Reducing the effects of lead-time bias, length bias and over-detection in evaluating screening mammography: a censored bivariate data approach