I want to perform a meta-analysis of prognostic models and I would like to pool calibration slopes since these were reported in various validation studies included in the review.
I found that one can use the valmeta() function from the metamisc package to pool O:E ratio or c-statistics. In the description of the function, it is described that the function can be used to pool calibration slopes, but I can’t find further information on the required arguments (there are only arguments for O:E ratio or c-statistics).
Therefore, my questions are:
- Can the function that pools O:E ratios be used to pool calibration slopes?
- If not, does anyone know functions in R (or even functions in other statistical programs) that can be used to pool calibration slopes?
Thank you very much in advance!
I talked to Thomas Debray, the author of the metamisc package. Meta-analysis of calibration slopes is currently not yet built into the function. However, one can apply “normal” meta-analysis (i.e., metafor package in R etc.) because calibration slopes do not require transformation.
Are you also going to present the individual slopes or perhaps slopes stratified by potentially relevant subgroups (e.g. age or disease based subgroups)
Pooling all calibration slopes into one average through meta-analysis might pool together groups in which the prognostic model is relatively miscalibrated which in itself could be a relevant detail to judge whether the model is suited for use in specific (sub)groups?
@scboone Yes, that’s a great idea!! I feel like there aren’t enough analyses of prediction models, specifically in which cohort they work best. This would surely help enhance the quality of future models/modifications of models.