A design for cancer case-control studies using only incident cases: experience with the GEM study of melanoma.