How to estimate sample size to meet specificity and sensitivity requirements

[#1]

Hi, I'm trying to calculate the sample size required for my test to meet the sensitivity and specificity requirements.

I have a binary test that detects the presence of bacteria in water, if it's positive bacteria is present, if it's negative it is not present.

I would like my test to have

92% sensitivity (with 95% CI of 90-94%)

and 

90% specificity (with 95% CI of 88-92%)

and I want the power to be 0.8

I want to know how many samples I need in order to meet this requirements, how many positive bacteria samples and negative bacteria samples.

How would I go about doing this calculation? I've had a look under DOE > Design Diagnostics > Sample Size and Power. But I'm still not clear on which I should be using and how.

Many thanks!

[#2]

Hi,

Considering that nobody has contributed yet, let me take a stab at your question. To the best of my knowledge, the Specificity and Sensitivity of a test do not depend on sample size. But, the precision of the estimates for Specificity and Sensitivity is dependent on sample size. Therefore, if I understand your inquiry correctly, you will not modify the Specificity or Sensitivity by adding more samples to your experiment.

However, if your question is about the mixture of positive and negative cases in your experiment, then you can easily derive those from the definition of Sensitivity and Specificity.

I hope it helps,

Best,

TS

[#3]

Yes, the Specificity and Sensitivity of a test will not change with the sample size. What I meant by the requirements is that the 95% CI are between the quoted values.

The more samples I test the more confident I can be that the "true" Specificity and Sensitivity of my test is close to the one I calculate. Does this make sense?

[#4]

Hi,

I don't believe that JMP has the direct capability to conduct this type of estimation but I found this public domain article that may help you:

https://ncss-wpengine.netdna-ssl.com/wp-content/themes/ncss/pdf/Procedures/PASS/Confidence_Intervals...

Best,

TS