A Novel Upper-Extremity Sensor-Based Approach to Predict COPD Adverse Outcomes in an Acute Setting

Background: Decisions about the intensity of treatment for patients with COPD are influenced by the ability to predict upcoming adverse outcomes after treatment. The 6-minute walk distance test is commonly used to assess functional capacity in patients with COPD for predicting adverse outcomes. Although the 6-minute walk distance showed adequate reliability and validity, it is often not feasible for frail patients. Therefore, an alternative objective, quick, and simple approach for assessing functional capacity in COPD is needed. Research Question: Is an upper extremity test an accurate and feasible method for assessing fnctional capacity individuals with COPD? Study Design and Methods: We previously developed and validated an upper extremity function (UEF) test, incorporating motor function kinematics and muscle force measures for assessing functional capacity in COPD. In this study, with the goal of longitudinal evaluation of the UEF test for predicting adverse outcomes, we recruited 192 hospitalized older adults that were admitted due to COPD exacerbation. In-hospital (ie, mortality, excessive length of stay, complications) and longitudinal 90-day (ie, acute COPD exacerbation, mortality, readmission) outcomes were recorded. We developed a risk stratification model using elastic net regularization for selecting optimum feature sets (kinematics and muscle model parameters) in combination with support vector machine to predict adverse outcomes. Results: Results from 10-fold cross-validation for model prediction showed, on average, accuracy of 78% in predicting in-hospital outcomes and accuracy of 76% in predicting 30- to 90-day longitudinal outcomes. Interpretation: Current findings suggested that the UEF test may provide an efficient method for risk stratifying older adults with COPD, with accuracy higher than other available tools within our recorded data set (ie, clinical frailty score and COPD assessment test with accuracies < 61%).