UCSF

Latent structure analysis can be used to determine sensitivity and specificity rates of human immunodeficiency virus antibody assays in the absence of previous clinical or laboratory results. The technique was applied to the analysis of data obtained when a panel of serum samples, collected as part of a large-scale screening project, were subjected to four conventional bioassays (ag121, p24, gp120, and an enzyme-linked immunosorbent assay). To determine the accuracy of this statistical approach, the results of latent structure analysis were compared with the known clinical diagnoses of patients from whom the samples were taken, and nearly 100% agreement was obtained. Although a two-class latent structure model had some predictive value, a three-class model more adequately explained assay patterns. The use of the four standard assays in conjunction with the statistical methods described here would largely reduce the need for confirmatory Western blot assays in analyses of large panels of samples.