Thanks for this excellent video! It's the only material than I found to understand how to do a meta-analysis with precalculated ORs and confidence intervals. Now, I want to know if you can indicate me how to performe outliers detections based on this results of metagen analysis. Thank u!
It depends. If the disease is rare, you can assume OR and RR are approximately the same. If the OR is reported from case cohort study, you can combine it with HR, RR, and OR. If the disease or condition is not rare, then I recommend using the conversion formulas to convert OR to RR or vice versa.
Thanks for this great resource. A few questions: 1) Is it possible to use metagen to calculate a pooled estimate from studies that only report an odds ratio and upper-lower 95% confidence interval (they don't provide standard error)? 2) Is it possible to pool adjusted and unadjusted ORs from observational studies?
if you use pre-calculated OR and upper-lower limits, do you have to take the log of these values or can they be inputted into the analysis without transformation?
@@beduboy1 Thanks for your Q. You don't necerrily need to calculate SE by yourself. The metagen accepts lower and upper CI values. It has lower and upper arguments. You just need to supply log(your_lower_interval), log(your_upper_interval) for lower and upper aruments then the point estimate also has to be passed in log(). That means for studies that reported SE, you can use SE. For studies that reported lower and upper, you can use log(lower) and log(upper).
Yes, you have to set backtransf = TRUE. Regarding your Q whether to pool adjusted and unadjusted together, it is your judgement. I usually do subgroup meta-analysis of adjutsed and unadjusted OR. The adjustmen factors are rarely identical. So I use my personal judgement to combine together. I call it Multivariable adjusted eventhough the adjustement factors are different. As I said, it is personal judgement and you may have to judge putting the advantage the synthesis into perspective. Good luck
Hello sir. While doing meta-analysis (reliability generalization) on R metafor, I am having trouble and I can't get over it. I want to do moderator analysis with a categorical variable (three-category) feature over "continent". For this, I use the following code; res
Based on what you described, I think the error is likely due to the three categories of the country variable in your model but you specified only two in your predict function. The model name in your predict function is also res4 buy your saved model is res. Pls check that. I cannot help you much unless I have a reproducible example.
@@r_pydatascience When I increase the two numbers (0,1) I specified in my predict function to three (i.e. 0, 1, 2), the analysis is performed. So how do I decide the order of these three numbers? (0, 1, 2) or (0, 2, 1) or some other order?
Thanks for this excellent video! It's the only material than I found to understand how to do a meta-analysis with precalculated ORs and confidence intervals. Now, I want to know if you can indicate me how to performe outliers detections based on this results of metagen analysis. Thank u!
Sure, I will do that. Thanks.
Nice video, can tou combine OR and RR in the same MA transforming the effect size to logarithmic scale?
It depends. If the disease is rare, you can assume OR and RR are approximately the same. If the OR is reported from case cohort study, you can combine it with HR, RR, and OR. If the disease or condition is not rare, then I recommend using the conversion formulas to convert OR to RR or vice versa.
Thanks for this great resource. A few questions: 1) Is it possible to use metagen to calculate a pooled estimate from studies that only report an odds ratio and upper-lower 95% confidence interval (they don't provide standard error)? 2) Is it possible to pool adjusted and unadjusted ORs from observational studies?
if you use pre-calculated OR and upper-lower limits, do you have to take the log of these values or can they be inputted into the analysis without transformation?
@@beduboy1 Thanks for your Q. You don't necerrily need to calculate SE by yourself. The metagen accepts lower and upper CI values. It has lower and upper arguments. You just need to supply log(your_lower_interval), log(your_upper_interval) for lower and upper aruments then the point estimate also has to be passed in log(). That means for studies that reported SE, you can use SE. For studies that reported lower and upper, you can use log(lower) and log(upper).
Yes, you have to set backtransf = TRUE.
Regarding your Q whether to pool adjusted and unadjusted together, it is your judgement. I usually do subgroup meta-analysis of adjutsed and unadjusted OR. The adjustmen factors are rarely identical. So I use my personal judgement to combine together. I call it Multivariable adjusted eventhough the adjustement factors are different. As I said, it is personal judgement and you may have to judge putting the advantage the synthesis into perspective.
Good luck
Hello sir.
While doing meta-analysis (reliability generalization) on R metafor, I am having trouble and I can't get over it.
I want to do moderator analysis with a categorical variable (three-category) feature over "continent". For this, I use the following code;
res
Based on what you described, I think the error is likely due to the three categories of the country variable in your model but you specified only two in your predict function. The model name in your predict function is also res4 buy your saved model is res. Pls check that. I cannot help you much unless I have a reproducible example.
You shoud check the examples detailed on the package. Use ?predict.rma and scrol down at the bottom. There are some good examples
@@r_pydatascience When I increase the two numbers (0,1) I specified in my predict function to three (i.e. 0, 1, 2), the analysis is performed. So how do I decide the order of these three numbers? (0, 1, 2) or (0, 2, 1) or some other order?
can i print fixed and common model odds ratio graph in 1 time?
Yes. Set fixed and common to TRUE and then both pooled effects will be displayed in your plot.