Comparing the accuracy of two diagnostic tests applied on the same sample

#1
Hi there,

I have recently come across some data that compares the accuracy of two diagnostic tests. We have 214 samples which are tested positive using a "gold standard" test. Using diagnostic test A, 145 / 214 (67.8%) were tested positive, whereas using diagnostic B, 213 / 214 (99.5%) were tested positive. Out of the total of 214, only one sample was tested negative with both test A and B.

We have not used the two diagnostic tests on samples that were tested negative using the "gold standard" test.

I am a bit confused about what is correct statistical test to use here, but decided to vote against the following tests:
- Since both of these two percentages are drawn from the same sample, I cannot use a two-sample t test for percentage (the two samples are not independent of each other).
- I figured that I also cannot use a one-sample t test for percentage because the two percentages were not mutually exclusive of each other (i.e. as they add up to more than 100%).
- I considered using a Chi square test, but I don't think I could form an unbiased "expected" frequency here.

A bit more reading led me to consider the McNemar's test, which (to my understanding) allow repeated testing of the same sample. I put my numbers into GraphPad QuickCalc online and it gave me a p value < 0.0001. I am still unsure if I did the right thing; if anyone could confirm that McNemar's test was the correct test, or if not, suggest a better test to use, I would very very much appreciate it.

Thank you for your time in advance!
Nancy
 

hlsmith

Omega Contributor
#2
You never say it, but you are trying to compare the results from the two diagnostics on the same 214 patient sample, correct?
 

hlsmith

Omega Contributor
#4
In my head mcnemar seems wrong since it is typically used on a 2x2 table that has a stratum variable. But you don't know how the diagnostic functioned on the negative tests so you have a 1x2 with a stratum.