Using evaluations of diagnostic tests: Understanding their limitations and making the most of available evidence

Diagnostic tests are commonly used in cancer medicine to screen for, diagnose, stage and monitor the progression of disease. An increasing variety of imaging and laboratory tests are available to oncology, all of which add to the diagnostic information obtained from the signs, symptoms, history and clinical examination of each individual patient. Assessments of the reliability, accuracy and impact of these tests are essential to guide optimal test selection and appropriate interpretation of test results. To make sense of a diagnostic investigation a clinician needs to be able to infer the probability that a patient has the disease in question according to the result of the diagnostic test. Test results rarely make a diagnosis 100% certain, but they may provide enough evidence to rule in or rule out a diagnosis in a pragmatic manner. That is to say, a test may only need to establish that the probability of the disease is above a known threshold for the benefits of treating patients as if they had the disease to outweigh on average the consequences of not treating them [1]. Within cancer medicine and other medical specialities, diagnostic tests can be divided into two groups: those that provide a numerical measure, and those that report a (human) perception [2]. For example, biochemical assays or cell counts yield numerical results, the absence or presence of a feature in an image is a human perception. The issues discussed in this paper apply to both circumstances. To facilitate their application to diagnostic decision making, numerical results are usually reduced to a set of categories. Most often results are divided into just two categories, one corresponding to a positive diagnosis, the other a negative diagnosis. However, there are often problems in deciding where the division between the categories should be. Occasionally three or more categories are used [3]: the benefits of doing this will be explored later in this article. Interpretation of images and other perception-based tests can similarly be viewed as a categorisation of results. Again the simplest case is when two categories are used, relating to the presence or absence of a feature, but occasionally results may be reported as three or more categories reflecting the degree of certainty with which a feature is perceived to be present or absent. Correct application of a diagnostic test depends on appropriate evaluations having been undertaken in relevant clinical settings. This paper first outlines the types of studies that are commonly used in the evaluation process. It then concentrates on evaluations that compare a new test with a reference standard, first by considering the biases these studies are susceptible to and secondly by considering the application of the results of evaluation from gynaecological cancer research.