It is important to note that in all these studies, including ours, ‘recovery’ in ambulation and upper limb function does not necessarily imply complete recovery. Many patients deemed to have recovered motor function using our operational definitions may still have had significant limitations in higher levels of mobility or more complex upper limb functional tasks. Several acute stroke studies have considered age (Dallas et al 2008, de Weerdt et al 1987, Hu et al 2010, Loewen and Anderson 1990, Meldrum JAK inhibitor et al 2004, Veerbeek et al 2011, Wandel et

al 2000), and severity of stroke (Au-Yeung and Hui-Chan 2009, Dallas et al 2008, Hu et al 2010) in their multivariate analyses to identify predictors of ambulation or upper limb function.

Only one study has found age and severity of stroke as significant predictors of ambulation. This study recruited patients from a stroke intensive care unit. Patients were included in that study only if they were referred for rehabilitation (Hu et al 2010). Another study that investigated the benefits of constraint-induced movement therapy in people six months after stroke also reported that age was a predictor for upper limb function (Fritz et al 2006). In these two studies, the cohorts might not be representative of patients seen early AC220 after stroke. Age and NIHSS have previously been shown to be strong predictors of mortality (Konig et al 2008, Weimar et al 2004), disability (Johnston et al 2007), and independence with activities of daily living (Johnston et al 2007, Konig et al 2008, Weimar et al 2004) in acute stroke cohorts. Consequently these predictors appear to have broad predictive utility. Their routine use in acute stroke units will facilitate external validation of our prediction models in other cohorts. One limitation of the Idoxuridine NIHSS is that it is a complex assessment that requires training to administer (Reid et al 2010). This potentially undermines its clinical usefulness. However online training and access to the scale (Kasner 2006)

have overcome some of these problems. An advantage of the NIHSS is that it provides information on a variety of stroke-related impairments that can be used by various health professionals in the acute stroke setting (Kasner 2006). The NIHSS can also be administered to patients who do not have good cognition or language, whereas this can be problematic with the MAS. We therefore recommend the use of the NIHSS in future prediction models of ambulation and upper limb recovery after stroke. The strengths of our study include the consecutive recruitment of patients seen early after stroke, the minimal loss to follow-up, the low risk of over-fitting of the prediction model, and the strong performance of the prediction models (discrimination and calibration results).