Zusammenfassung

Recently, an innovative upper extremity frailty (UEF) meter was developed and validated based on two inertial wearable sensors and combination of sensor-derived kinematics/kinetics and patient’s demographics including age, height and body mass. This study presents an optimized model to predict frailty status based on a single wearable sensor and without demographic/anthropometrics information. A dataset of 100 trauma patients (49 frail and 51 nonfrail) were retrospectively analyzed in which two inertial sensors were attached to elbow and forearm to quantify motor performance during a 20-second repetitive elbow flexion-extension task. The test protocol included performing elbow flexion/extension as fast as possible in supine position for 20-second. The classification accuracy of new algorithm was compared with previous method as well as Trauma-Specific Frailty Index as gold standard. We extracted totally 34 UEF sensor-derived parameters indicators of slowness, exhaustion, flexibility, and weakness. A multivariate linear regression model was used to identify independent predictors. Bootstrap technique was employed for generating training and validation dataset randomly with 1000 iterations. ANOVA statistic and recursive feature elimination technique were used for reducing and optimizing the UEF parameters. After training the model, 5 independent parameters were selected. Using the model, sensitivity of 85.3%, specificity of 78.5% and accuracy of 81.6% were achieved in the validation dataset. While new results were comparable with previous method, it allows identifying frailty using a single sensor and independent of subject’s demographic/anthropometrics information and thus making it a more practical tool for busy clinics without the need of specific training or additional measurements.