Hallo Adeel,

Sensitivity and specificity are not generally related to the prevalence of the disease in the population considered, since these are properties of the diagnostic tool. But this does not necessarily mean that these values are the same in all populations since they may well be related to characteristics of the individuals. For instance, serum ferritin is a relable indicator of iron deficiency anaemia in younger adults but poor in the elderly, due to its generally higher levels and to the influence of other chronic disorders in older people.

Positive and negative predictive value, on the other hand, are intrinsically related to the prevalence of the disease. This relationship follows from Bayes theorem; higher prevalences lead to higher positive and lower negative predictive values, assuming that sensitivity and specificity remain constant.

I agree with what Jeremy Franklin wrote and couldn't have said it better or more succinctly.

It is not necessarily so. Characterization of two populations in some variables may be different. Such as different food pattern, lifestyle, weather and so on.

Indeed, time, place and person are the main determinants in the epidemiological status of the disease.

Sensitivity and Specificity are characteristics of the diagnostic test and as a result are not influenced by the prevalence of a disease in the study population. I just realized Jeremy has said it all.

Yes , but may be difference in prevalence of a disease in different study populations depends on different characteristics in its population or different in time of study

As Jeremy Franklin said, sensitivity and specificity theoretically are the same regardless of the population being studied. However, in practice, due to variability in the measure of tests or disease, the observed sensitivity and specificity may vary between populations.

PPV, NPV, NNT and NNH all will change with a difference in prevalence between populations.

If we use the same valid and reliable measurement in all populations, we can not assume that the null hypothesis is always true, the matter imply failure to rejection.

Thanks fellows for your valuable input. It was very helpful.

Generally, sensitivity and specificity are not affected by the disease prevalence but PPV and NPV are so. In addition, when variability is seen in two different studies, their power (sample size) and also technical considerations (i.e. how the test can be done in different methods for example, different companies for primers, different measurement techniques and also the level of expertise of the operators in two studies) have to be regarded.

Another thing that one sometimes sees in the literature is that the sensitivity and specificity of a test are evaluated in populations with severe disease or no disease - since researchers often work in specialized clinical settings. Sensitivity in particular is likely to be better when those with the disease have very severe disease - this is an issue especially with patient-rated scales such as screening scales where a positive result is defined by exceeding a cut-point. Severely ill people will always exceed those cut-points, but the resulting sensitivity may not generalize to populations where there is a broader range of severity.

I can't agree more with Amir. The same tool can be used by one person in different population and the results will be different.

Here, we are talking about the tools and the study population. When we choose tools, we care for the specicity and sensitivity of the assays, which can be applied for any study population. Normally , we do not choose tests to serve different study populations. lalitha kabilan

Dear Adeel

In theory, the sensitivity and specificity of a test are independent of the prevalence of diseased individuals in the sample in which the test is being evaluated.

In practice, however, several characteristics of patients, such as stage and severity of disease, may be related both to the sensitivity and specificity of test and to the prevalence, because different kinds of patients are found in high- and low-prevalence situations.

Using a test to screen for disease illustrates this point. Screening involves the use of the test in an asymptomatic population in which the prevalence of the disease is generally low and the spectrum of disease favors earlier and less severe cases. In such situations, sensitivity tends to be lower and specificity higher than when the same test is applied to patients suspected of having the disease, more of whom have advanced disease. (Clinical epidemiology, Fletcher, 4th edition pages: 42-43)

Sensitivity and specificity are not affected by change in the prevalence of the disease in the population. But positive and negative predictive values are influences by it.

Hello Adeel Ahmed Khan.

In my opinion, must of the important parts has ben said by Jeremy Franklin (Sensitivity and Specificity are not generally related to the prevalence of the disease. Positive and negative predictive values are intrinsically related to the prevalence of the disease), Younes Jahangiri Noudeh (give attention to the sample size, technical considerations and the expertise of the operator) and Lalitha Kabilan (the sensitivity and specificity are properties of the diagnostic tool, which will not be affected by the characteristics of the population).

