Constructing a cut-off point for a quantitative diagnostic test.

This paper is concerned with the problem of constructing a suitable cut-off point for a quantitative diagnostic test from a sample of diseased and non-diseased individuals, when the specificity and/or sensitivity required for the resulting diagnostic classification rule are specified. Statistical procedures are discussed which yield an explicit cut-off point, together with the necessary confidence limits on the true specificity and sensitivity of the test operating at this point (by contrast to confidence limits referring to an unknown optimal choice of the cut-off point). Sample size formulae are derived, showing a relative efficiency of up to 1.5 for a new procedure introduced in this paper over the usual procedure based on tolerance limits.