Upper-limb motor impairments are common and affect quality of life. Shared-control robotic assistance systems driven by patients’ residual effort generation capacities are being developed. These systems require the integration, into their control scheme, of knowledge about the maximum voluntary torque achievable by patients at each joint angle. This study aimed to quantify the performance of simplified mathematical models primarily developed to fit healthy individuals’ torque–angle curves, as candidates for integration in the control scheme. 15 patients (6F, 9M, 62.7 ± 14.4 years) hospitalized in a French rehabilitation center for stroke ( n = 10), multiple sclerosis ( n = 4), or traumatic tetraplegia ( n = 1) were included. Passive, then maximum concentric and eccentric torques were measured in the seated position, in shoulder external–internal rotation and in elbow flexion–extension, at an imposed speed of 30°/s, using a Con-Trex® isokinetic ergometer. The normalized RMSE between modeled and experimental curves was 3.1 ± 2.0% of corresponding peak torques. Univariate linear models displayed no difference in nRMSE between mathematical models, but differences across patients ( p < 0.001, R 2 = 0.38). The distance between modeled and experimental curves was continuously lower than 10% of the peak torque over 92 ± 13% of the experimental range of motion. Regardless of the mathematical model used, torque–angle curve modeling was globally less effective than that for healthy individuals, while still allowing consideration for future use for robotic assistance for the majority of patients. Further investigation of patient-related factors affecting model quality will be necessary to assess results’ generalizability. ID-RCB: 2024-A01007-40, clinicaltrial.gov ID: NCT06608121.
Tomezzoli et al. (Sun,) studied this question.