Test concepts

To be honest, these days, many patients come to the hospital for tests rather than for consultations, and for (self-prescribed) treatment rather than diagnosis.It seems to us that when we offer a test (instead of history taking and physical examination), the patients are more satisfied and many of them still hold the belief that the tests are invariably correct.

And this is not a small mistake to make.

When we look at tests, we look at the test result (e.g. a blood test), compared to the actual result (whether the patient actually have that particular condition).

condition    test	present	absent
positive	true positive	false positive	TP/(TP+FP) = PPV
negative	false negative	true negative	TN/(TN+FN) = NPV
	TP/(TP+FN)=Sn	TN/(FP+TN)=Sp

Where:
PPV = Positive predictive value
NPV = Negative predictive value
Sn = Sensitivity
Sp = Specificity


Sensitivity and Specificity are inherent to the test itself - it depends on the test itself, and the cutoff value chosen only. When different values are chosen as the cutoff for a test, the sensitivity varies with specificity, and we can plot a graph that is known as the "Receiver operating characteristic curve" (Wiki on ROC curves)

Sensitivity measures how good the test is at detecting the condition when the condition is present. Specificity measures how good the test is at ruling out the condition when the condition is absent.

The detail for the ROC curve isn't that important - but it is important to know that if you pick a value that makes the test more sensitive, the specificity will go down, and if you pick a value that makes the test more specific, the sensitivity will go down. The clinical implication that, if you want doctors to have very few misses in diagnosing a condition, doctors are going to do a lot more tests on everybody, and most of these tests will turn out negative.

Take the sample ROC curve above as an example, if you want the sensitivity to go as much as 90% (i.e. 90% of patients who have the disease the test will be tested positive) then the specificity will fall to 40% (i.e. only 40% of those who do not have the disease will test negative - or, put it the other way round, 60% of those who do not have the disease will test positive). These patients who are so unfortunate to have a false-positive result will be subjected to confirmatory tests, which are often more invasive and incur morbidity and mortality - for example, a 0.25-0.5% risk of death following a diagnostic coronary angiogram.

And then we come to the positive and negative predictive value. We talked about the sensitivity and specificity being inherent to the test and being related to the cutoff value chosen - the positive predictive value and negative predictive value depend very much on the prevalence of the condition (i.e. the number of people having the condition in the population)

For example, let's say we have a population of 1000 (e.g. in a secondary school) that we are going to test for drug use. There are, let's say, 20 students abusing the drug. And our tests are 95% sensitive while 95% specific.

condition    test	present	absent
positive	19	49	PPV = 27.9%
negative	1	931	NPV = 99.9%

We can see that the tests is inappropriate for this application - We are going to have 49 false positive among these 1000 students. Imagine the psychological trauma and the effect of labelling in these students while confirmatory tests are being done.

Imagine the same test being done in another school where drug problem is very severe - 200 students are abusing the drug:

condition    test	present	absent
positive	190	40	PPV = 82.6%
negative	10	860	NPV = 98.8%

The result looks much nicer compared with the previous one - with the same test, same cutoff (and thus same sensitivity and specificity). We can see that this test is much more suited to the application in this particular school.

What does all these mean to the readers, then?

Looking at it, human decision making is also a test (even if no laboratory resources are used), It means that, the higher the quality (i.e. the lower the miss-rate) you require from the doctor, the more tests will be done on you, and many of them will eventually going to be negative, and worse still, some of them are going to be false positives.