How Should Results from Clinical Tests Be Integrated into the Diagnostic Process?

Abstract

When deciding if something is likely, we gather and weigh evidence. For instance, in the middle of winter, on a cold windy day, with clouds on the horizon, one might consider how likely it is to snow. The odds are already quite high, given the prevailing conditions, but we know that one additional piece of information—whether it is above or below freezing—will make it more or less likely. This is an example of Bayesian reasoning. Now, we find a thermometer giving a reading of 4° C and yet it snows. An alert observer would have noticed that there was ice on the puddles and the thermometer had been in direct sunlight. This illustrates the importance of careful observation and considering the context of measurements. With increasing pressures on clinicians’ time, many rely on the output of devices to make diagnoses and forget to place test results in the context of what he or she knew before he or she ordered the tests. Instead of using tests as an aid to diagnosis, “clinicians often forget this . . . and order tests ‘to diagnose’ or ‘to rule out’ conditions rather than to corroborate or challenge a clinical hypothesis.” Devices cannot diagnose our patients’ conditions, but the findings they provide frequently alter the probability that a subject has a particular condition. Shah et al’s article is a welcome step towards a better understanding of how results from diagnostic devices should be interpreted—in this case, for glaucoma diagnosis. The output from such devices is a statistical description of where the subject lies in relation to a normative database, not a statement on the presence or absence of glaucoma. The presence of any disease should be determined by the clinician from a synthesis of all data available. Although clinicians do this all the time, it is in a relatively haphazard fashion. Bayesian statistics formalize the process to refine clinical diagnoses. To apply this approach, the clinician should decide the likelihood that the patient has the disease before ordering the test (pretest probability) and combine this with the knowledge gained from the test. Only then can the clinician decide if the test result has increased or decreased the probability of disease being present. Judging the probability of glaucoma before a test requires the assessment of risk factors and a clinical examination. For most patients, it is obvious that they either have or do not have glaucoma, so no diagnostic testing is needed (although many diagnostic tests are also used for monitoring and objective documentation of disease status). However, in clinically uncertain cases, results of diagnostic tests can alter dramatically the probability that glaucoma is present. A strongly negative result may so decrease the probability that a patient can be assured that he or she does not have glaucoma, whereas a strongly