Cost-effectiveness of non-consensus double reading