I am evaluating the validity of a test by calculating its sensitivity, specificity, PPV, and NPP. I was asked to calculate Youden’s index (J) by this formula:
J=[sensitivity(c)+specificity(c)–1]
I am seeking the usage of the Youden's index. What information does Youden index add to my knowledge of the validity of the test ?
Thank you.
A value of 1 indicates that there are no false positives or false negatives, i.e. the test is perfect. Is it true for a value of -1?
Thanks Wolfgang
No. all the indices are .25-.55. Does that mean that the tests are weak?
Thank you
How can I maximize the Youden index of my test ? Is it something more than: [sensitivity+ specificity] - 1?
Is there any possibility to Youden index -1.58?? if this, what is the meaning? Wolfgang Reichl, Foroughossadat Mortazavi
Hi Wolfgang,
I agree that a J = .50 is good. Might you have a reference for this?
Thank you!
Nichole
Hello all,
I could be wrong, and I would appreciate responses from others regarding this, but my sense is that a J value of .50 indicates that a diagnostic or screening test is completely worthless; it would deliver results that are no better than tossing a coin.
Again, I could be wrong, and I have searched for information about this, and have found discrepant information, but the more I read the more I am convinced that a J value (or area under the curve in an ROC analysis) is increasingly better as it moves from .5 toward 1.
I have seen differing categorizations, but it seems that values of .90 to 1 should be regarded as extremely good, between .80 and .90 as quite good, between .70 and .80 as barely satisfactory, between .60 and .70 as poor, and < .60 as worthless. (Sorry, Nichole. Please correct me if I am wrong.)
I would welcome responses to this view of things. I am by no means 100% confident. In particular, I would like people to comment on whether J values correspond to area-under-curve values in an ROC analysis.
With J values that are less than .50, I suspect that the analysis was done "upside down", with people NOT having a condition being "flagged" rather than those who had that condition. That would be no great deal, and, under those circumstances, values that came closer to zero, away from .50, would be more desirable.
Again, I would appreciate comments concerning this perspective. I am regarding ResearchGate as a venue for an exchange of ideas and advice.
Robt.
Hello again,
I must apologise, but I was not thinking clearly when I wrote the post immediately above. I was not thinking clearly about the Youden index when I equated it with areas under the curve (AUCs) in an ROC analysis. They are not parallel. I think that the values I suggested are probably appropriate for AUC, but definitely not for the Youden index, which can range from zero to 1.
Although it is almost definitely better to have a higher Youden index than a lower one, I have a hunch that global metrics such as the Youden index could mask some important features of a test such as a major deficiency in either sensitivity or specificity.
But, again, I would appreciate others' views about these things.
Robt.
It is a single statistic that captures the performance of a dichotomous diagnostic test and often used in ROC analysis & Instrumentation. The J= 0.50 and more indicates valuable diagnosis.
Refer to the below Refs:
Youden, W.J. (1950). "Index for rating diagnostic tests". Cancer. 3: 32–35. doi:10.1002/1097-0142(1950)3:13.0.co;2-3.
Schisterman, E.F.; Perkins, N.J.; Liu, A.; Bondell, H. (2005). "Optimal cut-point and its corresponding Youden Index to discriminate individuals using pooled blood samples". Epidemiology. 16: 73–81. doi:10.1097/01.ede.0000147512.81966.ba.
- Badges
- Science topic
- Similar topics
- Philosophy
- Philosophy of Mathematics
- Informality