A new approach for fall risk assessment in hospitalized older adults based on gait data

Abstract

Background:Falling in older adults is one of the most common and serious problems leading to disability. Therefore, it is necessary to assess the risk of falls in older adults and take preventive measures in advance. The traditional risk assessment depends on the scale, which may be affected by the subjective factors of patients. However, in recent years, instruments have been developed to collect objective data related to gait in older adults.The aim of this study was to use objective gait data to predict fall risk in older adults. Methods:In this study, a total of 207 hospitalized older adults were recruited, and the Morse Fall Scale (MFS) and six-degrees-of-freedom (6-DOF) gait kinematic parameters of the lower limb joints were collected using a marker-based instrument. Based on the gait data, two important tasks in fall risk assessment were conducted, analysis of abnormal gait patterns and risk level classification. There were three fall risk levels corresponding to the scale, and an end-to-end attention-based convolution model was proposed to analyze gait kinematic data. Results: The model achieved an accuracy score of 0.878 and a recall score of 0.897 on the test set. In addition, we applied an attention-based heatmap to visualize the input data and features across the model. The color bars in the heatmap highly correlate with the level of fall risk and can serve as an indicator of the abnormal gait pattern. Conclusions: An end-to-end attention-based convolution model achieved a favorable result.Besides, the heatmap could serve as the indicator of risk level for each step and also provide further clues to the mechanism of falling. It has the potential to assist doctors in clinical work and contribute to further knowledge discovery.