the heterogeneity and scale of population samples.
Methods: In this pilot study, 46 participants wore a Geneactiv™ accelerometer for 7 days. An automated signal processing framework was developed to classify the accelerometer data as sleep/activity, and quantify these periods. The algorithm was designed for scaling to large numbers. Total variance of the accelerometer data was measured, along with per-day and per-night variance, and waking and sleep period lengths. Classification was compared against a non-automated manufacturer-provided scheme (not validated in older adults). The output measures were compared using variables with established relationships to PA and sleep quality (Pearson's correlation coefficients and two-sample t-tests).
Results: Valid data were available from 36 participants. Agreement of the proposed framework with the manufacturer's method was good (95% CI: 80.25–85.92%). Correlations were found between median daily waking activity and BMI (r = −0.334, P = 0.035) and weight (r = −0.394, P = 0.018). Trends of PA against gender, age and gait speed were visible, but did not reach statistical significance (P > 0.05). Processing time was 5 min per file.
Conclusion: Use of accelerometer measurement in studies of older adults appears to be feasible and avoids subjectivity. Population studies should be cognisant of the significant analysis overhead and use validated databases when developing and validating in-house algorithms. The developed algorithm enables scaling to larger numbers for use in population studies without need for manual processing. Further validation is required before scaling to the population level.