Early prediction of upper limb functioning after stroke using clinical bedside assessments: a prospective longitudinal study

 WHAT THE HELL IS IT GOING TO TAKE TO STOP STUPID RESEARCH LIKE THIS THAT ONLY PREDICTS FAILURE TO RECOVER? How many people need to be fired before we actually try to solve stroke to 100% recovery?

Early prediction of upper limb functioning after stroke using clinical bedside assessments: a prospective longitudinal study

• Margit Alt Murphy,
• Ahmad Al-Shallawi,
• Katharina S. Sunnerhagen &
• Anand Pandyan 

Scientific Reports volume 12, Article number: 22053 (2022) Cite this article

• 5 Altmetric

• Metrics details

Abstract

Early and accurate prediction of recovery is needed to assist treatment planning and inform patient selection in clinical trials. This study aimed to develop a prediction algorithm using a set of simple early clinical bedside measures to predict upper limb capacity at 3-months post-stroke. A secondary analysis of Stroke Arm Longitudinal Study at Gothenburg University (SALGOT) included 94 adults (mean age 68 years) with upper limb impairment admitted to stroke unit). Cluster analysis was used to define the endpoint outcome strata according to the 3-months Action Research Arm Test (ARAT) scores. Modelling was carried out in a training (70%) and testing set (30%) using traditional logistic regression, random forest models. The final algorithm included 3 simple bedside tests performed 3-days post stroke: ability to grasp, to produce any measurable grip strength and abduct/elevate shoulder. An 86–94% model sensitivity, specificity and accuracy was reached for differentiation between poor, limited and good outcome. Additional measurement of grip strength at 4 weeks post-stroke and haemorrhagic stroke explained the underestimated classifications. External validation of the model is recommended. Simple bedside assessments have advantages over more lengthy and complex assessments and could thereby be integrated into routine clinical practice to aid therapy decisions, guide patient selection in clinical trials and used in data registries.

Introduction

Early and accurate prediction of post-stroke recovery potential, ideally during the first week, is needed to assist selection of treatment approaches and inform patient selection in clinical trials¹. A range of models to predict upper limb motor outcome have been published^2,3,4,5,6. At the simplest level, measurable grip strength at 1 month⁵ or presence of shoulder abduction or finger extension at 3-days post stroke⁷ are suggested as plausible predictors for upper limb recovery. The more complex models propose various neurophysiological and neuroimaging techniques combined with clinical assessments^8,9,10.

There are, however, significant barriers to the clinical implementation of existing prediction algorithms. For example, use of more comprehensive scales, such as Fugl-Meyer Assessment, within days of stroke is a challenge in acute settings, particularly in patients with complex needs. The requirement to take repeated measurements within the first weeks¹¹ can also be problematic when typical length of stay within a stroke unit is less than 2-weeks^12,13. Further, the need to draw upon neurophysiological and neuroimaging techniques to determine the corticospinal integrity are costly and not accessible to most clinical practices^1,14. Prediction models using easily accessible clinical data i.e. simple clinical tests with routinely available equipment early after stroke^5,6,15 would be a more realistic solution for developing clinically usable prognostic algorithms.

Prediction models only including clinical assessments have shown to be inferior compared to the models combining clinical and neurophysiological or neuroimaging techniques^9,16,17. This seem to be particularly valid for patients with initial poor motor function^9,16,17. Recently, a prediction algorithm only including clinical bedside assessment reported an overall accuracy of 61% at predicting upper limb activity capacity at 3 months post stroke, although the sensitivity and specificity varied across the four outcome categories¹⁷. An external validation of a prediction model for upper limb activity capacity at six months post stroke, discriminating poor outcome (Action Research Arm Test < 10) and using shoulder abduction and finger extension as clinical predictor variables, showed high sensitivity (> 0.80) but lower specificity (0.40–0.70)¹⁸. For discrimination of a higher outcome level (Action Research Arm Test > 32), likewise, the sensitivity was high (> 0.92) but specificity was lower (0.28–0.60)¹⁸. These studies confirm that prediction models only using clinical assessments can provide clinically useful information, perform better than a chance alone and could therefore be considered as alternatives for more complex models¹⁷. Based on previous literature, to reach clinical relevance, a prediction algorithm only including clinical assessments should be expected reach an accuracy, sensitivity and specificity at least between 60 and 70%.

To target this clinical need we aimed to identify a set of clinical assessments feasible in routine practice early after stroke that can provide an accurate and differentiated prediction of upper limb activity capacity at 3-months post-stroke.