Improving Measurement Efficiency of the Cough Severity Index With Item Response Theory.

Abstract

The larynx is central to voice, breathing, swallowing, and cough-such that routinely tracking related outcomes can introduce burden. Within these assessments, the Cough Severity Index (CSI) is frequently deployed. We thus utilized item response theory (IRT) to determine: (1) which items confer the most information, (2) if proposed 5-item scales yield substantially differing discrimination and location parameters, compared to the 10-item instrument, and (3) whether a subset of items has more favorable characteristics as an initial assessment (i.e., for whether all 10 items are needed). Item discrimination, location, and information parameters were calculated from 4240 consecutive CSI responses. Based on these data, 5-item subsets were proposed, evaluated, and compared with each other and the 10-item instrument via information function curves and receiver-operator characteristic (ROC) analyses. Performance of 5-item mean scores < 1 as a potential initial assessment for the full results was also assessed. Discrimination parameters ranged from 0.65-2.76, with the greatest discrimination in items about embarrassment or avoiding places due to cough. These data, along with residual item correlations and location parameters, delineated a 5-item subset, which also incorporated cough when lying down, affecting voice, and upset by cough for comparative analysis. ROC-AUCs revealed no difference between the 5-item subset and parent 10-item instrument in ability to discriminate cough-specific clinical diagnoses. One 5-item subset performed better as an initial assessment (mean 4.9, 95% CI 4.7-5.2). IRT data identified a 5-item CSI subset which reflects the cough trait and provides an initial efficient assessment mechanism.