Alternative ways of estimating serological titer reproducibility

A quantitative measure of the reproducibility of serum antibody titers has recently been proposed (R. J. Wood and T. M. Durham, J. Clin. Microbiol. 11: 541-545, 1980). The measure advocated is "the probability that the maximum ratio of two distinct (integer) titers (obtained in the blind) on the same specimen will not exceed 2." This measure of the reproducibility of serological titers is considered to be a fixed probability for any given specimen and set of test conditions. Although it is a fixed constant during the time period of a study, there are alternative methods one might use to compute an estimate of it, using laboratory data. Four such methods of estimating test reproducibility are discussed and evaluated. The estimates obtained from the two principle methods are evaluated quantitatively by means of Monte Carlo computer simulation. The simulation results show that, from a given sample of replicate integer titers, these two principal methods yield estimates that are highly correlated. In addition, with moderate numbers of replicates (sample size) these methods provide estimates that are on the average properly directed at the true reproducibility values (that is, are essentially, unbiased), particularly when the true reproducibility of the test is at least 0.9. The reliability, or stability, of the alternative estimates is studied for selected sample size.