physioscience 2025; 21(S 03): S5-S6
DOI: 10.1055/s-0045-1812357
Abstracts
Präsentationen/Presentations
PS 3

External validation of the EPOS prediction model for independent gait in a Swiss stroke unit setting

Authors

  • L Vinzens

    1   Kantonsspital Winterthur, Institut für Therapien und Rehabilitation, Winterthur, Switzerland
    2   Berner Fachhochschule, Departement Gesundheit, Physiotherapie, Bern, Switzerland

  • J Veerbeek

    3   Luzerner Kantonsspital, Klinik für Neurologie und Neurorehabilitation, Luzern, Switzerland

  • M Betschart

    1   Kantonsspital Winterthur, Institut für Therapien und Rehabilitation, Winterthur, Switzerland
    4   Ostschweizer Fachhochschule, Institut für Gesundheitswissenschaften, St. Gallen, Switzerland
Further Information
Also available at
 

Background Gait disorders affect most acute stroke survivors and are a major contributor to long-term disability. Accurate prognosis of gait recovery is crucial for guiding rehabilitation and supporting informed decision-making. The Early Prediction of functional Outcome after Stroke (EPOS) model, which predicts independent gait within six months, was previously externally validated in two small, homogenous samples of first-ever stroke patients without pre-stroke disability. This study aimed to externally validate the EPOS model in a heterogeneous stroke population as a step toward clinical implementation in Switzerland.

Methods A prospective longitudinal cohort study was conducted in two Swiss acute stroke units. Patients were eligible if they could not walk independently within 72 hours post-stroke, regardless of stroke characteristics or pre-stroke status. Sitting balance and paretic leg strength were assessed within the first 72 hours after stroke. Independent gait was evaluated at three months. Model performance was determined using multiple metrics.

Results 280 patients were included (mean age 74 years, 58% male, median NIHSS 4). The model demonstrated acceptable overall fit (Brier score 0.14, 95% CI 0.11–0.18) and moderate discrimination (AUC-ROC 0.74, 95% CI 0.67–0.81). Gait recovery was slightly but significantly overestimated in the highest-probability group (98%) with adequate leg strength and sitting balance. Using a classification threshold of 0.5 – identifying only patients lacking both leg strength and sitting balance as unlikely to walk independently – the model achieved a sensitivity of 0.98 (95% CI 0.95–0.99), specificity of 0.32 (95% CI 0.21–0.45), and overall accuracy of 0.83 (95% CI 0.78–0.87).

Conclusions The EPOS model showed moderate predictive performance for regaining independent gait at three months when applied within the first 72 hours post-stroke in a heterogeneous cohort. For clinical use, a high-sensitivity threshold is recommended that classifies only patients lacking both sitting balance and hemiplegic leg strength as unlikely to become independent in gait, avoiding overlooking recovery potential. Further research is needed to evaluate the model’s impact on decision-making, patient outcomes, and healthcare resource use.



Publication History

Article published online:
23 October 2025

© 2025. Thieme. All rights reserved.

Georg Thieme Verlag KG
Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany