The second module, provided by Dr. Idris Guessous, will address the metrics of screening with concepts related to robustness, validity and impact. A quiz on screening metrics will complete this module.

Professor of epidemiology and public health and Director of the Institute of social and preventive medicine Lausanne University Hospital

Gillian Bartlett-Esquilant

Professor of Epidemiology and Research and Graduate Program Director and Associate Chair for the Department of Family Medicine at McGill University. University of Lausannne and McGill University

So we spoke about how robust a test can be.

Â

We discussed the validity of the test and before we move to the impact,

Â

I would like just to stress that

Â

the robustness and the validity of a test are actually related.

Â

And I said that how reliable the test is,

Â

is how wrong the answer will be no matter.

Â

But at least, it will be always wrong.

Â

This is how reliable the test is.

Â

Hopefully it will be correct,

Â

and this is what is used in sensitivity and specificity.

Â

So obviously, if you want to have a test that has high sensitivity and specificity,

Â

you have to have a test which is highly reliable.

Â

So of course, this is

Â

the common relation or the interaction between robustness and validity.

Â

Now, the impact.

Â

The impact of a test is measured using predictive value,

Â

either positive or negative.

Â

Again, here you see this way of

Â

expressing positive predictive value and negative predictive value.

Â

The positive predictive value would be the probability

Â

of having the disease given that the test is positive.

Â

Here, what is given is the result of the test.

Â

You have the information of the test.

Â

For sensitivity, you had the information on the disease.

Â

Disease is whatâ€™s given.

Â

Negative predictive value is defined as the probability, small p,

Â

of not having the disease, m-,

Â

and negative, given that the test is negative.

Â

You can use again a two-by-two table which

Â

is illustrated here to calculate predictive value.

Â

Again, here, what is given is the results of the test.

Â

And as you can understand,

Â

this is really what a clinician wants.

Â

What is given is the result of the test,

Â

is not the disease.

Â

He's using the information of the test to know the probability of the disease,

Â

which was exactly the opposite when we discussed sensitivity and specificity.

Â

Here, the probability of having the disease given that the test is positive would

Â

be simply calculated as the number of people in the Â«Â aÂ Â» cell,

Â

over the combined number Â«Â a+bÂ Â».

Â

What is given that I have this number of people with a test which is positive,

Â

a+b, the number that really have the disease,

Â

Â«Â aÂ Â», a over a+b.

Â

You can use now this table with data inside to calculate at home sensitivity,

Â

specificity and the predictive values, positive and negative.

Â

Just try to do it at home.

Â

You should end up with the same results expressed here.

Â

Again, try to start thinking about what is given.

Â

Is it the disease?

Â

Or is it the test which is given?

Â

In practice, we do have sensitivity,

Â

specificity, performance of predictive value for different tests.

Â

For example, FOBT in colorectal cancer,

Â

you can see that there are different range of performance and

Â

the positive predictive value in general of a FOBT is a range about 10%.

Â

Here, this is already the reason why you will have to use the test multiple times,

Â

to repeat the test,

Â

because the performance of the test itself

Â

once is not good enough to use it only one time,

Â

but you have to repeat the test multiple times.

Â

For predictive value, we can say that the greater the specificity,

Â

the greater the positive predictive value.

Â

And the greater the sensitivity,

Â

the greater the negative predictive value.

Â

We said that sensitivity and specificity do not depend on the prevalence of the disease.

Â

How much is the proportion of the disease in your population?

Â

Usually, this is fixed by people who run the test or construct the test.

Â

This is in contrast with predictive value

Â

that really depends on the prevalence of the disease.

Â

For instance, if you use this slide,

Â

you can see on the horizontal axis the prevalence of a disease from zero,

Â

no disease in the population, to 100.

Â

As you can see, the predictive value,

Â

either positive or negative,

Â

will really depend on using the same test with the same sensitivity,

Â

the same specificity, they will depend on the prevalence of the disease.

Â

The positive test predictive value will

Â

increase as the prevalence of the disease increases,

Â

and the negative predictive value will

Â

increase at the prevalence of the disease decreases.