Will you please stop with this fucking useless research on predicting failure to recover! And just do what survivors want! EXACT 100% RECOVERY PROTOCOLS! I 'd have you all fired.
Predicting later categories of upper limb activity from earlier clinical assessments following stroke: an exploratory analysis
Journal of NeuroEngineering and Rehabilitation volume 20, Article number: 24 (2023)
Abstract
Background
Accelerometers allow for direct measurement of upper limb (UL) activity. Recently, multi-dimensional categories of UL performance have been formed to provide a more complete measure of UL use in daily life. Prediction of motor outcomes after stroke have tremendous clinical utility and a next step is to explore what factors might predict someone’s subsequent UL performance category.
Purpose
To explore how different machine learning techniques can be used to understand how clinical measures and participant demographics captured early after stroke are associated with the subsequent UL performance categories.
Methods
This study analyzed data from two time points from a previous cohort (n = 54). Data used was participant characteristics and clinical measures from early after stroke and a previously established category of UL performance at a later post stroke time point. Different machine learning techniques (a single decision tree, bagged trees, and random forests) were used to build predictive models with different input variables. Model performance was quantified with the explanatory power (in-sample accuracy), predictive power (out-of-bag estimate of error), and variable importance.
Results
A total of seven models were built, including one single decision tree, three bagged trees, and three random forests. Measures of UL impairment and capacity were the most important predictors of the subsequent UL performance category, regardless of the machine learning algorithm used. Other non-motor clinical measures emerged as key predictors, while participant demographics predictors (with the exception of age) were generally less important across the models. Models built with the bagging algorithms outperformed the single decision tree for in-sample accuracy (26–30% better classification) but had only modest cross-validation accuracy (48–55% out of bag classification).
Conclusions
UL clinical measures were the most important predictors of the subsequent UL performance category in this exploratory analysis regardless of the machine learning algorithm used. Interestingly, cognitive and affective measures emerged as important predictors when the number of input variables was expanded. These results reinforce that UL performance, in vivo, is not a simple product of body functions nor the capacity for movement, instead being a complex phenomenon dependent on many physiological and psychological factors. Utilizing machine learning, this exploratory analysis is a productive step toward the prediction of UL performance.
Trial registration NA
Background
Wearable movement sensors allow for direct measurement of upper limb (UL) activity in daily life, i.e. performance [1]. Performance is operationally defined in the World Health Organization’s (WHO) International Classification of Function (ICF) model as activity in the unstructured, free-living environment, and is distinguished from capacity, operationally defined as the capability for activity in a structured or standardized environment [2, 3]. The most common wearable sensors used are accelerometers, from which numerous clinically relevant variables about UL activity can be computed to provide insight into how people with or without neurological impairment use their ULs in daily life [4,5,6,7]. Data extracted from bilateral, wrist-worn wearable sensors can be used to quantify UL performance variables measuring the duration [8, 9], symmetry [6, 8, 10, 11], magnitude [5, 7, 12], and variability of one or both limbs [5, 7, 12]. Each UL performance variable conveys slightly different information about the collective nature of UL use; multiple variables may provide a fuller understanding of the scope of UL performance in daily life [13]. As a solution to the multi-variable problem, we recently categorized UL performance in adult cohorts with and without stroke [14]. The most parsimonious solution was five categories of UL performance formed from five UL performance variables named: (A) Minimal activity/rare integration; (B) Minimal activity/limited integration; (C) Moderate activity/moderate integration; (D) Moderate activity/full integration; (E) High activity/full integration. The UL performance categories are multi-dimensional, with each category providing information about UL activity with respect to the different movement characteristics in adults with and without neurological UL deficits. Thus, the five categories of UL performance may provide a more complete measure of UL use in daily life [13, 14].
Early prediction of motor outcomes after stroke has tremendous clinical utility [15, 16]. Our next step, therefore, was to explore what factors might predict someone’s subsequent UL performance category. Predictive knowledge of subsequent outcomes can inform the delivery and specification of individualized rehabilitation services [17, 18]. This effort to predict an individual’s subsequent UL performance category is informed by the development of the PREP 2 algorithm [19, 20], which has demonstrated that prediction of an UL capacity (i.e. activity a person has the capability to do) category provides clinically-useful information to people with stroke and their families [21,22,23,24]. Advances in computing have improved upon old and led to new analysis techniques for building prediction models of UL outcomes after stroke. Recently, machine learning techniques of support vector machines (SVM) and tree-based methods (e.g., Classification and Regression Trees [CARTs]) have been used to classify people with stroke into categories with different ranges of UL capacity [17, 20, 25,26,27]. The PREP 2 prediction model was originally built and validated with a CART which resulted in the easy to interpret decision tree [20]. Machine learning techniques have the advantages of: (1) requiring fewer assumptions about the distributions of the data, (2) numerous options for non-parametric models, and (3) strong predictive capabilities [18, 26,27,28,29]. There are strengths and weaknesses to each machine learning technique. For example, the CART algorithm yields a single, easy to interpret decision tree (strength), but lower predictive accuracy on new, external samples because of high variance (weakness) [30]. An alternative to creating a single decision tree is to use ensemble classifiers like bootstrap aggregation (called “bagging”) or random forests [31]. These ensemble techniques rely on the collective judgment of many decision trees (hundreds or even thousands) in order to make a classification. These ensemble methods tend to have higher predictive power and reduce the risk of over-fitting relative to other CART methods, but at the expense of interpretability (as there is no longer one single decision tree to follow, but a whole forest of trees) [26, 31]. Capitalizing on the advantages of ensemble machine learning algorithms by applying them for prediction of UL performance outcomes could yield key insights into UL recovery post stroke.
The purpose of this study, therefore, was to explore how different machine learning techniques can be used to understand how clinical measures and participant demographics captured early after stroke are associated with the UL performance categories from a later post stroke time point. We utilized the same data set from which we had previously predicted the trajectory of single, continuous UL performance variables with regression techniques [32]. In this analysis, we attempt to predict the subsequent multivariate categories of UL performance that people with stroke fell into. We explicitly tested different machine learning methods to build predictive models with different input variables as predictors (also called feature sets) to explore how each method yields similar versus different results. Based on prior post stroke prediction models of UL capacity [17, 19, 33], performance [25, 34], and walking performance [18, 35,36,37], we hypothesized that the Shoulder Abduction Finger Extension (SAFE) measure of UL impairment [20], the Action Research Arm Test (ARAT) a measure of UL capacity [17, 32], the Area Deprivation Index (ADI) [38,39,40] and the Center for Epidemiological Studies Depression Scale (CES-D) would be the most important predictors of the subsequent UL performance category.
More at link.
No comments:
Post a Comment