Hi!
I was studying diagnostic meta-analysis by Jones et al (
Article Summary Receiver Operating Characteristic Curve Analysis Tec...
)and I tried to calculate the Q value from cited.
"Q is the intercept of the SROC and the antidiagonal line through the unit square. Its value indicates overall accuracy by finding where sensitivity and specificity are the same"
1. I started inserting the confuse matrix data and making a reitsma object from "mada" package
2. Then I gave it's value to the sroc() function with the return_function method set it to T in order to give me the generic function of sROC (let's call it f(x))
3. I then used uniroot() to find the (1-specificity) value where sROC and the anti-diagonal curve intercepts each other.
4. Finally, I gave the false positive rate back to f(x) and found the sensitivity, and so, the Q value.
Is there a more direct approach for this in R?
You can use the metabin package in R to solve for getting a common Odds Ratio, Tau sq, Tau, I sq and run a test of heterogeniety to get the Q value in the output.
Thanks @Jharna Katara, but the Q value I refer to is not the Q from cochrane heterogeneity, but the sensitivity value of a sRoc point where it's accuracy its at it maximum. It also can be refered as the interception of the sROC with the anti-diagonal arrow of the sensitivity vs (1-specificity) graphic.
I think the Q value if looking for the paper you sent is just the point you'll get from the meta-analysis, which is the pooled accuracy throughout possible thresholds of the test ?
I'm not sure I've catched the question correctly.
Maybe if you have many studies for different thresholds you could run several meta-analyses (1 for each threshold) and then deciding which is the optimal (first not differentiating costs of false negative and false negative (ie Se and S have the same value) as well as in a 2nd step incorporating misclassification cost term (using the ratio of cost btw FN and FP as well as several realistic prevalence scenari)).
- Badges
- Science topic
- Similar topics
- Biological Science
- Immunology
- Antibodies