Identifying the best window size, lead time and sensor combination for classification of injurious versus non-injurious patient transfer from bed to wheelchair.

Abstract

BackgroundPatient transfers are frequent occupational tasks of caregivers, often leading to lower back injuries. While early detection of injurious moments during patient transfers is helpful to lower the risk of such injuries, it remains a critical challenge.ObjectiveWe developed a classification model for injurious and non-injurious patient transfers using a support vector machine (SVM) with inertial measurement unit (IMU) sensor data of caregivers.MethodsSixteen young adults simulated one-person patient transfers from bed to wheelchair. During trials, the kinematics and kinetics of transfer movements were recorded with a motion capture system, IMU sensor, and force plate. A simple kinematic model was used to determine compressive forces at L5/S1, which were then used to define injurious transfers. For classification, features were extracted from the IMU sensor data to be used in SVM with different window sizes and lead times. Since characteristics of the compressive forces have been documented in our companion paper, we focus on the classification model in this paper.ResultsClassification models were able to differentiate injurious versus non-injurious transfers with accuracy in the range from 84.9-100%. The performance depended on the window size, lead time, and the number and location of IMU sensors. The recommended combination for clinical use would be a window size of 0.2 s, lead time of 0.3 s, and two sensors at both thighs.ConclusionsOur study presents high-performance risk detection models during patient transfers, informing the development of the application to help address musculoskeletal injuries in caregivers.