Читать книгу Cases in Medical Microbiology and Infectious Diseases - Melissa B. Miller - Страница 15

The clinical microbiology laboratory must balance the requirements of timeliness with those of accuracy.

As an example, consider the identification of a Gram-negative bacillus from a clinical specimen. If the organism is identified with the use of a commercially available identification system, an identification and an assessment of the probability of that identification will be made on the basis of biochemical test results and a comparison of these results with a database. So, if the result states that the organism is Enterobacter cloacae with 92% probability, the laboratory may very well report this identification. Assuming that the 92% probability figure generated by the commercial system is on target (many commercial systems do a worse job with anaerobic bacteria), this means that there is a probability of 8%, or about 1 time in 12, that this identification will be incorrect.

Certainly, it would be possible for the laboratory to perform additional testing to be more certain of the identification. The problem is that by doing so there would be a delay, perhaps a clinically significant one, in the reporting of the results of the culture. In some cases such a delay is unavoidable (e.g., when the result of the identification in the commercial system is below an arbitrary acceptable probability and manual methods must be used) or clinically essential (e.g., when a specific identification is required and the isolate must be sent to a reference laboratory for identification; an example is Brucella spp., which require prolonged therapy and are potential agents of bioterrorism).

Similarly, the methods most commonly used in the clinical laboratory for susceptibility testing are imperfect. The worst errors, from the clinical point of view, are those in which the laboratory reports an organism as susceptible to a particular antibiotic to which, in fact, it is resistant. In some cases, additional tests are employed to minimize the risk of this occurring. For example, in addition to standard testing using either an automated or a manual method, recommended susceptibility testing of Enterococcus includes the use of Mueller-Hinton agar in which the antibiotic vancomycin is present at a known concentration. Even if the results of the standard susceptibility testing indicate susceptibility to vancomycin, if there is growth of the Enterococcus isolate on the vancomycin-containing Mueller-Hinton plate, the organism is reported as resistant to vancomycin.

Unfortunately, very few such checks exist to correct erroneous bacterial susceptibility assays. In general, there is a delay in the ability of automated susceptibility methods to reliably identify newly described mechanisms of antibiotic resistance. As a result, manual methods are often required. The performance of automated susceptibility testing methods varies, and certain combinations of organism and antibiotic have an unacceptably high error rate. In such cases, backup methods, such as disk diffusion or MIC testing, should be employed. Laboratories with a significant number of susceptibility tests to perform commonly use automated susceptibility methods because of the labor-intensive nature of manual testing and the speed with which automated systems are able to report results—often in a few hours as compared with overnight incubation, as is the case with manual methods.

Diagnostic tests vary in their sensitivity and specificity. As an example, consider a hypothetical STI (sexually transmitted infection) clinic in which the rapid plasma reagin (RPR) test, a screening test for syphilis, is being evaluated in 1,000 patients with genital ulcer disease who are suspected of having primary syphilis:

		PRIMARY SYPHILIS
		PRESENT	ABSENT
RPR TEST RESULT	POSITIVE	420	60	Positive predictive value = 420/(420 + 60) = 0.88Positive predictive value = 88%
	NEGATIVE	220	300	Negative predictive value = 300/(300 + 220) = 0.58Negative predictive value = 58%
		Sensitivity = 420/(420 + 240) = 0.66Sensitivity = 66%	Specificity = 300/(300 + 60) = 0.83Specificity = 83%

On the basis of these data, the sensitivity of this test (the true-positive rate, calculated as true-positive results divided by the number of patients with disease) in primary syphilis is 66%. The specificity (1 minus the false-positive rate) is 83%. Note that in this high-prevalence population (the prevalence here is the total number of cases in which primary syphilis is present—640 divided by the total number of individuals, 1,000—and is thus 0.64 or 64%), the predictive value of a positive test is fairly good, at 88%. The positive predictive value of an assay varies with the prevalence of the disease in the population. This is a key point. An example of this in our syphilis serology example in a low-prevalence population will serve to illustrate the point.

The same RPR serologic assay is being used in a hypothetical population of octogenarian nuns, none of whom are or have been sexually active in at least 6 decades.

		SYPHILIS
		PRESENT	ABSENT
RPR TEST RESULT	POSITIVE	1	169	Positive predictive value = 1/170 = 0.006Positive predictive value = 0.6%
	NEGATIVE	0	830	Negative predictive value = 830/830 = 1.00Negative predictive value = 100%
			Specificity = 830/999 = 0.83Specificity = 83%

In this patient population, there is only one true case of syphilis, presumably acquired many years previously. The specificity of the test in this patient population is the same as it is in the individuals attending the STI clinic (in reality, it is likely to be different in different populations and also in different stages of syphilis). Because there is one case of syphilis, and 169 of the positive RPR results are false-positive test results, the positive predictive value in this patient population is only 0.6%. Clearly, this is a patient population in which the decision to test for syphilis using the RPR assay is not cost-effective.

In making a decision to order a specific test, the physician should know what he or she will do with the test results—essentially, how the results will alter the care of the patient. In a patient who the physician is certain does not have a specific disease, if the test for that disease has an appreciable rate of false-positive results, a positive test result is likely to be false positive and should not alter clinical care. Conversely, if the physician is certain that a patient has a disease, there is no good reason to order a test with a low sensitivity, as a negative result will likely be false negative. Tests are best used when there is uncertainty and when the results will alter the posttest probability and, therefore, the management of the patient.