Aggregating published prediction models with individual participant data: a comparison of different approaches