The potential of Latent Class Analysis in diagnostic test validation for canine Leishmania infantum infection

Abstract

Accuracy assessment of diagnostic tests may be seriously biased if an imperfect reference test is used such as parasitology in the diagnosis of visceral leishmaniasis. We compared classical validity analysis of serological tests for Leishmania infantum with Latent Class Analysis (LCA), to assess whether it circumvented the gold standard problem. Clinical status, three serological tests (IFAT, ELISA and DAT) and parasitological data were recorded for 151 dogs captured in an endemic area. Sensitivity and specificity estimates from the 2×2 contingency tables were broadly corroborated by LCA, but the latter method provided more precise estimates that were robust for the different fitted models. It furthermore yielded a higher prevalence of infection and indicated that parasitology was only 55% sensitive. LCA seems a promising technique for test validation, but caution is required when applying it to sparse data sets. The feasibility and applicability of LCA in infectious disease epidemiology is discussed.