Agreement of Specific Lung Sounds Auscultation by Veterinarians for the Detection of Bronchopneumonia in Calves

Lung auscultation is a common method for the routine diagnosis of calf bronchopneumonia. However, its repeatability among operators has been criticized. Determine agreement among veterinarians for specific lung sounds after a short tutorial to standardize the definition of lung sounds. Forty lung sounds from a larger dataset collected at 4 veal calf farms that housed 495-815 animals were submitted online to 10 different veterinarians. After a short tutorial on lung sound auscultation, the raters were asked to detect the presence of any abnormal sounds and to differentiate among wheezes, crackles, and bronchial sounds. Raw percentage of agreement (PA), Gwet's agreement coefficient type 1 (AC1), Krippendorff's alpha (Ka), and Fleiss kappa (KFleiss) were chosen as agreement indicators in the absence of a gold standard indicator to assess agreement. The different indicators were interpreted based on a priori reported benchmarks. The agreements were fair to good for almost all lung sound indicators. For the presence of any abnormal lung sound, the reported agreements (95% confidence intervals [CI]) were 0.781 (0.716-0.845), 0.646 (0.514-0.777), 0.403 (0.351-0.455), and 0.293 (0.137-0.493) for PA, AC1, Ka, and KFleiss, respectively. The same indicators were 0.769 (0.694-0.845), 0.615 (0.446-0.784), 0.426 (0.378-0.475), and 0.425 (0.293-0.563) for wheezes, 0.754 (0.685-0.823), 0.643 (0.503-0.782), 0.21 (0.146-0.275), and 0.208 (0.097-0.327) for crackles, and 0.636 (0.571-0.701), 0.345 (0.179-0.512), 0.182 (0.131-0.232), and 0.18 (0.081-0.279) for bronchial sound detections, respectively. Agreement among raters auscultating calf respiratory sounds was higher than previously reported. However, improvement is still possible to increase auscultation agreement.