ComBat before or after filtering

Question

[averagelife]

Hi,

I have 2 microarray data sets with patient data that were obtained over several years. There is a strong batch effect that I want to adjust using combat. I read in earlier posts, some suggest it doing prior to low exp. filtering but many do it following afterward. In my understanding combat expects the data to be normalized/cleaned, so isnt it better to perform after filtering? Which one is the correct way?