Some theoretical considerations
End points: What is positive and negative?
Most modern commercial assays for chlamydial antigen or nucleic acid are quantitative tests, which output their results in the form of optical densities or other continuous variables. No test is perfect. For all tests there is a certain amount of background 'noise' arising from various technical factors such as hydrophobic and ionic interactions, the stringency of hybridisation, the 'proof reading' capability of nucleic acid polymerases etc. A crucial question is what is the end point , the lowest value considered to be an unambiguous positive? How wide is the underlying 'grey area' of uncertainty where repeat testing, perhaps using another method, is indicated? Nowadays this question is usually resolved by the kit manufacturers, who specify an exact protocol to be followed plus, from trials, the end point values for negative, uncertain and positive results which in their experience are likely to be the most useful practically. These values are not usually the statistical cut off, which is the mathematically ideal cut off, but represent a clinical cut off, which attempts to take account of overall medical and epidemiological factors. Factors to be considered include the clinical objective of the laboratory test and whether it is necessary to have a high positive predictive value, or a high negative predictive value. Clearly, as the manufacturer's clinical cut off is not tailored for the precise circumstances of an individual clinic, it is to a small extent a compromise tending to err on the side of safety. The real clinical cut off is influenced by the prevalence of the disease in the test population and by one's own judgement of the importance of false positives and false negatives.
The medical cut off is influenced by:
With modern nucleic acid based amplification tests this has become less of an issue because their sensitivity is of the order of 90% or more and their specificity is close to 100%. However the antigen detection enzyme immunoassays (EIAs) have a sensitivity in the order of 70% and poorer specificity. Thus sensitivity, specificity and their trade-off relative to the prevalence of disease in the test population becomes a real issue. Where the prevalence of disease in a test population is less than ~5%, the best modern tests with specificities approaching 100% are required if there is not to be an unacceptable level of false positives.
- the performance of the test
- the prevalence of the disease in the test population
- a risk analysis
Sensitivity and Specificity
In actual fact, we never know the true sensitivity and specificity of a diagnostic test, only an estimate for it based on sample populations at different centres. The more quality studies, the better the estimate of the true value. Thus the results of a study are only valid for a particular category of patient and a particular sampling site. The performance of a test on a urine sample is likely to be significantly different to its performance on a urethral swab sample. The relationship of an imperfect test to the truth [which only God knows] is given by the so-called 'truth' table:
Table 1: The Truth Table.
Sensitivity is the probability [a / (a + c) in the table] that a true positive has been correctly classified as positive by the test.
Specificity is the probability[d / (b + d) ]that a true negative is correctly classified negative by the test
A false negative is the probability [ c / (a + c) ] that a true positive has been wrongly classified
A false positive is the probability [ a/ (a + d) ] that a true negative is wrongly classified as positive.
Since humans are not God, it is necessary to take another test, or combination of tests, as the reference standard and there is much debate as to what is an appropriate "gold standard", see: Discrepant analysis. It is important to realise that it is IMPOSSIBLE to better the reference standard, whatever it is, as clearly shown in the paper of Chernesky et al., 1999, the results of which are tabulated in Table 2. It can be seen there is considerable opportunity caveat lector to influence the presentation of test results by the reference standard that is chosen, while only a few percentage points difference in specificity may have a dramatic influence on the positive predictive value of a test in populations with a low prevalence of infection [see: worked examples].
Table 2. Effect of choice of reference standard on the percentage sensitivity and specificity of various rapid diagnostic tests for chlamydial antigen in male urine or for urinary tract inflammation. The value for 100% in each cell of the table varied between 44 and 73 specimens. The manufacturer of the test is indicated in brackets. The data are from the paper of Chernesky et al., 1999.
| || |
| || || |
|POSITIVE || |
| ||NEGATIVE || |
MeWard April 2002.
NEXT: Some worked examples.
| Rapid Test
|| Chlamydiazyme® (Abbott) with blocking reagent
|| Ligase chain reaction (Abbott LCx®) confirmed by fluorescent monoclonal Ab
| Testpack® Chlamydia (Abbott)
| Surecell® Chlamydia (Kodak)
| Clearview® Chlamydia (Unipath)
| Leukocyte Esterase Dipstick
Chernesky, M., Jang, D., Krepel, J., Sellors, J. & Mahony, J. (1999). Impact of reference standard sensitivity on accuracy of rapid antigen detection assays and a leukocyte esterase dipstick for diagnosis of Chlamydia trachomatis infection in first-void urine specimens from men. Journal of Clinical Microbiology 37, 2777 - 2780. Full article
Registered users: Please add any comments below.