RESUMO
Reclassification of endocervical atypical immature metaplasia (AIM) into reactive changes and neoplastic lesion is often challenging. We aimed to accurately reclassify AIM on hematoxylin and eosin (HE)-stained sections without auxiliary immunohistochemistry (IHC). A total of 133 AIM diagnosed by punch biopsy were reclassified by IHC for p16 and Ki-67 into high-grade squamous intraepithelial lesion (HSIL) or negative for intraepithelial lesion or malignancy or low-grade squamous intraepithelial lesion (NILM/LSIL) as a reference. Nuclear features significantly associated with HSIL on HE-stained sections were extracted by multivariate logistic regression analysis. Propensity score (PS) of HSIL was calculated in each case and cut-off was determined by receiver operation characteristic (ROC) curve analysis. As a result, AIM was reclassified into 104 NILM/LSIL and 29 HSIL by IHC. Compared with reference diagnosis, accuracy of pathologists' subjective diagnosis was 54.9% (kappa coefficient, 0.208). Three nuclear features on HE-stained sections, ie, nuclear enlargement with anisokaryosis, nuclear hyperchromasia, and mitosis, were significantly associated with HSIL. The ROC curve analyses revealed that PS and number of nuclear features were significant predictors of HSIL. Diagnostic accuracy of PS-based diagnosis was 76.7% (kappa, 0.447). When AIM with 2 or more of the 3 nuclear features was diagnosed with HSIL, diagnostic accuracy was 77.4% (kappa, 0.448). Nuclear feature-based diagnosis significantly improved diagnostic accuracy on HE-stained sections compared with subjective diagnosis and may be useful when IHC is not available. However, a considerable proportion of AIM would still remain misdiagnosed and IHC for p16 and Ki-67 should be mandatory for accurate reclassification.
