Predicting independent dressing after stroke using path and neural network analyses

Various motor and cognitive functions affect dressing skill. Evaluating the hierarchy of factors that influence performance in daily dressing and predicting this performance are crucial for effective rehabilitation. This study aims to clarify the interrelationships among motor function, cognitive function, dressing skill, speed of dressing, and performance of dressing in daily living, and to predict the latter using a convolutional neural network (CNN) model. We conducted a retrospective study. A total of 243 stroke patients were enrolled in this study, which comprised 2 rounds of data analysis. The first round involved all 243 patients and aimed to identify the hierarchical structure in the performance of dressing in daily living. The second round focused on 153 of these patients to predict dressing performance among those able to dress themselves in a rehabilitation setting. In the first round, path analysis was used to examine the interrelationships among motor function, cognitive function, dressing skills, speed of dressing, and performance of dressing in daily living. In the second round, a predictive algorithm for dressing in daily living was developed using a CNN. The model's performance was evaluated using the area under the receiver operating characteristic curve. The model analyzing the performance of dressing in daily living revealed statistically significant causal relationships among the path coefficients for the severity of paresis, unilateral spatial neglect, attention disorder, dressing skill, and speed of dressing. Speed of dressing directly influenced the performance of dressing in daily living, exhibiting the highest standardized path coefficient in each of the path models. The CNN model, which predicts the speed of dressing and the performance of dressing in daily living, performed well, achieving the highest area under the receiver operating characteristic curve of 0.939 ± 0.015. The severity of paresis, unilateral spatial neglect, attention disorder, and dressing skill had indirect effects on the performance of dressing in daily living by influencing the speed of dressing. The speed of dressing could predict the performance of dressing in daily living with high accuracy. Our results suggest that the performance of dressing in daily living strongly depends on the speed of dressing.