Performance of specific morphologic features in distinguishing low-grade squamous intraepithelial lesions from high-grade squamous intraepithelial lesions in borderline cases: a College of American Pathologists Cytopathology Committee multiobserver study.

INTRODUCTION: Distinguishing between low-grade squamous intraepithelial lesions (LSIL) and high-grade squamous intraepithelial lesions (HSIL) can be difficult on certain Papanicolaou (Pap) tests, hindering interobserver concordance. We investigated the variables influencing the interpretation of LSIL versus HSIL in Pap test slides rejected from the College of American Pathologists PAP education program. MATERIALS AND METHODS: Eleven cytologists, who were unaware of the reference interpretation, examined 21 Pap slides (11 submitted as LSIL and 10 as HSIL) rejected from the PAP education program and recorded the number of LSIL cells, HSIL cells, keratinized dysplastic cells, LSIL clusters with mixed HSIL cells, atypical squamous metaplasia, atypical glandular cells, the presence of inflammation or infectious organisms, and the overall interpretation (LSIL or HSIL). We evaluated the significance of these 11 variables using a nonlinear mixed model analysis. RESULTS: LSIL had greater concordance (92 of 121 responses; 76.0% concordance) than HSIL (68 of 110 responses; 61.8% concordance; P < 0.001). The only predictors of misclassified cases were the number of atypical squamous metaplastic cells and the number of HSIL cells (P < 0.001). The more of these cells identified, the more likely the reviewers were to classify the slide as HSIL. The reproducibility of the diagnosis was fair (Gwet's agreement coefficient, 0.33). CONCLUSIONS: Interobserver reproducibility is a challenge for a subset of cases with features intermediate between LSIL and HSIL. Atypical squamous metaplasia and dysplastic nuclei with a nuclear/cytoplasmic ratio greater than one half of the cell volume (HSIL) present on a Pap test influenced the likelihood that a reviewer would interpret the case as HSIL rather than LSIL.