The economics of diagnosis

Any population can be divided into two groups, one with the presence of a given disease or condition, and the other without. Diagnosis consists of using tests to sort the population into these groups. Diagnostic tests use a threshold value of a diagnostic variable to distinguish between disease-positive and disease-negative individuals. The analysis of error in diagnostic tests has typically been undertaken using receiver-operator characteristic (ROC) curves. More recently, economic value of information (VOI) methods have characterised the costs and consequences of testing. This paper develops a new method for economic test evaluation, which we call ROTS analysis. The ROTS curve plots the costs and effects of changing test thresholds, in cost-effectiveness space. We illustrate the use of our method with a worked example, and show how it can answer three key questions: (1) Is there any test that is worth doing? (2) What is a test's optimum operating point in terms of sensitivity and specificity? (3) If two tests are available, which is best? We contrast the merits of our method with those of established ROC and VOI analysis. We argue that ROTS analysis more clearly reveals the link between changing test thresholds and the cost-effectiveness of different treatments.