Sensitivity and specificity remain the statistical measures by which veterinary surgeons select their tests in an attempt to reach a confident diagnosis. These values are (usually) determined from ‘pure’ populations of animals with and without a particular disease using the gold standard reference test. In the ideal world sensitivity and specificity would both be 100%. This would mean that the test was perfect at both ruling out (excluding) and ruling in (diagnosing) a particular disease. Whilst this is desirable, in reality this is almost impossible to achieve. In reality what is achieved is a compromise between each measure. We will return to this concept in a future article.
Sensitivity represents the true positive rate – i.e. the number of animals correctly identified with the condition. Specificity represents the true negative rate – i.e. the number of animals correctly identified as not having the condition. Whilst applying sensitivity and specificity measures, we must remember that these have (usually) been determined in a pure population of affected and unaffected animals. This aspect must be considered in order to understand how these tests perform in particular clinical settings where the population is poorly defined……
Clearly when trying to diagnose a particular disease, we do not know whether it is present or absent, hence the reason for the test (!). Test performance becomes heavily influenced by the individual case and setting. The strengths of each measure must therefore be considered when attempting to interpret the test. So, to consider the terms slightly differently:
- sensitive tests are best at ruling disease out (SeNsitivity OUT – SNOUT)
- specific tests are best at ruling disease in (SPecificity IN – SPIN).
Therefore, if we want to use each test to the best of its capability we must refine our clinical approach and choose whether we want to exclude or diagnose the condition. This may seem obvious, however in many clinical situations vets are looking to diagnose a particular condition when they may be better excluding it. This is particularly true when seeing a case for the first time, as many different diseases are often being considered at this point. Unless a multitude of clinical signs are present at the time of first presentation (raising a very strong suspicion for one particular disease), a rule-out approach has the greatest strength.
It is therefore important to refine the clinical question that needs answering in order that the test result is reliable. In order to get the best out of the tests you are using it is desirable to know whether you should be trying to rule out a particular disease or rule it in. So using the little aide mémoires of SNOUT and SPIN, let’s use hyperadrenocorticism as an example:
If we consider dog 1 that is only suffering with PU/PD we would want to exclude it from our list of differentials as the likelihood of this being the cause is reasonably low when only one clinical sign is present.
If we consider dog 2 presenting with PU/PD, polyphagia, a pot-belly, lethargy and alopecia we would be highly suspicious of hyperadrenocorticism and therefore we would look to diagnose the disease not rule it out.
Under these situations, a sensitive test would be most appropriate for dog 1, whilst a specific test would be most suitable for dog 2 (SNOUT and SPIN).
We hope this helps with some aspects of your clinical decision-making and in selecting the most appropriate test for your case. This will help to avoid errors in result interpretation and ultimately misdiagnosis.