A predictive model for functionality improvement after stroke rehabilitation

This study develops a simple predictive model for identifying stroke patients who have a better chance of showing improved activities of daily living (ADL) outcomes following a stroke. The cohort of 489 stroke patients was divided into testing and training groups. Multivariate logistic regression analysis was conducted for each model. Four models were compared using the C statistic (AUC), Akaike's information criterion (AIC), and other metrics. The best model was assessed using a nomogram. Univariate analysis revealed that several variables measured significantly higher at discharge than at admission, including manual muscle testing, standing, and so on. Multivariate logistic regression analysis revealed that activities-specific balance confidence, Brunnstrom recovery stage for lower extremities, standing, the mini-balance evaluation systems test, and the Hamilton anxiety scale were independent predictors of ADL. Model 1 was found to be more accurate for the prediction of ADL (AUC: training, 0.916 [0.889−0.943] and test, 0.887 [0.806−0.968]; AIC: training, 257.42 and test, 76.79) than model 2 (AUC: training, 0.850 [0.894−0.806] and test, 0.819 [0.715−0.923]; AIC: training, 314.44 and test, 83.78), model 3 (AUC: training, 0.862 [0.901−0.823] and test, 0.830 [0.731−0.929]; AIC: training, 307.76 and test, 86.55), and model 4 (AUC: training, 0.862 [0.901−0.823] and test, 0.833 [0.733−0.932]; AIC: training, 305.8 and test, 86.28). A multivariate model can be used to predict functionality improvement, as measured by ADL, following hospitalization with a stroke.