Hello, everyone.
I am studying the influence of one biomarker on multiple disease characteristics, and I would like to calculate its cutoff point. I created univariate ROC curves to investigate the diagnostic accuracy of this biomarker for each characteristis, and it's obviously that I have obtained different cutoff values. Could I combine this AUCs in order to find one general, optimal cutoff point? (something like multivariate analysis with multiple outcomes?)
Thank you in advance for you help!
Julia Rogovaia You could consider using Youden's J-statistic as the best cut-off that maximizes both sensitivity and specificity.
As @Jan Lisec pointed below, you would have several values of J. What you need is max ( Abs (J)) that maximizes both.
Hi,
If I understand you correctly, than you do obtain multiple cutoff values -- one for each chracteristic -- for a given sensitivity/specificity. Combine these cutoff's into one would mean that for every characteristic you will effectively change sensitivity/specificity. You now have to define what 'optimal' means for you. :) Do you want to maximize sensitivity for all characteristics? Do you want to weight characteristic A with 90% and B/C with 5% each? Something else?
Bests, Jan
Dear Julia,
I totally agree with Jan Lisec , without a well defined aim you can not choose an optimal cutoff aggregation method.
Ákos
Ákos Bede-Fazekas , Jan Lisec , Ashutosh Karna thank you all so much!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R