I am presently performing a Meta-analysis for a Systematic Review. Most of the included Randomized Controlled Trials (RCTs) in the review have reported Least Square (LS) Mean and Standard Error (SE) for the each respective group. Whereas, few RCTs have Arithmetic Mean and SE.
In this case, how do I use this Information to perform Meta-analysis (continuous outcomes) by using Review Manager (Version 5.3).
Is there any method/formula to convert LS mean to Arithmetic Mean?
Please help ....
Thank you
Are least square means simply a mean "adjusted" for various covariates in a model. And the arithmetic mean is unadjusted. Presumably trusting the randomisation to deal with any confounding. So, and I'm happy to be corrected on this, I'd see both means as being estimates of the same thing, so could be combined using the estimate and standard error, using the inverse-variance method.
But... because some estimates are adjusted for covariates, and others are not, this is a potential source of heterogeneity, and you'd be wise to explore the extent to which results differ between the methods, and even whether it matters what covariates are adjusted for.
I've not come across this exact situation myself before, so I would welcome any comments from more experienced colleagues, but that's my best guess. So I think you should be able to do all that in RevMan.
Agree with Darren, least square means are derived from a linear model and adjusted for covariates. If you combine the two sorts of means you may want to use models (IVhet; QE) that factor in heterogeneity more robustly and thus use MetaXL (www.epigear.com)
least square is a adjusted mean and it can be a source of hetrogenity,
What if all the trials in your meta analysis have reported their data as least mean squares and standard error ? And further, all of these has used same model and covariates for calculating the least mean square of respective groups? Can we simply use least mean squares in place of mean and calculate the standard deviation from standard error by Review Manager (Version 5.3) calculator ? Or we should go for inverse-variance method by using the least mean square difference and standard error?
you should calculate the effect size separately and manually for each paper. I assume for your case that the effect size is the standardized mean difference (SMD).
Cooper et al* proposed methods to calculate SMD from means and marginal means (LSM) see page 215
* Cooper, Harris, Larry Vernon Hedges, and Jeffrey C. Valentine. "The handbook of research synthesis and meta-analysis 2nd edition." The Hand. of Res. Synthesis and Meta-Analysis, 2nd Ed.. Russell Sage Foundation, 2009. 1-615.
You can also see: https://pdfs.semanticscholar.org/d88e/7804b7df89d229fd5db7a1c567b94ff18f33.pdf
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis