Could someone possibly suggest a solution to how to overcome problems with meta-regression when working with effect estimates with very narrow confidence intervals?
I'm using meta-regression to look at (for example) whether differences in effect estimates may be explained by the % of the study population who have kidney impairment. However when I try meta-regression, I get an error message relating to the fact that for some huge studies, the lower (or upper) CI, given to 2dp, is the same as the OR (ie OR 2.36 (95% CI 2.35-2.36)). I have extracted the effect estimate and CI from the studies, so I don't have any option to increase the precision of the results (ie to increase the number of decimal places). So far, to check my code and generate approximate results, I have added 0.01 to the upper CI where this matches the OR, and meta-regression runs fine. However I don't think this would be a great option to present in a manuscript as essentially I've falsified the results!
************************************************** ******
meta set angio_logOR angio_loglci angio_loguci, random(reml) civartolerance(0.1) studylabel (Author)
effect size variable angio_logOR must be < CI upper limit, angio_loguci
************************************************** ****
Has anyone else experienced this problem and/or can think of a solution?
Thanks so much
Jemima
I'm using meta-regression to look at (for example) whether differences in effect estimates may be explained by the % of the study population who have kidney impairment. However when I try meta-regression, I get an error message relating to the fact that for some huge studies, the lower (or upper) CI, given to 2dp, is the same as the OR (ie OR 2.36 (95% CI 2.35-2.36)). I have extracted the effect estimate and CI from the studies, so I don't have any option to increase the precision of the results (ie to increase the number of decimal places). So far, to check my code and generate approximate results, I have added 0.01 to the upper CI where this matches the OR, and meta-regression runs fine. However I don't think this would be a great option to present in a manuscript as essentially I've falsified the results!
************************************************** ******
meta set angio_logOR angio_loglci angio_loguci, random(reml) civartolerance(0.1) studylabel (Author)
effect size variable angio_logOR must be < CI upper limit, angio_loguci
************************************************** ****
Has anyone else experienced this problem and/or can think of a solution?
Thanks so much
Jemima
