What do you mean by combined sens ans spec ? The diagnostic accuracy ?
See: Article Measures of Diagnostic Accuracy: Basic Definitions
Remember that you first have to optimize your cutoff for the diagnostic purpose, for example by using ROC curves and a "gold" standard.
Hi,
I calculated sensitivity and specificity of each biomarker, but I want to calculate the sensitivity and specificity after measuring both biomarkers.
Thank you in advance.
You could put the biomarkers into a logistic regression model: outcome ~ bm1 + bm2. The model can yield the probability of the outcome and from this a ROC curve and AUC can be obtained. Note, if the biomarkers are not normally distributed, you should log them or take their square root before putting into the model.
Hi John,
Thank you for your answer. I also read this way:
https://radiopaedia.org/articles/sensitivity-and-specificity-of-multiple-tests
Which is your opinion?
Best,
Maria
Hello Maria,
That kind of approach is "OK" if there are well established (eg in clinical guidelines) thresholds for each of the biomarkers. If you want to maximise sensitivity then combine then with an OR. If you want to maximise specificity then use an "AND". This will be most powerful if the biomarkers are both related to the outcome but not strongly related to each other (if they are perfectly correlated to each other then combining them should not make any difference).
However, I'd recommend only do this if you have no choice but to use established thresholds. There is much more power and likelihood of a good result to combine them in a model as I suggested [outcome ~ bm1 + bm2. & try outcome ~ bm1 + bm2 + bm1*bm2. Note bm1*bm2 is symbolic for an interaction term between the biomarkers).
John
Hi John,
Thank you very much for your thorough reaponse.
Best regards,
Maria
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication