Frailty detection in older adults via fractal analysis of acceleration signals from wrist-worn sensors.

Abstract

Purpose: Frailty is a reversible multidimensional syndrome that puts older people at a high risk of adverse health outcomes. It has been proposed to emerge from the dysregulation of the complex system dynamics of physiologic control systems. We propose the analysis of the fractal complexity of hand movements as a new method to detect frailty in older adults.

Methods: FRAIL scale and Fried's phenotype scores were calculated for 1209 subjects-72.4 (5.2) y.o. 569 women-and 1279 subjects-72.6 (5.3) y.o. 604 women-in the pubicly available NHANES 2011-2014 data set, respectively. The fractal complexity of their hand movements was assessed with a detrended fluctuation analysis (DFA) of their accelerometry records and a logistic regression model for frailty detection was fit.

Results: Goodness-of-fit to a power law was excellent (R${}^2>0.98$). The association between complexity loss and frailty level was significant, Kruskal-Wallis test (df = 2, Chisq = 27.545, p-value $<0.001$). The AUC of the logistic classifier was moderate (AUC with complexity = 0.69 vs. AUC without complexity = 0.67).

Conclusion: Frailty can be characterized in this data set with the Fried phenotype. Non-dominant hand movements in free-living conditions are fractal processes regardless of age or frailty level and its complexity can be quantified with the exponent of a power law. Higher levels of complexity loss are associated with higher levels of frailty. This association is not strong enough to justify the use of complexity loss after adjusting for sex, age, and multimorbidity.