Background: Substantial uncertainty surrounds the specificity of the Direct Agglutination Test (DAT) for visceral leishmaniasis (VL) in clinical suspects, since no good gold standard exists for unequivocally identifying diseased subjects. We explored the Latent Class Analysis (LCA) modelling technique to circumvent this problem.
Patients and methods: Data on 149 clinical suspects recruited in 1993-96 during a multicentre study in Sudan were re-examined. Clinical data, lymph node and bone marrow aspirate and DAT results were available. IFAT was performed in 1997 on stored filter paper blood of 80 individuals. Classical Validity Analysis (CVA) in a 2 x 2 contingency table with parasitology as a gold standard was compared with the parameter estimates produced by the best fitting LCA model.
Results: The sensitivity estimates of DAT produced by CVA (98% (89%-100%)) were almost exactly reproduced by LCA. The specificity estimates by LCA were substantially higher than those obtained in CVA. Specificity of DAT depended, however, on whether the subject was treated for VL before. In subjects without prior treatment, CVA estimated DAT specificity at 68% (56%-79%), whereas LCA estimated it at 85% (63%-100%).
Conclusion: LCA modelling proved a useful tool, as it gave consistent estimates of test characteristics and allowed for control of confounding factors and interaction effects. Since VL is a life-threatening disease for which expensive but effective and safe treatment exists, a clinical suspect in an endemic area should be treated on the basis of a positive DAT result.