External Validation of the Time to Walking Independently After Stroke (TWIST) Prediction Algorithm within 72 Hours Poststroke

BACKGROUND: The TWIST algorithm was developed to predict the time to walk independently after stroke based on clinical assessments performed 1 week poststroke. The objective of this study was to externally validate this algorithm within 72 hours after stroke in a heterogeneous sample as a next step towards clinical implementation in an acute stroke unit setting. METHODS: This prospective longitudinal observational cohort study included adult stroke patients unable to walk independently (Functional Ambulation Categories [FAC] <4) within 72 hours poststroke. Within this timeframe, the predictors Trunk Control Test (TCT) and paretic hip extensor strength were assessed. Patients were categorized into 'independent 6 weeks', 'independent 12 weeks', or 'dependent 12 weeks' based on the algorithm's prediction. After 6 and 12 weeks, walking ability was evaluated with the FAC. The algorithm's performance was assessed by area under the curve (AUC), classification measures, and accuracy. In case of insufficient performance for an outcome category, differences in patients' characteristics were explored. RESULTS: 146 participants were analyzed. Multiclass AUC was 0.73, the sensitivity for 'independent 6 weeks' was 0.83 (95% CI 0.74, 0.90) and for 'independent 12 weeks' 0.23 (95% CI 0.09, 0.44). The specificity for 'dependent 12 weeks' was 0.80 (95% CI 0.72, 0.87). CONCLUSION: Application of the TWIST algorithm within 72 hours poststroke can inform on who will regain independent gait at 6 weeks but not about who will regain independent gait between 6 and 12 weeks. Before initiating clinical implementation, the impact of predictor assessment timing and the optimization of predictions for 'independent 12 weeks' should be investigated. CLINICAL TRIAL REGISTRATION: ClinicalTrails.gov (Identifier: NCT05039047).