Lung ultrasonography used as a diagnostic test for respiratory disease diagnosis in calves: Systematic review and meta-analysis using a Bayesian latent-class modeling approach.

Bovine respiratory disease complex is a common disease that commonly affects calves in the form of bronchopneumonia. There is currently no affordable, perfectly accurate reference standard test, and the routine diagnosis is commonly based on clinical signs assessment or other ancillary tests. Lung ultrasonography (LUS) has emerged as a practical calf-side test that can be done routinely. However, information on this diagnostic test's accuracy is limited, especially because no test reaches the accuracy of a gold standard comparator test. The objective of this systematic review and meta-analysis was to determine the accuracy of LUS while adjusting for imperfect accuracy of available reference standard (RS) tests, including clinical scoring systems, using a Bayesian latent-class meta-analysis approach. A structured literature search was performed, and from 875 studies screened, 26 studies reported at least one 2 × 2 table with cross-classification of LUS (positive vs. negative) calves and RS test (positive vs. negative). The RS tests included were Wisconsin Clinical Respiratory Score (WCRS, n = 16), California Clinical Respiratory Score (n = 4), other clinical sign combinations (n = 5), clinical score combined with bronchial lavage results (n = 1), thoracic radiograph (n = 1), expert panel diagnosis (n = 1), and necropsy (n = 1). Various LUS thresholds for test positivity were reported, but 2 thresholds of maximal consolidation depth, ≥1 cm or ≥3 cm, were most commonly reported in 12 and 8 studies, respectively. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 assessment for risk of bias and applicability revealed that in addition to limitations associated with an imperfect RS test, it was difficult to know if LUS results were interpreted without the knowledge of RS test results. A Bayesian latent-class meta-analysis accounting for the imperfect accuracy of WCRS was performed for both positivity thresholds. Eight studies were available for LUS ≥1 cm. The median pooled sensitivity (95% Bayesian credible intervals) and specificity were 68.4% (51.7%-87.4%) and 91.2% (78.5%-99.4%), respectively. Five studies were available for LUS ≥3 cm. The median pooled sensitivity and specificity were 58.2% (38.4%-80%) and 95.8% (86.7%-99.4%), respectively. The LUS appeared as a moderately sensitive test with a relatively high specificity. The models revealed heterogeneity that could not be further investigated due to the low number of available studies. This is the first reported meta-analysis to determine LUS accuracy accounting for the RS test uncertainty. This study helps to identify the gaps of knowledge and reporting issues, such as standardization of LUS scanning protocol and reporting maximal consolidation in a continuous way, as well as to provide raw datasets to continue gathering information on LUS accuracy for the diagnosis of calf respiratory disease.