Editorial: Pre-Test Probability Matters – A Comment Regarding “Shotgun” Testing
Occasionally I receive calls regarding unexpected laboratory results. A typical query might involve an extensive workup for a potential infectious disease in which one of many agent-specific antibodies (usually IgM antibodies) ordered on a patient returns a positive result that is discordant with clinical findings. I am generally asked whether this could represent a false positive. Invariably, a thorough review of processes and procedures yields no evidence that a given test is operating outside expected performance characteristics. The process from there becomes an exercise in educated guessing: Is the result near the positive threshold or is it well beyond the threshold? Were the reference values for the day unusually low or high? Were there an unusual number of positive results for that particular run? Each of the questions might be potentially useful, but ultimately there is no magic formula for determining whether an individual positive result is true or not. What may be most useful is a thorough evaluation of the clinical scenario in which the test was ordered, and the statistical and epidemiologic issues surrounding the meaning of any given laboratory result.
The meaning of a “positive” test of any kind is dependent on a number of pre-test variables, including the clinical probability that a patient has the condition the test is being used to detect. This tenet is met with varying skepticism throughout the health care community, but remains an important principle in laboratory medicine. The most astute medical practitioners treat laboratory results not as yes-or-no diagnostic switches, but as ways to either increase or decrease the probability of a given disease relative to the probability that existed before the results were received.
These ideas may seem abstract or ill-defined (or they may seem like ways for laboratory directors to dodge accountability), but they are actually embedded principles, quantifiable by inferential statistics. Let’s take, for example, a test for which 1% of the general population tests positive regardless of clinical status (for instance from known rates of analytic interference of serum or plasma elements, or known rates of cross-reacting antibodies). If a patient’s clinical signs and symptoms place them in a group for which the prevalence of the disease is 20%, then the predictive value of a positive test (assuming, for the sake of argument, 100% sensitivity) is 96%. That is, 96 of every 100 patients in that group that test positive for disease will actually have that disease; this is likely to be a clinically useful result. Conversely, if a patient’s clinical signs and symptoms place them in a group for which the prevalence of disease is 1%, then the predictive value of a positive test is only about 50%. That is, only about half of the patients with a positive result will have the disease—a much less clinically useful result. This discordance is due to the fact that, given a consistent rate of false positive results for a given assay, as true positives decrease in the population, the probability of any given positive result being false increases. An extreme example would be a disease for which populational prevalence is zero. In that case, all positives are by definition false positives, and the predictive value of a positive result is zero.
There may be some utility in assessing for certain disorders even in low-probability settings, since the negative predictive value of a negative result is often quite high for very sensitive assays. However, uncertainty as to the meaning of a positive result in these settings might abrogate the usefulness of the negative result. Again, an informed approach from an astute clinician as to how the laboratory information will be used is essential.
There is a culture in medicine that resists the principle that a laboratory test (whether a routine chemistry panel, a next-generation molecular test, or a traditional surgical pathology evaluation) is a means by which to hone pre-test probability, rather than the magic answer to a given medical question. To our credit, the field of pathology and laboratory medicine has established such a high level of reliability for most tests that each result is treated as that magic answer, and deviations from perfection are attributed to some ill-defined “lab error” instead of to the natural performance characteristics of a given test within a given patient population.
Many laboratory tests enjoy “gold standard” status for the answering of many medical questions. Every test, however, is influenced by the populational statistics that underlie the disease for which testing is done. Consequently, every patient is best served by medical ordering practices that take into account the pre-test probability of a given medical condition. “Shotgun” testing doesn’t just drive up the cost of health care; it also measurably decreases the value of medical information.