Objectives: To assess computer-assisted (neural network based) cervical smear screening as a primary tool for the early detection of cervical dysplasia.
Design: Longitudinal cohort study.
Setting: Cytology laboratory reviewing cervical smears taken by general practitioners in a mass screening program in the Netherlands.
Subjects: 846 women who developed (pre-)neoplasia of the cervix in the seven years after the baseline smear, and 5217 controls.
Interventions: Cervical smears were evaluated both by conventional light microscopy and with use of the PAPNET Testing System by the same cytotechnologists.
Main outcome measures: Seven year histological and cytological follow-up results were obtained for all women from a nation-wide pathology database.
Results: Conventional screening diagnosed dysplasia or carcinoma in the baseline smears of 458 (54.1%) of the 846 women who were diagnosed with (pre-)neoplasia during follow-up, whereas computer-assisted PAPNET analysis detected such lesions in 462 (54.6%) of these women. In the control population of 5217 (86.0%) women, in whom follow-up revealed no cervical dysplasia, conventional screening gave false positive results in 210 (4.0%) and computer-assisted PAPNET analysis gave false positive results in 207 (4.0%) smears. The areas under the receiver operation curves (AUC) were 80% (95% confidence interval, 78 to 82%) and 79% (95% confidence interval, 77 to 81%) for conventional and PAPNET-assisted screening, respectively.
Conclusions: The PAPNET Testing System has similar diagnostic value as the conventional screening of Pap smears when used for primary screening.