Reproducibility of classification and correction for verification bias as determinants of performance of Papanicolaou smear cytology in the screening setting: experience from the New Independent States of the former Soviet Union cohort study.