Probabily you have 0 events in your groups, (one of them or both).

Thanks Vicente, but that is not the case. I have a problem with the extracted OR from the relevant study, it's already calculated in a published paper. I am not using the events.

Ok, you can use the function escalc on the package metafor into R, this function give you a wide range of transformed options for ORs, I hope this info help you

I do not use R, however will try to look into it. Thanks.

Best to attach an excel sheet with your computations as well as the mismatched output from a study that does not match so we can assist you further

Hi Cem. If you look at the following link

http://cfgd.cochrane.org/sites/cfgd.cochrane.org/files/public/uploads/Generic%20Inverse%20Variance%20method.doc

Study 2 is consistent with the original data but study 5 is not consistent , may be I am missing something 

I do not have this problem when I used comprehensive mets analysis version 2 

if you need any help let me know 

thanks

Fatma

Dear Doi, I cannot attach it here for confidential reasons, however can send you an e-mail if you'd like. Thanks.

Hi Fatma,

Yes that's inconsistent indeed. In my case, the possibility that the authors could have done a mistake for the CI limits in the paper getting stronger. 

I'll send you the data and maybe you can have a look.

Many thanks

Best

I would like to thank to Suhail A. Doi and Fatma A Etwel for working on this issue. The CIs reported in the relevant studies are incorrect. 

Generic inverse-variance (IV) method approach is implemented in its most basic form in RevMan (latest version 5.3.5), and is used behind the scenes in certain meta-analyses of both dichotomous and continuous data.

The IV method is so named because the weight given to each study is chosen to be the inverse of the variance of the effect estimate (i.e. one over the square of its standard error [SE]). Thus larger studies, which have smaller SEs, are given more weight than smaller studies, which have larger SEs. This choice of weight minimizes the imprecision (uncertainty) of the pooled effect estimate.

A fixed-effect meta-analysis using the inverse-variance method calculates a weighted average as (see the attachment for formula).

The results of a study may be expressed as a rate ratio, that is the ratio of the rate in the experimental intervention group to the rate in the control group (see Section 9.4.8 Meta-analysis of counts and rates, Cochrane Handbook for Systematic Reviews of Interventions). Suppose EE events occurred during TE participant-years of follow-up in the experimental intervention group, and EC events during TC participant-years in the control intervention group. The rate ratio is (see the attachment for formula).

The (natural) logarithms of the rate ratios may be combined across studies using the generic IV method (see Section 9.4.3.2 The generic inverse variance outcome type in RevMan, Cochrane Handbook for Systematic Reviews of Interventions). An approximate standard error of the log rate ratio is given by (see the attachment for formula).

Bradburn et al. found that (see Section 16.9.5 Validity of methods of meta-analysis for rare events, Cochrane Handbook for Systematic Reviews of Interventions) many of the most commonly used meta-analytical methods were biased when events were rare (Bradburn 2007).  The bias was greatest in inverse variance and DerSimonian and Laird odds ratio and risk difference methods, and the Mantel-Haenszel odds ratio method using a 0.5 zero-cell correction. As already noted, risk difference meta-analytical methods tended to show conservative confidence interval (CI) coverage and low statistical power when risks of events were low.

The random effects model will tend to give a more conservative estimate (i.e. with wider CI), but the results from the two models usually agree where there is no heterogeneity. When heterogeneity is present the random effects model should be the preferred model: https://www.medcalc.org/manual/meta-generic.php.

An example with similar discrepancy in effect measure and CIs (published), http://www.ncbi.nlm.nih.gov/pubmed/22068607 (Off-pump versus on-pump coronary artery bypass surgery in patients aged 80 years and older: institutional results and meta-analysis) (see the attachment).

http://handbook.cochrane.org/chapter_9/9_4_8_meta_analysis_of_counts_and_rates.htm

http://handbook.cochrane.org/chapter_16/16_9_5_validity_of_methods_of_meta_analysis_for_rare_events.htm

http://handbook.cochrane.org/chapter_9/9_4_3_2_the_generic_inverse_variance_outcome_type_in_revman.htm

Hi Sharanbasappa, the random effects model is neither conservative nor is it useful for meta-analysis. Pls see www.epigear.com under MetaXL to find out why I say this. 

Hi Yusuf,

the correct formula to calculate the SE of lnOR is

RQ(1/a+1/b+1/c+1/d).

With this formula, the OR and CI output is similar to the original value, both with the random effect that with the fixed effect model.

The formula for the SE of the ln Risk Rate

RQ(1/Ee + 1/Ec)

or  the SE of the ln Risk Ratio

RQ[(1-pe)/Nepe+(1-pc)/Ncpc]

is different.

This could justify the discrepancy.

I hope I have been helpful. Carmen

Thank you Carmen. Very good point. However, even we assume that the erroneous risk estimates are RRs and not ORs, should the upper and lower limits of the CI not be symmetrical in the log scale since it is ln(RR) ± 1.96 RQ[(1-pe)/Nepe+(1-pc)/Ncpc] ?

They are not symmetrical in our calculations which suggests that the reported CI intervals in the relevant studies are not correct.

Best,

Yusuf

What are the discrepancies you have?  You can even manually calculate ORs/RRs and CIs to check if reported ones are correct as long as enough data is reported.

In log scale, CIs should be symmetrical.

Thank you Mehmet, 

Unfortunately, raw data are not available for the calculation and some of the ORs are adjusted. 

There’s probably a misunderstanding.

The lnOR CI (or lnRR CI) is certainly symmetrical, but the CIs reported in meta-analyses and studies are generally that of OR, not that of LnOR

The upper and lower limits of the OR CI are the exponents of the limits of lnOR CI, then can be asymmetric.

IC lnOR = lnOR± 1.96 RQ[(1-pe)/Nepe+(1-pc)/Ncpc] = upper value ------ lower value

IC OR = exp (upper value) -------  exp (lower value)

These values, being the exponents, do not derive directly from the formula…± 1.96 …,  so can lose the symmetry.

Best regards

Carmen

As far as I understand, the discrepancy is the mismatch between reported CIs and calculated CIs. So you had better provide the data in order for us to check what is going on.

Another possible explanation might be using log10 instead of loge (ln) which would result in different CIs. Revman has a calculator, did you check the results with that calculator? How are the reported lower and upper bounds when calculated with Revman's calculator? Aren't they the same as reported ones?

The CIs yielded in RevMan are different than the original ones. I can e-mail the data if you would like to take a look.