Background: Automated cytology devices have utility in quality assurance applications, but the effectiveness of these devices in primary screening is unknown.
Methods: Enrollment smears obtained from 7323 women participating in a population-based study sponsored by the National Cancer Institute were screened manually in Costa Rica and then evaluated independently in the U.S. with the PAPNET system (Neuromedical Systems, Inc., Suffern, NY), a semiautomated, neural network-based device. Smears with abnormal PAPNET images were microscopically rescreened and then diagnosed by a U.S. cytopathologist. ThinPrep slides (Cytyc Corporation, Boxborough, MA), prepared from rinses of the cytologic sampler, and cervigrams (National Testing Laboratories, Fenton, MO) were also evaluated. Women with any abnormal cytologic diagnosis or a positive cervigram were referred for colposcopy with biopsy and definitive therapy if indicated.
Results: Based on the U.S. cytotechnologist's review of the PAPNET images, 1017 (13.9%) of 7323 smears were selected for manual screening, resulting in the selection of 492 (6.7%) possibly abnormal slides for referral to the U.S. pathologist. Ultimately, 312 smears (4.3% of the total) were diagnosed as containing squamous cells of undetermined significance or a more severe abnormality (> or =ASCUS), resulting, hypothetically, in the referral of 66.5% of women with a final diagnosis of a squamous intraepithelial lesion or a more severe abnormality (> or =SIL) and 86.0% of patients with > or =high grade SIL. Conventional microscopic screening performed in Costa Rica resulted in the hypothetical referral of 6.5% of patients with > or =ASCUS for colposcopy, including 69.5% of patients with > or =SIL and 79.8% of those with > or =high grade SIL.
Conclusions: In this study, PAPNET-assisted cytologic screening accurately identified smears obtained from women with high grade SIL or carcinoma. Determination of the clinical cost-effectiveness of PAPNET-assisted screening in routine practice awaits future study.