Abstract
A model consisting of overlapping Gaussian distributions of disease and reference values has been used to calculate the effects of analytical imprecision on the proportions of patients wrongly classified by a test. Using published data the model has been applied to the urinary excretion of 11-hydroxycorticosteroids in the diagnosis of Cushing's syndrome. At the lowest levels of imprecision encountered in hospital laboratories, doubling the imprecision increased false negatives (missed diagnoses) from 3·6 % to 4·1 % and false positives from 9·4 % to 11·2 %. Although improved imprecision may thus produce only marginal improvements of misclassification, there is no level of imprecision below which further improvement does not improve misclassification. It is suggested that the concept of an ‘acceptable level of imprecision’ should be replaced by the concept of costing improvements of imprecision, equating the benefits in terms of patient classification with cost in terms of additional resources.