Calculating sensitivity and specificity of a *battery* of tests

#1
Hello everyone! I need to calculate the specificity, sensitivity, positive and negative predictive values of a battery of neuropsycholocial tests..that is, right now there is no such thing as a GLOBAL score for this battery, it is just a collection of tests...but I need to be able to show those psychometric values to justify the use of this battery of tests over other batteries. The tests have different values:

Test 1. Score goes from -20 to 20 (20 is best)
Test 2. Score goes from 0 to 17 (17 is best)
Test 3. Score goes from 0 to 20 (20 is best)
Test 4. Has 4 subscores that go from 0 to 5 (5 is worse)
Test 5. Has 3 subscores: one goes from 0 to 5 (5 is best), one goes from 0 to 5 (5 is best), one goes from 0 to 15000 (seconds, 15000 is worse).

Any ideas will be helpful!! Thank you
 

vinux

Dark Knight
#2
Hello everyone! I need to calculate the specificity, sensitivity, positive and negative predictive values of a battery of neuropsycholocial tests..that is, right now there is no such thing as a GLOBAL score for this battery, it is just a collection of tests...but I need to be able to show those psychometric values to justify the use of this battery of tests over other batteries. The tests have different values:

Test 1. Score goes from -20 to 20 (20 is best)
Test 2. Score goes from 0 to 17 (17 is best)
Test 3. Score goes from 0 to 20 (20 is best)
Test 4. Has 4 subscores that go from 0 to 5 (5 is worse)
Test 5. Has 3 subscores: one goes from 0 to 5 (5 is best), one goes from 0 to 5 (5 is best), one goes from 0 to 15000 (seconds, 15000 is worse).

Any ideas will be helpful!! Thank you
define specificity, sensitivity, positive and negative predictive values
What statistical help you need?
 
#3
sensitivity = probability of a positive test among patients with disease
specificity = probability of a negative test among patients without disease

Probably I need a ROC curve to determine this, but I am having trouble figuring out how to put all the subscores together to calculate sensitivity and specificity.
 
#4
If I have undestood your problem right you need as many ROC curves as the tests you want to evaluate.

If the tests are testing for the same thing then you should choose the test that its corresponding ROC curve has the largest AUC. (Area Under the Curve), or at least say that it is the best.


Leo