Interobserver variability of radiographic methods for the evaluation of left atrial size in dogs

The objectives of this retrospective, observer agreement study were to (a) test variability of radiographic left atrial dimension (RLAD) and vertebral left atrial size (VLAS) measurements among observers with different levels of expertise in thoracic radiology and cardiology, (b) assess whether one method is better than the other in detecting left atrial enlargement (LAE), and (c) assess the agreement among RLAD, VLAS, and American College of Veterinary Internal Medicine (ACVIM) classes. Seventy-four dogs (eight healthy and 66 with mitral valve disease) with thoracic radiographs and echocardiography performed on the same day were reviewed. Thirty showed echocardiographic LAE. Left atrial dimension was quantified using RLAD and VLAS by six different operators with three levels of clinical experience in veterinary cardiology/radiology. Vertebral heart score and fourth thoracic vertebra (T4) were also measured. Differences in T4, vertebral heart score (VHS), RLAD, and VLAS measurements were found among six operators and among the three levels of clinical expertise as well as between veterinary cardiology readers and veterinary radiology readers (P < .05). The area under the receiver operating characteristic (AUC) curve for VHS showed good performances for all observers and level and type of expertise; the AUC for RLAD and VLAS was suboptimal only for the radiology student. Our RLAD and VLAS cutoffs (1.9 and 2.43 v, respectively) were better related to qualitative radiographic than quantitative echocardiographic LAE evaluation. Radiographic LA dimension and VLAS showed an increase proportional to the worsening of the ACVIM class. In conclusion, these results allow us to affirm that RLAD and VLAS are reproducible measurements for detecting LAE. Better performances are associated with clinical expertise and background.