To the Editor: White emphasised problems that can arise if medical decisions are overly reliant on the results of laboratory tests.1 He relates the case of a patient who, because of a peculiarity of her immunology, consistently produced a false-positive test result. In probabilistic language, the issue is are there risks of both random errors and patient-specific errors? It is important to distinguish between them. Suppose a test has a false-positive rate of 10%. If this is truly random error, the probability of two false-positive results in the same person is 1%, and the probability of three false-positive results is tiny. But if it is due to there being 10% of healthy people for whom the test is invalid and who consistently give a positive result, the probability of two false-positive results in the same person is 10%, and the probability of three false-positive results is 10%!

Major textbooks of medicine have excellent chapters on decision-making. These warn about limitations of sensitivity and specificity (eg, that data from the general population may not apply to people who have tested positive in screening). But, other than this, little is said about reasons for errors in testing, and the consequences for how sensitivity and specificity should be used. In most cases, the impression given is that errors occur completely randomly. However, it appears that White's example, in which repeated testing led to repeated errors, is not unique.

Lee2 writes as follows: "Suppose a low-risk patient has an abnormal lung ventilation–perfusion scan. Obtaining that same test result over and over will not truly raise that patient's probability of coronary disease further and further."

Perhaps Goldman3 had something similar in mind when writing, "It may be quite difficult to distinguish random laboratory errors from test results that might be falsely positive or negative because of coexistence of a process that can affect the test".

Lists of possible reasons4,5 for errors include both short-acting (eg, distracting external noise, and biochemical effects of foods recently eaten) and long-term (eg, physical handicaps, and demographic factors) influences.

I wonder if information about tests should routinely include separate random and patient-specific components of sensitivity and specificity. For example, it might be stated that a false-positive rate of 15% arises from 10% random errors and 5% patient-specific factors, or that a false-negative rate of 10% arises from 3% random errors and 7% patient-specific factors. This is the conclusion I have been led to by White's article.