External Validation of prognostic models predicting independent gait in stroke patients.

Abstract

Rationale: After stroke, early prediction of the likelihood that a patient will recover independent gait can provide valuable information to clinicians, patients and their families. Yet, there are no externally validated prognostic models predicting independent gait after stroke. The aim of this study is to externally validate prognostic models that predict independent gait after stroke and adjust if needed. Design: Validation study using baseline data from the Danish Stroke Registry and discharge data from adult stroke patients following rehabilitation from Hammel Neurorehabilitation and Research Centre. Method: A systematic literature search identified two prognostic models (Kwah- model and EPOS- model) predicting independent gait with or without walking aid. Both models used predictors easily obtained in clinical practice and quality assessment with the PROBAST. To be able to validate the models, predictor variables were matched to proxies in the validation cohort. The area under the receiver-operating characteristic curve (AUC) and calibration curves were calculated. If AUC was < 0.7, adjustments were made to re-estimate the intercepts and regression coefficients. Results: Two studies met the selection criteria, both scoring ‘high’ risk of bias. We were unable to validate the EPOS- model, because there was no suitable proxy in the validation cohort. The Kwah- model used National Institute of Health Stroke Scale (NIHSS) score and age to predict independent gait, measured with the Motor Assessment Scale (MAS). In the validation cohort, the Scandinavian Stroke Scale (SSS) was used as a proxy for NIHSS and the Functional Independent Measure for the MAS. The model showed an AUC=0.68 (95%CI: 0.65–0.71) and overestimation in the calibration curve. Re-estimation of the intercept and/or regression coefficients showed little improvement (AUC between 0.61-0.69). Finally, adding a new predictor (SSS leg) to the model improvement of discrimination of the model (AUC 0.79). Conclusion: External validation of the Kwah- model showed insufficient discrimination and calibration in our validation cohort. Updating the model by adding SSS leg score showed optimal calibration and fair discrimination in this dataset. The updated model needs external validation to determine if it is valid and whether it can be used in clinical practice.