Specificity vs. Sensitivity

[fuzzywuz]

---

Members don't see this ad.

Hey all,

I'm hoping someone can explain the plot on P51 in FA 2011.

It's basically a 2 bell curves that overlap at the ends. The curve represents Disease or No Disease. X axis = Test results (positive/negative). Y axis = Frequency of test result (it bisects the X axis).

Anyway, FA indicates that point A (which is the peak of the No disease curve) represents 100% sensitivity. Why is that the case? The area under this particular curve represents true negative (excluding the overlap part). If sensitivity is TP, why would it represent 100% true negative?


Can someone pleaseee explain it to me!?

Thanks in advance!!!

 

[ih8mcat]

There's a really good biostats question on uworld that offers a really good explanation to what you're asking. It's been a month since I did it, so naturally I've forgotten already 😀

 

[fuzzywuz]

Yeah, I jsut had that question today! (Maybe we're talking about the same one) They offered a good explanation on the different parts of the graph. But they didn't present it the way FA did (w/ points A, B, C).

I can understand UWorld's explanation, but what really bugs me is the placement of A, B, C... especially A and C (which represents 100% sens and spec respectively). Still can't understand why... ><

 

[SirGecko]

So I think the problem you are having is that the lines are not indicating which bell curve to look at. They are cutoffs or test thresholds that you would set when designing a test. You look at both bell curves. What is to the right of the line represents a positive test. (and to the left is a negative test)

So: line a has all of the disease bell curve and half of the no disease bell curve. So 100% sensitive 50% specific. Line B has only half of the disease bell curve and none of the no disease bell curve. So 50% sensitive and 100% specific.

Hope that helped.

 

[lackli]

Hey all,

I'm hoping someone can explain the plot on P51 in FA 2011.

It's basically a 2 bell curves that overlap at the ends. The curve represents Disease or No Disease. X axis = Test results (positive/negative). Y axis = Frequency of test result (it bisects the X axis).

Anyway, FA indicates that point A (which is the peak of the No disease curve) represents 100% sensitivity. Why is that the case? The area under this particular curve represents true negative (excluding the overlap part). If sensitivity is TP, why would it represent 100% true negative?


Can someone pleaseee explain it to me!?

Thanks in advance!!!

Hopefully I understand what you're asking here.

The points A, B an C are referring to different cutoff points that you can use to for a diagnostic test, say PSA levels for prostate cancer. If you want to catch everyone who has prostate cancer, then you are going to set the cutoff really low, say at 4. Conversely if you set your cutoff really high, like a 10 for PSA, then you are catching fewer people with the disease, but are also saying fewer people without the disease are testing positive.

Point A:
Because the ratio of TP to FN is dependent on your cutoff, at a low cutoff you're going to have more TPs and fewer FNs. Remember sensitivity is TP/(TP+FN). If you have zero FN, then sensitivity is TP/TP=1. Another consequence of this is that all you're negative tests will also be true negatives.

Point C:
Just the opposite of point A. At high cutoff values you are going to no FP. Specificity is TN/(TN+FP). So everyone who is testing negative at this point is going to be a true negative and specificity=1.

Point B: best balance between the two; all things being equal you want the lowest area under the curve. Of course it also depends on other factors such as how bad the disease you are screening for is, how expensive the test is etc to set cutoff points.

 

[SlaveOfTCMC]

In simple terms (for first pass of this information):

100% sensitivity means you are trying to capture everyone who has the disease. You don't care if you get false positives, because the goal is to cast a wide net and get everyone who does have the disease.

"Kill em all and let God sort em out" kind of fits 100% sensitivity

100% specificity means you want to ensure that every result you get contains the actual disease. You don't care if you don't get them all. You simply do not want anyone who does not have the disease.


This is a very layman description of that graph of yours. I hope it helped, but do not rely on this only.

Rather, understand the statistical terminology as mentioned by the previous poster in order to get the firmest grasp of the subject

 

[hyrule]

I honestly never understood that graph
I just memorized what each area stood for, where 100% sensitivity and 100% specificity were... and that the smaller the "under the curves" the better

 

[fuzzywuz]

Hey thanks for the replies! It definitely helps!

 

[SlaveOfTCMC]

again, this is the reason why a screening test must have high sensitivity and a confirmatory test must have high specificity.