Hello Everyone,
I am using STATA v10. I am performing a meta-analysis of diagnostic test accuracy. When there are 4 or more studies, I've been performing a bivariate analysis using the 'midas' command without difficulty. However, when this model fails to converge, or when I have three studies, then I've been using 'metan' to determine the logit sensitivities and specificities with their respective 95% CI. This works very well but I need to calculate the LR+ / LR- / DOR and I2 along with their 95% CI. I can calculate LR+/LR- and DOR manually, but not sure how to calculate their respective 95% confidence intervals.
Here is an examples for the sensitivity point estimate:
list studyid tp fp fn tn if parameter2==4
+-----------------------------+
| studyid tp fp fn tn |
|-----------------------------|
20. | 20 9 8 17 31 |
21. | 21 4 2 6 9 |
22. | 22 50 42 15 43 |
+-----------------------------+
metan b1 se_b1 if parameter2==4 , randomi nograph z /* summary logit sensitivity */
Study | ES [95% Conf. Interval] % Weight
---------------------+---------------------------------------------------
20 | -0.636 -1.444 0.172 34.34
21 | -0.405 -1.671 0.860 29.09
22 | 1.204 0.627 1.781 36.57
---------------------+---------------------------------------------------
D+L pooled ES | 0.104 -1.239 1.447 100.00
---------------------+---------------------------------------------------
Heterogeneity chi-squared = 15.25 (d.f. = 2) p = 0.000
I-squared (variation in ES attributable to heterogeneity) = 86.9%
Estimate of between-study variance Tau-squared = 1.1972
Test of ES=0 : z= 0.15 p = 0.879
. di invlogit(.104) /* sensitivity point estimate */
.52597659
. di invlogit(-1.239) /* sens LCI */
.2246101
. di invlogit(1.447) /* sens UCI */
.8095363
The same is easily done with the specificity. I can then use the sensitivity and specificity to generate the LR+ and LR- and diagnostic odds ratio.
However, I do not know how to calculate the 95% confidence intervals for the LR+ / LR- or DOR. Furthermore, I do not know how I can calculate the 95%CI for the I-squared statistic.
Any help would be greatly appreciated.
Don Griesdale
I am using STATA v10. I am performing a meta-analysis of diagnostic test accuracy. When there are 4 or more studies, I've been performing a bivariate analysis using the 'midas' command without difficulty. However, when this model fails to converge, or when I have three studies, then I've been using 'metan' to determine the logit sensitivities and specificities with their respective 95% CI. This works very well but I need to calculate the LR+ / LR- / DOR and I2 along with their 95% CI. I can calculate LR+/LR- and DOR manually, but not sure how to calculate their respective 95% confidence intervals.
Here is an examples for the sensitivity point estimate:
list studyid tp fp fn tn if parameter2==4
+-----------------------------+
| studyid tp fp fn tn |
|-----------------------------|
20. | 20 9 8 17 31 |
21. | 21 4 2 6 9 |
22. | 22 50 42 15 43 |
+-----------------------------+
metan b1 se_b1 if parameter2==4 , randomi nograph z /* summary logit sensitivity */
Study | ES [95% Conf. Interval] % Weight
---------------------+---------------------------------------------------
20 | -0.636 -1.444 0.172 34.34
21 | -0.405 -1.671 0.860 29.09
22 | 1.204 0.627 1.781 36.57
---------------------+---------------------------------------------------
D+L pooled ES | 0.104 -1.239 1.447 100.00
---------------------+---------------------------------------------------
Heterogeneity chi-squared = 15.25 (d.f. = 2) p = 0.000
I-squared (variation in ES attributable to heterogeneity) = 86.9%
Estimate of between-study variance Tau-squared = 1.1972
Test of ES=0 : z= 0.15 p = 0.879
. di invlogit(.104) /* sensitivity point estimate */
.52597659
. di invlogit(-1.239) /* sens LCI */
.2246101
. di invlogit(1.447) /* sens UCI */
.8095363
The same is easily done with the specificity. I can then use the sensitivity and specificity to generate the LR+ and LR- and diagnostic odds ratio.
However, I do not know how to calculate the 95% confidence intervals for the LR+ / LR- or DOR. Furthermore, I do not know how I can calculate the 95%CI for the I-squared statistic.
Any help would be greatly appreciated.
Don Griesdale
Comment