Flexibly modelling the baseline risk in meta-analysis