I haven't read the article and I have no experience in the relevant field, but I think it's a simple idea: P(disease|sugar) = 50 * P(disease|calories from sugar).

That is, if all you know is that you've consumed n calories, we think you probably you have probability p of getting the disease; if you've consumed n calories' worth of sugar, we think you have probability 50p of getting the disease.