This study aims to identify, using random forest analysis, specific parameters contributing to improved walking speed in the 10-meter walking test (10mWT) through Welwalk gait training in stroke patients. This knowledge is expected to guide the optimization of rehabilitation programs using Welwalk.
From March 2021 to November 2022, 21 stroke patients who underwent Welwalk gait training at a convalescent rehabilitation hospital were included. Inclusion criteria were patients with Functional Ambulation Categories (FAC) of 2 or less at rehabilitation start and improved to FAC 3 or higher by the end. Evaluations included age, sex, paralyzed side, Stroke Impairment Assessment Set (SIAS), Barthel Index (BI), Functional Independence Measure (FIM), 10mWT, and multiple gait parameters measured by Welwalk.Random forest analysis was used, with the dependent variable being 10mWT walking time and independent variables being Welwalk parameters and physical functions at the end of rehabilitation. The training data proportion was set at 60%, with 500 trees. Model accuracy was evaluated using the Out-Of-Bag (OOB) error. Statistical analysis was performed using R version 4.2.1.
Based on random forest analysis, among the top 10 variables considered most important for the results of the 10mWT, six Welwalk parameters were selected: ①Average heel load (Mean Decrease Gini: 2125.5), ②Proportion of stance phase (2106.1), ③stride length (1969.1), ④Knee extension assist (1682.1), ⑤Coefficient of variation of swing time (1618.9), and ⑥Coefficient of variation of knee flexion angle (1604.0). The Out Of Bag (OOB) error rate was 3.71%. The model's prediction accuracy was 87.6%. The median (IQR) at the start and end are as follows: ①Average heel load: 64 (14) → 52 (12), ②Proportion of stance time: 71.9% (5) → 69.8% (7.6), ③stride length: 41.7 cm (11.9) → 56.7 cm (34.0), ④Knee extension assist level: 9 (1) → 2 (3), ⑤Coefficient of variation of swing time: 0.21 (0.09) → 0.12 (0.08), ⑥Coefficient of variation of knee flexion angle: 0.11 (0.04) → 0.11 (0.07).
Parameters such as average heel load, proportion of stance time, stride length, knee extension assist level, and coefficients of variation for swing time and knee flexion angle significantly contribute to improving walking speed. These findings may help set standardized assist parameters for enhancing walking speed in robot-assisted gait training. Future research should optimize rehabilitation based on these parameters and assess long-term effects on walking ability.
Clinically, training utilizing these parameters can efficiently enhance patients' walking abilities. Real-time evaluation enables optimal, patient-specific training plans, improving treatment outcomes. By advancing knowledge in robot-assisted rehabilitation, physical therapists can effectively use the latest technologies to provide higher-quality care.
Walking Speed
Robot
