Are there different gait profiles in patients with advanced knee osteoarthritis? A machine learning approach. (August 2021)

• Record Type: Journal Article
• Authors: Leporace, Gustavo; Gonzalez, Felipe; Metsavaht, Leonardo; Motta, Marcelo; Carpes, Felipe P.; Chahla, Jorge; Luzo, Marcus
• Is Part Of: Clinical biomechanics. Volume 88(2021)
• Publication Date: 2021-08
• Subjects: Self-organizing maps -- Kinematics -- Osteoarthritis -- Gait -- Artificial intelligence -- Three-dimensional; Biomechanics -- Periodicals; Osteopathic medicine -- Periodicals; Biomechanics -- Periodicals; Osteopathic Medicine -- Periodicals; 612.76
• DOI: 10.1016/j.clinbiomech.2021.105447
• Languages: English
• ISSNs: 0268-0033

Abstract

Abstract: Background: Determine whether knee kinematics features analyzed using machine-learning algorithms can identify different gait profiles in knee OA patients. Methods: 3D gait kinematic data were recorded from 42 patients (Kellgren-Lawrence stages III and IV) walking barefoot at individual maximal gait speed (0.98 ± 0.34 m/s). Principal component analysis, self-organizing maps, and k-means were applied to the data to identify the most relevant and discriminative knee kinematic features and to identify gait profiles. Findings: Four different gait profiles were identified and clinically characterized as type 1: gait with the knee in excessive varus and flexion (n = 6, 14%, increased knee adduction and increased maximum and minimum knee flexion, p < 0.01); type 2: gait with knee external rotation, either in varus or valgus (n = 11, 26%, excessive maximum and minimum external rotation, p < 0.001); type 3: gait with a stiff knee (n = 17, 40%, decreased knee flexion range of motion, p < 0.001); and type 4: gait with knee varus 'thrust' and decreased rotation (n = 8, 19%, increased and reduced range of motion in the coronal and transverse plane, respectively, p < 0.05). Interpretation: In a group of patients with homogeneous Kellgren-Lawrence classification of knee OA, gait kinematics data permitted to identify four different gait profiles. These gait profiles can be a valuable tool for helping surgical decisions and treatment. To allow generalization, further studies should be carried with a larger and heterogeneous population. Highlights: A machine learning approach was applied to 3D motion analysis during gait. Four gait profiles have been found in same stage knee osteoarthritis patients. Gait features in each group had been associated previously with knee functionality. These gait profiles can be a valuable tool for helping therapeutic decision-making. Future studies should compare knee pain and functional capacity among groups.