[R-meta] Pooling studies with binary and continuous outcomes

[Blackman, Graham]

Dear R Special Interest Group for Meta-Analysis,

I hope this email find you all well.

I'm currently completing a meta-analysis (using the metafor package) looking at biological predictors of treatment response in patients suffering from psychosis. The predictor variable is recorded on a continuous scale.

Here�s the challenge�

Some studies report outcome as a binary variable (responder vs non-responders) and others report outcome as a continuous variable (% change in symptoms). For the latter, studies typically report the correlation coefficient. I would like to combine these different study designs to increase the statistical power.

Here�s my question�

Is it generally appropriate to convert the effect sizes for a correlation analysis (Pearson�s correlation coefficient) to Cohen's D?

If so, that solves the dilemma � if not, are they are other solutions?

Any advice greatly appreciated!

Best wishes,

Graham


--------------------------------------------------------------

Dr Graham Blackman BSc MBChB MRCPsych
Clinical Research Fellow
Section of Cognitive Neuropsychiatry, Psychosis Studies, 6th Floor
Institute of Psychiatry, Psychology and Neuroscience, 16 De Crespigny Park, Camberwell, London SE5 8AF

Email: graham.blackman using kcl.ac.uk<mailto:graham.blackman using kcl.ac.uk>
Telephone: 02078485228



	[[alternative HTML version deleted]]