2
$\begingroup$

Reversing the reference category

Hello! I am doing a meta-analysis. I explain the problem with an example: All the study included in the meta-analysis have reference B (OR) except one that has reference A (OR). For this study, unfortunately, I do not have the data available to calculate the Odds Ratio inverted directly from Table 2X2.

Is there any possibility of getting OR B vs A directly from OR A vs B?

Thanks in advance!

$\endgroup$
4
$\begingroup$

Using standard notation of:

  • $a$ being number of events in non-reference group
  • $b$ being number of non-events in non-reference group
  • $c$ being number of events in reference group
  • $d$ being number of non-events in reference group

we have

$$ OR = \frac{ad}{bc} $$.

Switching groups means switching roles of $a$ and $c$ and of $b$ and $d$, so

$$ OR_{switch} = \frac{bc}{ad} = \frac{1}{OR} $$.

For meta-analysis you probably need to re-calculate SE as well. Recall that

$$ SE(log(OR)) = \sqrt{\frac{1}{a}+\frac{1}{b}+\frac{1}{c}+\frac{1}{d}} $$

so

$$ SE(log(OR)) =SE(log(OR_{switch})) $$

For SE without logarithm, see this post and Maarten Buis answer. By delta method

$$ SE(OR) = SE(log(OR)) \cdot OR $$ so

$$ SE(OR_{switch}) = SE(log(OR_{switch})) \cdot OR_{switch} = \frac{SE(log(OR))}{OR} = \frac{SE(OR)}{OR^2}$$

If you wish to work exclusively on log-scale (as @mdewey suggested) it's even more straightforward:

$$ SE(log(OR_{switch})) =SE(log(OR)) $$

and since

$$ OR_{switch} = \frac{1}{OR} $$

we have that

$$ log(OR_{switch}) = log(\frac{1}{OR}) = -log(OR)$$

$\endgroup$
2
  • 1
    $\begingroup$ The OP might find it simpler to work exclusively in the log scale. Do you want to add edit that into your post? $\endgroup$
    – mdewey
    Aug 22 '17 at 15:34
  • $\begingroup$ @mdewey Edit done. $\endgroup$ Aug 23 '17 at 9:30

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.