Effects of increasing cutoff to increase sensitivity

[MudPhud20XX]

---

Members don't see this ad.

Here is one of the Kaplan questions:

For most screening test, changing the cutoff value to increase the sensitivity of the test will also increase what property of the test?

A. Accuracy
B. Discriminant validity
C. Negative predictive value
D. Positive predictive value
E. Specificity

I chose D, but the answer is C.

So if you increase the cutoff, you are more certain that this person truly has this disease, so won't that increase the positive predictive value? I am confused : (

Many thanks in advance....

 

[Missorleans]

sensitivity is TP/TP+FN, so to change the cutoff to increase sensitivity would be lowering the cutoff value, decreasing the number of false negatives.
lowering the cutoff would decrease the number of false negatives, so since NPV = TN/TN+FN, NPV would also increase

i think the problem here is just that you read the question as "increasing the cutoff value" instead of just "changing the cutoff value"

 

[igcgnerd]

Changing the cut off to make a test more sensitive means lowering the threshold number of what would be considered a positive test.
This lowers the rate of false negatives. Remember that Negative predictive value is the likelihood that someone without the disease will test negative. By eliminating false negatives, you are increasing the true negatives. This in turn increases negative predictive value.
Conversely, changing the threshold of a + test to make more people + will lower the positive predictive value because you are increasing the number of false positives.

 

[MudPhud20XX]

Thank you!

So in diabetes, let's say that you want to make the fasting glucose serum test more sensitive, shouldn't you increase the cutoff like from 90 to 110? In that way, you will reduce the false negative, correct? Ah so that's why you affect the NPV I guess, right?

But then when I think of PPV and NPV, these are affected by prevalence so I feel like again changing the cut off to make the test more sensitive shouldn't affect any of these parameters. Sorry for the confusion : (

 

[MudPhud20XX]

sensitivity is TP/TP+FN, so to change the cutoff to increase sensitivity would be lowering the cutoff value, decreasing the number of false negatives.
lowering the cutoff would decrease the number of false negatives, so since NPV = TN/TN+FN, NPV would also increase

i think the problem here is just that you read the question as "increasing the cutoff value" instead of just "changing the cutoff value"

yup that completely makes sense! you decrease FN, thus affect NPV, but at the same time doesn't it also increase TP, thus may affect PPV?

 

[Missorleans]

yup that completely makes sense! you decrease FN, thus affect NPV, but at the same time doesn't it also increase TP, thus may affect PPV?

PPV = TP/TP+FP

yes, TP increases, but so does FP, so the denominator is still increasing even though the numerator is increasing too. If it follows the curves in that image, then FP will be increasing more than TP increases as you lower the cutoff value, so PPV would decrease.

 

[thehundredthone]

So in diabetes, let's say that you want to make the fasting glucose serum test more sensitive, shouldn't you increase the cutoff like from 90 to 110? In that way, you will reduce the false negative, correct? Ah so that's why you affect the NPV I guess, right?

If the criteria for a disease is to have high value, then by increasing the cut off, you're making it harder for diabetics to get their club class membership. This means that real diabetics who don't have enough sugar in their blood will get left out. Although that makes your club very specific about who it lets in, many also consider that it is very insensitive on your part.

By decreasing the cut off, you're making the test more inclusive. The unintended consequence is that you also pick up those without the disease. Basically it will pick up everyone who has the disease, and then some. But, and more pertinently, that also means that a negative test will mean that the person truly does not have the disease (low false negative). Thus sensitivity = TP/(TP+FN). The domain TP+FN includes only diseased individuals, i.e. the prevalence of the disease is not a factor.

Practically speaking, this is the value of sensitivity, to be able to have confidence in the negative result. This comes at the cost of the positive result, meaning that unless the specificity is also high, the positive test result only indicates that further testing is required to rule in/rule out the disease.

Now the NPV tells you the probability of the lack of disease with a negative test. We've already established that the value of sensitivity lies in the negative test, hence it becomes clear that the more sensitive the test, the better its NPV. The caveat is that the NPV domain includes true negatives, i.e. individuals without the disease. NPV=TN/(TN+FN). This is why disease prevalence influences the NPV. Even if the test is sensitive (low FN), if the disease is very common then the NPV will not be high since FN may be comparable to TN.