Assessing elderly persons' fall risk using spectral analysis on accelerometric data--a clinical evaluation study.

Falls are among the leading causes for morbidity, mortality and lasting functional disability in the elderly population. Several studies have shown the applicability of accelerometry to detect persons with a high fall risk. Most of these studies have been conducted under laboratory settings and without clear definition of 'fall risk' reference measures. The aim of our work is to provide a simple unsupervised method to assess the fall risk of elderly persons as measured by reference clinical fall risk assessment scores. Our method uses parameters computed by spectral analysis on triaxial accelerometer data recorded in a clinical setting, and is evaluated using simple logistic regression classifier models with reference to three clinical reference scores. The overall prediction accuracy of the models ranges from 65.5-89.1%, with sensitivity and specificity between 78.5-99% and 15.4-60.4%, respectively. Our results show that our simple method can be used to detect persons with a high fall risk with a fair to good predictive accuracy when tested against common clinical reference scores. Our parameters are independent of specific test procedures and therefore are suited for use in an unsupervised setting. Our future research will include the evaluation of our method in a large prospective study.