Clearly, a test has to be requested and interpreted in the context of other findings [1]. However, it is also important to ensure that the test’s results are interpreted appropriately and in an ‘evidence-based’ way. For example, using a single cut-off point that gives ‘positive’ and ‘negative’ results may not be good enough as the number needed to treat (NNT) and thus the risk/benefit balance often changes with each actual value of the test used to select patients for treatment [2]. The NNT for each test result can be worked out by designing test-treatment randomised trials to study the effect of using different selection criteria and their resulting values on treatment outcome (and diagnostic criteria).

The specificity and likelihood ratio (equal to the sensitivity divided by ‘one minus the specificity) is useful for assessing the accuracy of tests for population screening. Such likelihood ratios are also advocated for calculating ‘posterior probabilities’ from ‘prior probabilities’ in diagnostic reasoning but this approach is of dubious validity [3]. In practice, clinicians use a ‘prior’ finding to suggest a differential diagnosis and then look for other findings that occur commonly in at least one of those differential diagnoses and rarely in at least one other. This gives a ‘sensitivity ratio’ or ‘differential likelihood ratio’ (not a standard likelihood ratio). There is a differential diagnosis ‘theorem’ that can be used to calculate diagnostic probabilities during such reasoning [4], including in decision analysis.

In order to practice evidence-based medicine properly, the diagnostic process leading to sensible selection of patients likely to benefit from treatment has to be understood more clearly, and especially if value based pricing of drugs [5] is to be used in the NHS.

References

1. di Ruffano LF, Hyde CJ, McCaffery JK, Bossuyt PMM, Deeks JJ. Assessing the value of diagnostic tests: a framework for designing and evaluating trials. BMJ 2012; 344: e686

2. Llewelyn D E H, Garcia-Puig, J. How different urinary albumin excretion rates can predict progression to nephropathy and the effect of treatment in hypertensive diabetics. JRAAS 2004, 5; 141-5.

3. Llewelyn H, Ang AH, Lewis K, Abdullah A. The Oxford Handbook of Clinical Diagnosis, 2nd edition. Oxford University Press, Oxford 2009, p760.

4. Llewelyn H, Ang AH, Lewis K, Abdullah A. The Oxford Handbook of Clinical Diagnosis, 2nd edition. Oxford University Press, Oxford 2009, pp754 – 772.

5. Hawkes N. UK government pushes ahead with value based pricing of drugs. BMJ 2011; 343: d4632

Competing interests: None declared