Over- and Underestimation of the Sensitivity of a Diagnostic Malignancy test due to Various Selections of the Study Population

Abstract

In characterizing indicators of malignancy, sensitivity and specificity are commonly presented. These estimates, however, are dependent upon the structure of the basic data. A simple model is given, explaining why some sensitivity estimates do not give a realistic picture of an indicator. The model is based on the assumption of three categories of diseased individuals–those having the illness in such a form that the indicator is not sensitive at all, those for whom the indicator is of great practical value, and finally those who are so seriously ill that the sensitivity of the indicator is one hundred percent, but the diagnosis is apparent anyway. Applied to data concerning tumor markers, an estimate of a ‘future relevant sensitivity’ can be obtained. The model can also be used when comparing results from studies with different population structures.