Lab Tests and False Positives

[Kluver_Bucy]

Advertisement - Members don't see this ad

If you order a lab test with 25 different tests like a complete metabolic profile. What is the chance one of the test result is a false positive? I don't even now where to start!

 

[doc05]

it's very high. remember that the "normal range" only accounts for ~95% of the population, so you will almost certainly have 1 or more false positives if you order a set of 25 labs.

 

[Mumpu]

It depends on the sensitivity of the individual tests. A 5% false positive (95% sensitivity) over 25 tests gives you a 25 x (0.05) = 125% chance of a false positive. Bear in mind that this is still statistics -- you may well have 25 accurate values.

The rule of thumb is to always correlate labs with the patient.

 

[doc05]

It depends on the sensitivity of the individual tests. A 5% false positive (95% sensitivity) over 25 tests gives you a 25 x (0.05) = 125% chance of a false positive. Bear in mind that this is still statistics -- you may well have 25 accurate values.

The rule of thumb is to always correlate labs with the patient.

where did you study statistics? 25 x 0.05 is the wrong calculation.

 

[Kluver_Bucy]

I'm screwed. One of the 3rd years at my school had this exact question on his step 1, and he didn't know how to do the calculation. It seems unfair to get this kind of question on a licensing exam, but it is fair game. Somebody please solve this problem because one of us will get this question. Thanks a bunch to the smarty on SDN that knows the answer.

 

[Janders]

I'm screwed. One of the 3rd years at my school had this exact question on his step 1, and he didn't know how to do the calculation. It seems unfair to get this kind of question on a licensing exam, but it is fair game. Somebody please solve this problem because one of us will get this question. Thanks a bunch to the smarty on SDN that knows the answer.

First off, we wouldn't use sensitivity here. Sensitivity would be used for false negative.... sensitivity tells you how often the test will come up positive IF the patient IS positive.

We want to use specificity, which tells you how often a test comes up NEGATIVE IF the person is NEGATIVE. Lets assume 95% specificity.

If you are 95% specific, and run 25 tests you have 0.95^25 chance of getting them all right (0.95 to the 25th power). this comes out to about 0.277, so 27.7% that all tests are correct. 72% chance you get at least one error!

(at least, I think this is how it works! 😉 )

 

[Horseradish99]

First off, we wouldn't use sensitivity here. Sensitivity would be used for false negative.... sensitivity tells you how often the test will come up positive IF the patient IS positive.

We want to use specificity, which tells you how often a test comes up NEGATIVE IF the person is NEGATIVE. Lets assume 95% specificity.

If you are 95% specific, and run 25 tests you have 0.95^25 chance of getting them all right (0.95 to the 25th power). this comes out to about 0.277, so 27.7% that all tests are correct. 72% chance you get at least one error!

(at least, I think this is how it works! 😉 )

How could they give us a question like that if it requires you to calculate something to the 25th power? I may be smart, but I'm not THAT smart! 😛

 

[Mumpu]

Haha, you guys are right. Serves me for staying up until 2 am. 🙂

Doc, I was thinking of repeated t-tests on the same set of data. Janders is right, it's to the 25th power. And yes, specificity.

(hangs head in shame)

Horseradish, it's only one question out of 350. Don't sweat it.

 

[dynx]

If you order a lab test with 25 different tests like a complete metabolic profile. What is the chance one of the test result is a false positive? I don't even now where to start!

I may be over thinking this and I would really have to see the question to know how I would answer it but I'll put my two cents in anyway. First off, for a metabolic profile you would be using continuous data not catagorical so if 0 or unable to determine was an option I would probably lean heavily towards choosing that. For disease states requiring interpretation of continuous data with rigorous cut offs (EX. drug tests...positive if above a certain level negative if below) you can interpret false positives etc since the disease state may not be present even at the level above the cutoff. For cholesterol....a cholesterol of 300 (meassured level) is high cholesterol, end of story. That is, the dissease state is present BY DEFENITION if the meassure is that high. The meassurement may be f-ed up or a lab artifact but thats not the same as false positive. If you had to choose a value I believe you would have to be given the false positive rates for each of the test and add them together. You are looking for the probablility of one test showing up FP OR another test and so on and so one...the sum of the individual odds of a FP. To approach it from the other side...what are the odds that this one is correct AND this one is correct you could take the individual odds to the 25th power as another poster said but something tells me that adding them all up is a lot easier.