Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes. (12th March 2010)