Eliciting uncertainty for complex parameters in model-based economic evaluations: quantifying a temporal change in the treatment effect