External Validation of Two Prediction Models for Independent Gait After Stroke

Regaining independent gait is one of the main goals in stroke rehabilitation and early prediction of gait outcome is important to guide discharge planning at acute stroke units, design rehabilitation and inform patients and relatives. In the last decade, two easy-to-apply prediction models for gait were developed: the EPOS model in the Netherlands and the TWIST model in New Zealand. Although the models' performance in the development cohorts was good, this does not automatically mean that the models are ready for application in clinical practice, as it is unknown whether their performance is also good in an independent cohort from a different country and with different patient characteristics. Such external validation is an essential step towards clinical implementation of prediction models. A mobility-related problem is the occurrence of falls after stroke. Walking is among the Top 3 activities during which stroke patients fall, with the other two activities being transferring or sitting in a wheelchair. Especially soft tissue injuries after a fall are common and in 1-15% of the patients, the fall results in a fracture. Apart from the costs that arise from these injuries, falls have a negative impact on the patient's physical functioning and psychological status, with an increased dependency and fear of falling, resulting in a reduced quality of life. A systematic review found 12 studies that developed fall risk prediction models for either inpatient rehabilitation stroke patients or those living in the community. Important predictors for falls are the presence of hemi-inattention, fall history and balance deficits. However, none of the models had an acceptable performance and predictors were not always captured by a validated assessment, which is an important prerequisite for an unbiased prediction model. The primary aim of this study is to externally validate the EPOS and TWIST models for independent gait after stroke in a heterogeneous sample of subjects admi