Inter-laboratory evaluation of serological tests using Bayesian latent class models: A case study for bovine viral...

Accurate estimation of diagnostic test performance is crucial for epidemiological studies and disease control programs. Bayesian latent class models (BLCMs) provide a robust statistical approach to estimate these parameters in the absence of a gold standard test. This study aimed to establish a proof of concept for interlaboratory diagnostic test evaluation and to develop metrics for model fit validation, using serological detection of bovine viral diarrhoea as a case study. A total of 485 samples were collected from France, the Netherlands, Sweden and the United Kingdom and tested in four laboratories using six commercial ELISA kits. We initially fit a 6-test-4-population Hui-Walter model with both minimally informative and strong priors, as well as covariance terms. Model fit was assessed through four novel posterior predictive metrics, targeting the multinomial response frequency (LPmf), test-specific positivity (LPtp), pairwise crude agreement (LPag) and population-specific re-estimation of sensitivity and specificity (LRse/LRsp). BLCM results showed that almost all tests exhibited high sensitivity and specificity (>95 %). In addition, the model fit metrics identified one test breaching the assumption of constant test performance across populations which was therefore removed from the final model. This highlights the importance of robust model validation strategies to ensure reliable estimates. Our findings demonstrate that the joint evaluation of diagnostic tests across laboratories using BLCMs is both feasible and effective, providing robust accuracy estimates while reducing the burden on individual laboratories. As this approach does not require characterized samples, it is readily adaptable for evaluating diagnostics for emerging diseases without established gold standards.