Using historical data for Bayesian sample size determination