Identifying type 1 and 2 diabetes in population level data: assessing the accuracy of published approaches