Development of a Gait Independence Prediction Model in Patients With Stroke in a Convalescent Rehabilitation Ward: A Comparison of Decision Tree and Random Forest Models.

aimIn this study, we aimed to examine the clinical utility of a classification and regression tree (CART) model for predicting independent ambulation at discharge based on physical function at admission to a convalescent rehabilitation ward, by comparing its performance with that of a random forest (RF) model. Seventy-three patients with stroke admitted to a convalescent rehabilitation ward were included.

methodsIndependent ambulation at discharge was defined using the Functional Independence Measure (FIM) locomotion item (walk/wheelchair): patients with a score ≥6 and ambulation as the primary mode of mobility were classified as independent, whereas those with a score <6 or wheelchair use as the primary mode of mobility were classified as nonindependent. The dataset was randomly divided into training (70%) and validation (30%) sets, and CART and RF models were developed using the training data and evaluated using the validation data.

resultsIn the CART model, patients with a Trunk Impairment Scale (TIS) score <9 were classified as gait-independent when the FIM cognitive score was ≥30.5. Among patients with a TIS score ≥9, those aged <76.5 years were classified as independent, whereas those aged ≥76.5 years were classified as independent when the FIM cognitive score was ≥22.5. The area under the receiver operating characteristic curve was 0.832 and 0.856 for the CART and RF models, respectively, with no significant difference between the two models according to the DeLong test (p = 0.58).

conclusionThese findings suggest that the CART model demonstrates discriminative ability comparable to that of the RF model and can hierarchically visualize the likelihood of gait independence based on admission assessments, thereby supporting clinical decision-making and intervention planning in convalescent rehabilitation wards.