How should cost data in pragmatic randomised trials be analysed?

How should cost data in pragmatic randomised trials be analysed?