Combining two ORs with the same control group

#1
Hi everybody. I am doing a meta-analysis of observational studies. As the disease I am working on has different stages, some of the studies reported ORs(95%CI) for each stage separately in comparison to a single (shared) control population. I wonder if we can combine these two ORs (and their 95%CI) given that they reflect two case group but a single control group. Some other info:
*The sample size for each case group is different
*we have the sample size for each of the groups but not the frequency of exposure (otherwise we could use just the raw data)

Any ideas?

Thanks
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Also, are the odds ratios calculated just based on the two variables or are they from a multiple logistic regression model controlling for other variables?
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
Exactly what data do you have, and do you know the program used? It might help to try and recreate the contingency table the best you can.
 
#5
Exactly what data do you have, and do you know the program used? It might help to try and recreate the contingency table the best you can.
I really appreciate your help. I am attaching the paper. Under table 4 I want the mild-moderate and severe ID to be combined for item "resuscitation required"

Thanks
 

hlsmith

Less is more. Stay pure. Stay poor.
#6
Two options:

Try to calculate this or contact corresponding author for exact numbers.

Calculation (forgive me, I am not the greatest at stuff like this): I will do the Mild group and let you do Severe group. I would be as exact as possible with decimal places to circum-navigate rounding errors.

You can log base e transform the odds ratio into its beta coefficient from the regression model: 1.56 = 0.4446858

Now you can get the Standard error from the confidence interval: log base e of 1.45 equals 0.3715636, now minus this from from the 0.444 number and divide by 1.96 to get 0.0372449, which is your approximate standard error.

Now do this for severe...
 

hlsmith

Less is more. Stay pure. Stay poor.
#7
I was next going to propose weighting them and getting a pooled odds ratio, however you have polynomial regression, so I am slightly unsure if that approach is applicable since it is the outcome that is trichotomized not the variable. However resuscitation is the outcome based on the groups.

Perhaps others may weigh in (like maartinbuis, who seems good in these matters).


See following: http://stats.stackexchange.com/ques...-pooled-odd-ratios-in-meta-analysis/9495#9495
 

hlsmith

Less is more. Stay pure. Stay poor.
#8
I threw loose numbers into the R package and got:

OR 1.58 (CI: 1.53, 1.64), but I did this very quickly since I am on my way out the door, good luck and I will check back later.
 
#11
Hi, thanks a lot. I did not have access to internet last day. I have just seen these. Let me try and I will come back to you soon!
Hi I did it. All I needed was the R command you sent me and how to calculate SE. Thanks a lot. I think it is fine. I also contacted several senior biostatisticians. So far only you helped me!
And BTW, I used the data on table 1 which came from univariate analysis.
 

hlsmith

Less is more. Stay pure. Stay poor.
#12
Good idea using Table 1, this may make these data better for inclusion with other studies in a meta-analysis.
 
#13
Good idea using Table 1, this may make these data better for inclusion with other studies in a meta-analysis.
A problem remains however. When I uses that method of weighting two ORs, it took into account the same control group twice which makes the OR overweight in the meta-analysis. I am working on it to find a solution although I will probably be unable to do so without data of 2x2 tables. Anyway thanks for everything and please share anything you think might help.