ABSTRACT
BACKGROUND AND PURPOSE: The ARAT (Action Research Arm Test) has been used to classify upper limb motor outcome after stroke in 1 of 3, 4, or 5 categories. The COVID-19 pandemic has encouraged the development of assessments that can be performed quickly and remotely. The aim of this study was to derive and internally validate decision trees for categorizing upper limb motor outcomes at the late subacute and chronic stages of stroke using a subset of ARAT tasks. METHODS: This study retrospectively analyzed ARAT scores obtained in-person at 3 months poststroke from 333 patients. In-person ARAT scores were used to categorize patients' 3-month upper limb outcome using classification systems with 3, 4, and 5 outcome categories. Individual task scores from in-person assessments were then used in classification and regression tree analyses to determine subsets of tasks that could accurately categorize upper limb outcome for each of the 3 classification systems. The decision trees developed using 3-month ARAT data were also applied to in-person ARAT data obtained from 157 patients at 6 months poststroke. RESULTS: The classification and regression tree analyses produced decision trees requiring 2 to 4 ARAT tasks. The overall accuracy of the cross-validated decision trees ranged from 87.7% (SE, 1.0%) to 96.7% (SE, 2.0%). Accuracy was highest when classifying patients into one of 3 outcome categories and lowest for 5 categories. The decision trees are referred to as FOCUS (Fast Outcome Categorization of the Upper Limb After Stroke) assessments and they remained accurate for 6-month poststroke ARAT scores (overall accuracy range 83.4%-91.7%). CONCLUSIONS: A subset of ARAT tasks can accurately categorize upper limb motor outcomes after stroke. Future studies could investigate the feasibility and accuracy of categorizing outcomes using the FOCUS assessments remotely via video call.