Clinical utility of diagnostic tests

The clinical utility of a diagnostic test is its capacity to rule diagnosis in and/or out and to make a decision possible to adopt or to reject a therapeutic action. It can be integrated into clinical prediction rules for specific diseases or outcomes.

Factors determining the utility

A. Association between test results and disease is a must.

B. The pre-test probability of a disease

C. The demand of the testee in regard to the post-test probability of disease according with the given test result in order to rule in or out the disease and to accept or reject a particular therapeutic action.

More explanation of the factors mentioned above

In regard to A. In the case of no association the post-test probability of disease is independent of the positivity or negativity of the test result and is always equal to the pre-test probability. In other words, the test result does not change the degree of (un)certainty of presence or absence of the target disease: the test is useless. If association occurs the degree of post test probability of disease increases with positive test results and decreases with negative ones. Association increases the degree of certainty by which the hypothesis of the disease can be adopted given a positive test result and by which the hypothesis can be rejected given a negative test result.

In regard to B. The pre-test probability of disease influences the post-test probability. Pre-test probability is the probability that a person suffers from a disease before the test is executed. A high pre-test probability will tend to allow (much) easier to confirm the hypothesis of the presence of the target disease, a low pre-test probability of disease will tend to allow (much) easier to accept the hypothesis of absence of the disease.

In regard to C. It is the user of the test (or the testee) who determines which degree of certainty that is needed to decide to the presence or absence of the target disease and/or to take the adeaquate decisions in regard to therapy. Thus it is possible that someone is of opinion that the test result is useful while another thinks that the test (on its own or in the given combination of other test results and/or data) is useless.

Conclusion

Out of B and C an important conclusion can be deducted. It is not because the association between a test result and the presence or absence of the disease is weak that the test is necessarily useless since a high pre-test probability and/or a ‘low’ degree of demanded certainty by tester and/or testee can make a test with a weak relation (a modest likelihood ratio) to the target disease useful (allow to decide to the presence or absence of the target disease given a positive test result despite the weak relationship). An analogue reasoning can be made for ruling out the target disease.

On the other hand a test result on a test showing a strong association with the absence or presence of the target disease can be useless because of a ‘low’ pre-test probability and/or a too high demand for the degree of certainty. An analogue reasoning can be made for ruling out the diagnosis of the target disease.

Every test that shows an association between test results and the target disease is potentially useful. If it is not on its own thought to be useful then combination of it with other test results and/or data can potentially lead to a post-test probability that is thought to be high enough to rule the diagnosis in or low enough to rule the diagnosis out.

Tests can be useful to rule disease in or out or to rule both disease in (positive test result) and out (negative test result)

An example

A formula for the calculation of the post-test probability of disease is given by:

NK = PR*LR/(PR*(LR-1)-1)

Wherein NK = post-test probability of disease and PR = pre-test probability of disease and LR = likelihood ratio.

Let PR = .1 and LR+ = 10 and our demand for certainty = 95% then the post-test probability equals 52.6% and this is far from sufficient to accept the hypothesis of the presence of the target disease. Otherwise let PR = 90% and LR+ = 3 then NK = 96.4% what suffices to accept the presence of the target disease since the degree of certainty thought to be sufficient was 95%. Although a LR+ = 10 points to a much greater association between a positive test result and the presence of the target disease than a LR+ = 3, an LR+ = 3 can suffice for ruling a disease in while it is possible that a LR+ = 10 does not suffice. If the demanded degree of certainty should have been as high as 97% then both pre-test probabilities and LR’s should not have been sufficient to rule the diagnosis in. In this example both the crucial role of the pre-test probability and the demand of the degree of certainty for the usefulness of a positive test result are illustrated.

Further reading

• Attia J (2003). "Moving beyond sensitivity and specificity: using likelihood ratios to help interpret diagnostic tests". Australian Prescriber 26 (5): 111–113. ISSN 0312-8008. http://www.australianprescriber.com/upload/pdf/issues/107.pdf#page=15. 
• Davidson M (2002). "The interpretation of diagnostic tests: a primer for physiotherapists". Australian Journal of Physiotherapy 48 (3): 227–232. ISSN 0004-9514. PMID 12217073. http://ajp.physiotherapy.asn.au/AJP/vol_48/3/AustJPhysiotherv48i3Davidson.pdf. 
• Deeks JJ, Altman DG (2004). "Diagnostic tests 4: likelihood ratios". BMJ 229 (7458): 168–169. doi:10.1136/bmj.329.7458.168. PMC 478236. PMID 15258077. http://www.pubmedcentral.nih.gov/articlerender.fcgi?tool=pmcentrez&artid=478236. 
• Pewsner D, Battaglia M, Minder C, Marx A, Bucher H, Egger M (2004). "Ruling a diagnosis in and out with “SpPIn” and “SnNOut”: a note of caution". BMJ 329 (7459): 209–213. doi:10.1136/bmj.329.7459.209. PMC 487735. PMID 15271832. http://www.pubmedcentral.nih.gov/articlerender.fcgi?tool=pmcentrez&artid=487735. 
• Muldoon MF, Manuck ST, Matthews KA. Lowering cholesterol concentrations and mortality: a quantitative review of primary prevention trials. BMJ.1990;301:309-14.