Understanding the illustrations

Portraying quantitative information so that it is both remembered and used is non-trivial. Conventional text and tables offload the data from the writer but aren’t ideal for the reader. Infographic displays work better, according to cognitive psychologists. Stylistically, editors and web designers prefer a standardised format, but much of the visual impact and retention value of infographics is lost if they all look the same. A compromise has been adopted here:

Diagnostic test performance is portrayed as bubble charts, plotting sensitivity against specificity, with bubbles the same size but with margin definition portraying evidence quality – sharp is high quality; the more blurred the lower the quality. Different tests will get different colour bubbles.

Likelihood ratios are a good shorthand description of test quality but are not intuitively easy for clinicians to use. The figure below illustrates how various levels of likelihood ratio will affect post test probability:

It is difficult to remember how tests (and that includes clinical signs, even aspects of the history) perform diagnostically – human brains are really not set up to memorise comparative sensitivities/specificities or likelihood ratios as numbers – they tend to blur into each other, and even more confusingly, different numbers from different sources make it particularly difficult to keep track. Again, a picture placing a test into context relative to other similar tests is often more helpful and easier to remember. Like this:

This is a familiar plot of sensitivity versus specificity: a test with high sensitivity is helpful if negative – it serves to ‘rule out’ a conditions (SNOut – negative with high sensitivity rules out), whereas a positive test with high specificity helps corroborate a suspected diagnosis (SPIN – specific test positive rules in). The problem is that studies measuring test performance are often small, with challenging gold standard comparators, and performance is heavily influenced by pre-test probability (prevalence often used as a surrogate for this.) In a balloon plot such as the above, the size of the circle could thus represent a confidence interval around either sensitivity or specificity, or (and this option is used here) the stepped likelihood ratio – big circle = LR+ of about 7-10 (shift probability by a very useful 40%); medium circle = LR+ of 3-6 (shift probability by 25 to 30%) and a small circle = LR+ of 2 – probability shift of 10 to 20%)