[MUSIC] Okay, now to further illustrate the concept of predictive value depending on the prevalence, you can look at this table. Which shows that for a given sensitivity and specificity value of 90% and 95%, you will see that if the prevalence of the disease is 0.1%, 1%, 5%. Or if 1 out of 2 people have the disease, the predictive positive value will increase from about 2 to 94%. Which mean that if you use a test and you apply it to a really high risk individual, let's say, HIV test, in a population at high risk, for example, a drug users. You will trust the results of the test more in this population because of the high value of the predictive value, positive predictive value, than if you apply these same tests in the general population, or in a population in whom the prevalence of the disease is lower. So basically, there is a correlation, and we often relate prevalence, sensitivity, specificity, and predictive value using a formula that is expressed here. Which is basically illustrating how the three PPV, positive predictive value, sensitivity, and specificity, Se and Sp, are related to the prevalence of the disease, P. I will not go in detail into this formula, but if you know the prevalence of the disease, if you know the sensitivity and the specificity, you can calculate the predictive positive and negative value of the test. The determinants of the predictive value are the sensitivity, the specificity of the disease in the pre-clinical phase in the target population. Let's move now to the measures of association. I know how good my test is. Now in screening, you have to know how good It is made in the population. Meaning if I applied screening program, if I have applied a screening test, did the population have any benefit in terms of mortality or in terms of morbidity? We use, generally, two indicators of measures of association, either the relative risk or the odds ratio, depending on the design of the study. And in screening, and you will learn about the importance of using the appropriate outcome measures in different models discussing bias, but ideally, in screening, you should use mortality outcomes. This is the outcome or the measures that will be unbiased, That will not suffer from what we call over diagnosis bias, length or lead time bias, which are discussed in other modules. So the evaluation of screening must be based on measures of disease occurrence that will not be affected by what we discussed early diagnosis, except to the extent that early treatment is beneficial. You can use overall mortality, or you can use disease-specific mortality, for example, breast cancer mortality. In general, the screening programs may not have power enough to show overall mortality decreases, because you have competing risks. And most of the time, we may report disease-specific mortality, how much I decreased the mortality of breast cancer by using a breast cancer screening program. There are also, when we discuss measures of association, the same limitation or the same importance of study design can be discussed in screening. For example, IDD data should come from experimental studies, usually randomized controlled trials. There are many biases that can come from non-experimental study, and we, in general, rely on study designs that are RCTs, randomized control trials. [MUSIC]