I have a medical diagnostic test $A$ and test $B$. Test $B$ is the current gold standard. Test $A$ has an excellent negative predictive value and if it is negative, test $B$ is not performed as it is invasive.

However if test $A$ result is positive above certain cut off, test $B$ is performed to confirm the results. I have about $600$ patients who had test $A$, and about $200$ patients who tested positive with test $A$ were then tested with $B$ to confirm the findings.

I want to test the accuracy of test $A$ against test $B$, for the $200$ hundred positive results on test $A$:

  1. How can I get the sensitivity/specificity analysis ? as I have positives & negatives for test $B$ but only positives for test (A).
  2. What statistical methods will be suitable to compare the accuracy of test (A) against test $B$, the gold standard.

1 Answer 1


Let's start with a cross-tabulation of both variables:

enter image description here

Sensitivity requires that you have the true positives (A) as well as the column total (A+C). Specificity requires that you have the true negatives (D) as well as the column total (B+D).

In your case, you have 200 patients who tested positive for diagnostic test A, which would be the row total (A+B). As it is, I don't think you can calculate anything other than the positive predictive value, which is the number of true positives (A) divided by the number of positive calls (A+B). You would need all 600 patients with both test results to assess the sensitivity and specificity of diagnostic test A.

  • $\begingroup$ Thank you very much for your reply, What you have said was my real apprehension that I wont be able do the desired sensitivity & specificity calculations as only the positives for test ( A) had test (B), and not the negatives. You have clarified my confusion about this point. is there any other statiscal analysis I can perform on the available data to compare the accuracy for test A against test B. $\endgroup$
    – IUHK
    Aug 15, 2018 at 23:03
  • $\begingroup$ Let me clarify a bit more, I am comparing the accuracy of Cardiac CT angiography(test A) against invasive coronary angiography (Test-B)in quantification of degree of stenosis. It is believed that cardiac CT over estimated the degree of stenosis . if the patients have >50% stenosis, they undergo invasive angiography. Thats why only 200 out of the 600 patient were included as they had stenosis greater than the cut off 50%, and needed invasive test (B), we want to see if the estimation by CT (test-A) was , confirmed by Test (B),i.e Accuracy of test (A) . what statistical option are available? $\endgroup$
    – IUHK
    Aug 15, 2018 at 23:14
  • $\begingroup$ You could examine how much test B is a good predictor of test A, using linear regression, the continuous value of test A as a dependant variable and the continuous value of test B as an independant variable. The coefficient would give you an estimate as to how much, in average, test A over or underestimates test B. If you don't have those continuous values, and only have dichotomous variables (i.e. variables coded 0 and 1, for under or over the cutoff values), I don't think you can do anything, as every participant has undergone test A, and therefore there is no variability. $\endgroup$ Aug 16, 2018 at 20:39
  • $\begingroup$ Thank you again for your reply; actually you have clarified a lot to me; i am not good at statistics. I spent a good bit of time collecting the data without taking an advice and pre planning the peject. It looks like i will be left with a garbage at the end but at least you clarified my confusion añd i wont be wasting any more time on this. I am indebted to you for that. My data is collected in categoties normal =0; minimal 1; mild 2; moderate 3; severe 4; the data colected as per patient; per artery and per artery segment ( 16 segments ) . $\endgroup$
    – IUHK
    Aug 16, 2018 at 21:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.