Proof-of-concept of bayesian latent class modelling usefulness for assessing diagnostic tests in absence of diagnostic standards in mental health.

This study aimed at demonstrating the feasibility, utility and relevance of the Bayesian Latent Class Modelling (BLCM), not assuming a gold standard, when assessing the diagnostic accuracy of the first hetero-assessment test for early detection of occupational burnout (EDTB) by healthcare professionals and the OLdenburg Burnout Inventory (OLBI). We used available data from OLBI and EDTB completed for 100 Belgian and 42 Swiss patients before and after medical consultations. We applied the Hui-Walter framework for two tests and two populations and ran models with minimally informative priors, with and without conditional dependency between diagnostic sensitivities and specificities. We further performed sensitivity analysis by replacing one of the minimally informative priors with the distribution beta1,2 at each time for all priors. We also performed the sensitivity analysis using literature-based informative priors for OLBI. Using the BLCM without conditional dependency, the diagnostic sensitivity and specificity of the EDTB were 0.91 (0.77-1.00) and 0.82 (0.59-1.00), respectively. The sensitivity analysis did not yield any significant changes in these results. The EDTB’s sensitivity and specificity obtained by a BLCM approach are better compared to the previous studies when EDTB was evaluated against OLBI, considered as a gold standard. These findings show the utility and relevance of BLCM in the absence of a gold standard.