Sensitivity is not enough. Response to: Postoperative delirium portends descent to dementia

Editor—I welcome the thoughtful discussion by Aranake-Chrisinger & Avidan about testing for postoperative cognitive decline.¹ However, the penultimate paragraph is potentially misleading because it makes two common statistical errors. Firstly, there is confusion about the meaning and importance of test sensitivity. In addition to sensitivity, a test's utility needs to be evaluated in the context of its specificity and the disease prevalence. A test with 99% sensitivity can be clinically useless if it has poor specificity, or is used in a low prevalence population.²³ Their equating of a 50% sensitivity with a 50% positive predictive value is another commonly made mistake. The SNOUT (negative result in a sensitive test rules out the disease) acronym is unhelpful for the same reasons. Secondly, the article states that repetition of a test can help its sensitivity. This is only true if repeated testing produces results with some degree of conditional independence.⁴ There is no benefit from repeated testing if keeps giving the same result.

http://ift.tt/2ATwrrK