Negative bias in exposure-response trends in occupational studies: modeling the healthy workers survivor effect.