I don't know how they made the calculation, but here is how I would do it:
Take a random sample of people (say a 1,000) properly stratified among different demographics that may indicate a higher/lower prevalence for diabetes such as gender, age, ethnicity, income levels, regional, lifestyle...