Identifying gait subgroups in low back pain patients with artificial intelligence: implications for individualized interventions.

Purpose: The following study aimed to investigate the existence of different gait profiles in patients with low back pain (LBP) and assess their clinical characteristics.

Methods: This is a cross-sectional retrospective study that included individuals with chronic or acute LBP. Three-dimensional gait kinematics were assessed, including mean angles, range of motion (ROM), and coordination of the trunk and pelvis across three planes. An artificial intelligence algorithm was leveraged to identify distinct gait profiles using principal component analysis, self-organizing maps, and K-means clustering techniques. Clinical characteristics, such as demographics, hip and trunk passive ROM, and hip strength, were compared across profiles using the Kruskal-Wallis test with Bonferroni adjustment at a 5% significance level.

Results: 111 patients were analyzed (mean age 44.6; 56% females). Five distinct gait profiles were identified. Profile 1 (Flexed Trunk) was mainly characterized by an increased lateral trunk ROM and flexed trunk (4.1; 5.2; p < 0.05). Profile 2 (Lumbar Rectification), by trunk flexion and posterior pelvic tilt (3.4; 7.2; p < 0.05). Profile 3 (Pelvic Impairment Profile), by excessive pelvic ROM and maximum angle in the coronal plane (2.9 and 11.1; p < 0.05) and pelvic anterior tilt (12.2; p < 0.05). Profile 4 (Trunk Extension and Excessive Rotation), by trunk extension, and excessive trunk axial ROM (-3.0; 7.1; p < 0.05). Profile 5 (Tight Axial Control), by a prominent pelvis-trunk in-phase component in the coronal and axial planes with pelvic predominance (20.2%; 45.6%; p < 0.001). Profiles 1 and 2 predominantly consisted of males with higher body mass (> 77.3%; >76.8 kg; p < 0.001), while profiles 3 and 4, females with lower body mass (> 86.2%; <65 kg; p < 0.001). Profile 4 displayed increased hip passive ROM, and profiles 1 and 5 decreased (p < 0.05). No significant differences were found in age, hip strength, and passive trunk ROM between profiles (p > 0.05).

Conclusion: Each of the five profiles identified reveals distinct kinematic and physical characteristics, providing meaningful insights into clinical implications, associated pathologies, anatomical structures at risk, and management.