My question adresses specifically continuous variables. When does the first method is indicated over the second? When is it counter-indicated? How to calculate it? Can I calculate it on RevMan?
Hi Giovanni,
there is an interesting article by Stephen Senn, dealing with this question
"Change from baseline and analysis of covariance revisited.",
Statistics in Medicine 01/2007; 25(24):4334-44. DOI:10.1002/sim.2682 pp.4334-44
https://www.researchgate.net/publication/6865737_Change_from_baseline_and_analysis_of_covariance_revisited?ev=srch_pub
In short, an ANCOVA using baseline measures as covariate improves the efficiency of the estimated effect due to reducing residual variance. On the other hand, when dealing with non-randomized trials (eg observational trials), you may observe Lord's paradox, when there a differences in covariates between groups at baseline.
I don't know if RevMan has capabilities to calculate ANCOVAs (I don't think so), but certainly any statistical software (R, SPSS, STATA, SAS etc) can do an ANCOVA.
hth
Article Change from baseline and analysis of covariance revisted
Dear Eik,
thank you so much for your enlightening answer!
Right now, I'm conducting a systematic review with few trials (although all of them are randomized), so I need to conduct the best analysis possible.
Many thanks
Giovanni.
Hi Giovanni,
that’s a different matter. Adjustment for baseline measures is on individual(patient) level, but you want to conduct a meta analysis based on aggregated data.
Depending on how few your "few" is, you have to choose between a random or fixed effects meta analysis, both are possible in RevMan (I would recommend random effects, whenever possible). Points to consider are discussed here http://www.cochrane-net.org/openlearning/html/mod13-4.htm
Differences in (aggregated) baseline data between studies, if available, would then also indicate heterogeneity between study populations, advocating a random effects model.
hth.
Dear Eik,
I've already ran the sensitivity analysis, assessed risk of bias and quality of evidence with the GRADE approach as well as heterogeneity of the pooled data and I chose to run the meta-analysis with a random effects model.
Thus, my question relies on which effect size measure would be more appropiated to be used. Cochrane handbook suggests the use of mean differences or mean differences adjusted to baseline valuesbut I really don't know what is the main difference between both, which would be more appropiate to conduct, generally speaking.
Are both effect measures available in all studies? If so, then fortunately a lot changed since the reviews of Altman and Doré (1990)
http://www.sciencedirect.com/science/article/pii/014067369090014V
and Austin et al (2009) https://www.researchgate.net/publication/26776077_A_substantial_and_confusing_variation_exists_in_handling_of_baseline_covariates_in_randomized_controlled_trials_a_review_of_trials_published_in_leading_medical_journals
Then according to Stephen Senn (see above) the adjusted estimate is unbiased and more efficient.
Article A substantial and confusing variation exists in handling of ...
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis