Elimination of "lead time" bias in assessing the effect of early breast cancer diagnosis.