Validation, Categorizing, and Prediction of Upper Limb Outcomes after Stroke

Author's School

Graduate School of Arts and Sciences


Author's Department/Program

Movement Science


English (en)

Date of Award

Spring 5-15-2022

Degree Type


Degree Name

Doctor of Philosophy (PhD)

Chair and Committee

Catherine E. Lang

Committee Members

Linda Van Dillen


The incidence and costs of stroke in the United States are projected to rise over the next decade because of the aging population. Declining stroke mortality over the past few decades means that more people survive stroke and live with physical, cognitive, and emotional disability. Stroke remains one of the leading causes of disability in the United States because very few survivors experience a full recovery of their upper limb. Upper limb recovery after stroke is critical to performing activities of daily living and physical and occupational therapies are one of the only treatment options to address these challenges. The World Health Organization’s (WHO) International Classification of Functioning, Disability, and Health Framework (ICF) informs our understanding of the importance of measuring upper limb changes across measurement levels, showing that improvements seen in one level (i.e. domain) do not directly transfer to another. Knowing this, it is important to evaluate existing prediction models of motor outcomes after stroke while simultaneously developing novel tools available to researchers and clinicians to facilitate measurement of the upper limb across the ICF domains. This dissertation work performs an external validation of an existing prediction model of upper limb capacity (UL; capability measured in the clinic) after stroke, identifies and defines categories of UL performance (actual UL use in daily life) in people with and without neurological UL deficits, and explores how early clinical measures and participant demographic information are associated with subsequent categories of UL performance after stroke.Recently, prediction algorithms of upper limb capacity after stroke have been developed to facilitate treatment selection, discharge planning, and goal setting for clinicians and their clients. Prediction models have tremendous clinical utility because they aid in the clinical decision making required to select the appropriate efficacious and emerging interventions that afford improvements in upper limb functional capacity, measured by standardized assessments in the therapy clinic. Prior to wide spread implementation of existing prediction algorithms into routine rehabilitation care, however, it is necessary to understand how small healthcare system differences and availability of neurophysiological assessments affect external validation of the models. In Chapter 2, we test how well an algorithm with clinical measures, developed for use in another country, applies to persons with stroke within the United States. Knowing the importance of measurement across ICF domains, it is necessary to develop tools that facilitate clinical decision making and implementation of upper limb performance data into routine rehabilitation care. The use of wearable sensor technology (e.g., accelerometers) for tracking human physical activity have allowed for measurement of actual activity performance of the upper limb in daily life. Data extracted from accelerometers can be used to quantify multiple variables measuring different aspects of UL activity in one or both limbs. A limitation is that several variables are needed to understand the complexity of UL performance in daily life. As a solution to the multi-variable problem, it would be helpful to form categories of UL performance in daily life. If natural groupings occur among multiple UL performance variables calculated from accelerometry data, then these groupings could facilitate clinical decision making and implementation of upper limb performance data into routine rehabilitation care. In Chapter 3 we identify and define categories of UL performance in daily life in adults with and without neurological deficits of the upper limb. Prediction of motor outcomes after stroke have tremendous clinical utility, however there have been limited efforts to develop prediction models of upper limb performance (i.e., actual upper limb activity) in daily life after stroke. With advances in computing power, it is possible to capitalize on machine learning techniques to predict upper limb performance after stroke. These techniques allow for predicting a multivariate categorical outcome. This is important because it provides more information about the expected upper limb outcome to people with stroke, their families, and clinicians than a single continuous variable or a binary category (e.g., good or poor). Chapter 4 of this dissertation explores how different machine learning approaches can be used to understand the association between early clinical measures and participant demographics to the UL performance categories from a later post stroke time point. Our findings provide strong support for the importance of measuring recovery of the UL across ICF domains, not just with impairment and capacity level measures. Collectively this work provides preliminary measurement tools that could eventually be available to rehabilitation clinicians following subsequent validation efforts. Additionally, this work provides a rich exploration into the strengths, weaknesses, and limitations of analytical methods and their impact on validation efforts.