The Case for a Bayesian Approach to Benefit-Risk Assessment