Performance of an enzyme-linked immunosorbent assay for the diagnosis of Brucella ovis infection in rams

Abstract

Aim. To describe the performance characteristics (sensitivity and specificity) of an enzyme-linked immunosorbent assay (ELISA) for the diagnosis of Brucella ovis infection in rams. Methods. Sera from a negative (n = 2535) and a positive (n = 589) reference population were tested in an ELISA for anti-B. ovis antibodies and cut-off values calculated from the raw, log10-transformed and fitted data. Statistical methods were used to fit curves to the frequency distribution of the data and receiver-operated characteristics analysis used to optimise the cut-off values. Results. Analysis of the frequency distribution of the positive ELISA values suggested a normal distribution of the data, whereas, in the case of the negative population, a Pearson type IV curve appeared to best fit the data. The cutoff values calculated as the mean plus 1.96 standard deviations (s.d.) from the raw, log-transformed and fitted ELISA data did not differ markedly. The differences were much greater at the mean plus 3.09 s.d. cut-off, with the cut-off value calculated from the log-transformed data giving a much better estimate of specificity. Optimisation (minimisation of classification error) of the cut-off calculated from the fitted curves suggested varying cut-off values, depending on the prevalence of B. ovis infection. Discussion. Calculation of cut-off values from curves that were fitted from the observed data give more accurate estimations of the performance characteristics of an assay than traditional calculations from observed values. They also allow the calculation of optimal cut-off values taking into account the prevalence of B. ovis infection and give additional information about the performance of the assay at cut-off values varied according to the epidemiological situation.