Then I will suggest that consider always the surces of variability:

- Real differences among subjetcs (differences in the attribute)

- Different measurments instuments

- Expertise of operators

So, you can reduce variation by:

- Using the same screening tool

- Calibrating the operators

Mansour Rezaei sugest the effect of time, this can be a triky one, since thru time there culd be differences in the diagnosis criteria for certain diseases (criteria for diabetes in 1980's and 2000's), but still, if you perform an adecuate sampling procedure, this will not affect the sensitivity and specificity of the screening tool.

RC

This could be a tangential question, but could explain difference in prevalence in different populations:

If an adult with BMI > 30 is considered overweight in USA, would you use the same measure (BMI>30) for overweight prevalence in adults of Nigeria?

Jagdish, please repost your question as a different thread, Its an important question and there are lots of answers to it but not under this thread.

Complex question. Pity I cannot go in detail due to time constraints as I have dealt with this question too quite in depth!

Have a look, perhaps useful paper

http://www.jstor.org/discover/10.2307/3702427?uid=3738880&uid=2&uid=4&sid=21103924668333

and also look at term "spectrum bias".

Textbooks generally say that Se and Sp are property of the tool and should be standard in all populations/ various prevalence but in reality it is different and there are several proofs about it in the literature.

one more possibly useful paper: http://onlinelibrary.wiley.com/doi/10.1002/%28SICI%291097-0258%2819970515%2916:9%3C981::AID-SIM510%3E3.0.CO;2-N/abstract

Two more papers from meta-analysts of diagnostic accuracy, they are the people who had to deal with this issue as well. Perhaps more technical papers but i hope still useful for you:

http://biostatistics.oxfordjournals.org/content/12/4/710.short

http://onlinelibrary.wiley.com/doi/10.1002/sim.3627/abstract

Indrani, read my question again and read the main question for this thread. My question is connected with the original question on association/ or lack of between prevalence, sensitivity, specificity, and predictive values.

This topic raised a good question that was answered very well from a theoretical viepoint by all participants in this forum . Only Predictive Value Positive and Predictive Value Negative are afffected by prior-probability=prevalence of disease in the source population, not sensitivity and specificity. I 'd like to add just that maybe what the inquirer is seeing/observing is difference in the sensitivity (Se) and specificty (Sp) attributed simply to differences in the deploymen or statust of the screening/diagnostic tool/equipment. For example, he/she may have observed different levels of Se and Sp use of diagnostic tool A in populations X and Y because technicians applying diagnostic test had different levels of training of the tool/equipment (thus used it differently between populations) or tools/equipmente had different levels of callibration in the different population sites.

Jagdish, I think that I understand why Indrani said that. The second question that you asked, about the use of BMI cutpoint of 30 to identify obesity in a different population seems as though it touches upon the generalizability of the instrument to other population and may have less to do with the prevalence of obesity in that population. That instrument or measure itself is what you would use to define obesity in that population but how applicable is it? Might one population have significantly thicker frames and be more muscular (criticisms that people sometimes have when using BMI in some populations, like bodybuilders)?

It's a tricky distinction, though, so I see the confusion.

Julia, indeed there was some confusion (part of it could be the way I wrote my initial comment). I assume it is all clear now, Adjourned!

There is also the issue of spectrum bias in the assessment of validity of a measurement instrument. If you select a sample wth the wrong spectrum of disease you may get estimates of sensitivity and specificity that will not be generalizable to the populations in which you want to use the test. For example, a poor test may be highly sensitive in a population characterized by a too many people with severe disease or too many people with excellent health. The performance of the test may be better in this population than in a population where a larger proportion of people are more in the middle range.

I should have added an example to my comment above. Suppose that you wish to validate a depression rating scale and you choose a sample that is 50% volunteers (probably all of upbeat mood and energy) and psychiatric inpatients admitted for depression. A low qualtiy scale could probably distinguish quite well between these two groups and appear to have excellent sensitivity and specificity. Yet, when used in a more realistic spectrum of respondents it could perform quite poorly. This is usually called spectrum bias.

Just to emphasized what has been said by about all the contributors. Prevalence only affects the positive and negative predictive values, not necessarily the sensitivity and specificity. In addition, I think prevalence is a characteristic of a population, so one cannot just increase the prevalence of a population (except in hypothetical scenarios). Perhaps, the ideal question should do with changing populations with different disease prevalence