1
$\begingroup$

$lnOR = -0.909 - (-1.707) = 0.798$ and its standard error is $SE(lnOR) = \sqrt{0.2182 + 0.3182} = 0.386$. The results of the adjusted indirect comparison suggest that x was less effective than y for (odds ratio $2.22$; 95% CI: $1.04, 4.72$).

How was the the confidence interval worked out in that example?

$\endgroup$
3
  • $\begingroup$ Note: $\sqrt{0.2182 + 0.3182} \ne 0.386$. The $0.386$ appears to be correct (since it yields the correct CI bounds), but something is off with the square-root part. $\endgroup$
    – Wolfgang
    Apr 16 '16 at 14:57
  • $\begingroup$ Can you explain what the numbers in your post refer to? $\endgroup$
    – mdewey
    Apr 17 '16 at 8:25
  • $\begingroup$ @wolfgang I think each one should have been squared first $\endgroup$
    – mdewey
    Apr 17 '16 at 13:56
1
$\begingroup$

The CI was computed with $$\exp[0.798 \pm 1.96 \times 0.386] = (1.04, 4.73)$$ (the discrepancy between the 4.72 and the 4.73 is due to rounding error).

$\endgroup$
5
  • $\begingroup$ Thank you, can you please tell me what a wide CI accompanying RR can indicate in a meta-analysis of trials? Would the inclusion of fewer studies in a meta-analysis yield wider CIs? $\endgroup$
    – Harose
    Apr 16 '16 at 15:51
  • $\begingroup$ It indicates that the parameter that is being estimated is estimated impressively. More studies will tend to lead to more precise estimates and hence a tighter CI (unless the addition of the studies introduces additional heterogeneity). $\endgroup$
    – Wolfgang
    Apr 16 '16 at 17:19
  • $\begingroup$ By the way, you talk about the RR (also in the title before I edited it) but the value you posted is a (log) odds ratio and not a (log) risk ratio. Different things. $\endgroup$
    – Wolfgang
    Apr 16 '16 at 17:22
  • $\begingroup$ Yes, I am carrying out a meta-analysis to estimate RR using Bucher's method. The above example was one I found on the net to help me implement it into my own analysis. However, the result I got has an extremely wide confidence interval (0.2 to 12.2) making me wonder where I am going wrong. $\endgroup$
    – Harose
    Apr 16 '16 at 17:45
  • $\begingroup$ How can one work out SE from given a confidence intervals? $\endgroup$
    – Harose
    Apr 17 '16 at 11:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.