|
Abstract:
|
In cases where a decision about binary categorization of disease status is made on the basis of measurement on a trait, the determination of the cutoff point for the trait is an important problem. There are several criterion considered in literature for selecting the diagnostic cutoff point for classifying a subject in to disease or disease-free group. Notable among these are the probability of misclassification criterion, sum of sensitivity and specificity maximization criterion, maximal chi-squared criterion and a distance minimization criterion. We provide a quantitative assessment of different methods of obtaining diagnostic cutoff points in terms of performance of the resulting diagnostic quality measures such as sensitivity, specificity, predictive value positive and predictive value negative assuming models for the distribution of the underlying trait and the measurement error.
|
