A comparison of cross-sectional and prospective algorithms for falls risk assessment

Falls are the most common cause of injury and hospitalization and one of the principal causes of death and disability in older adults worldwide. Accurate identification of patients at risk of falls could lead to timely medical intervention, reducing the incidence of falls related injuries along with associated costs. The current best practice for studies of falls and falls risk recommends the use of prospective follow-up data. However, the majority of studies reporting sensor based methods for assessment of falls risk employ cross-sectional falls data (falls history). The purpose of this study was to compare the performance of sensor based falls risk assessment algorithms derived from cross-sectional (N=909) and prospective (N=259) datasets in terms of false positive rate. The utility of any classification algorithm is clearly limited by a high false positive rate. An estimate of the false positive rate for both cross-sectional and prospective algorithms was determined using an inertial sensor data set of 611 TUG tests from 55 healthy control subjects, with no history of falls. We aimed to determine which falls risk assessment algorithm is more effective at classifying falls risk in healthy control subjects. The cross-sectional algorithm correctly classified 94.11% of tests, while the prospective algorithm, correctly classified 79.38% of tests. Results suggest that sensor based falls risk assessment algorithms generated using cross-sectional falls data, may be more effective than those generated using prospective data in classifying healthy controls and reducing associated false positives.