Strength and Weakness of Multi-level Sectioning in the Detection of HSIL: A Study Based on Inter-observer Variation and Imunohistochemical Markers
AP Pinto, NF Schlecht, ES Cibas, CP Crum. Federal University of Paran, Curitiba, PR, Brazil; Albert Einstein College of Medicine, Bronx, NY; Brigham and Women's Hospital, Boston, MA; Brigham and Women's Hospital and Harvard Medical School, Boston, MA
Background: Distinguishing high-grade squamous intraepithelial lesions (HSIL) from their mimics can be difficult. Additional paraffin sections are commonly requested in order to improve diagnostic accuracy. The current study evaluated the performance of multi-level section analysis by multiple reviewers to differentiate HSIL from reactive or atrophic epithelium with atypia.
Design: 66 archived specimens (41 cervical biopsies, 13 cone products, 11 endocervical curettings, and 1 endocervical polyp) with up to 4 H&E stained slides (original plus levels, total of 234 slides) were independently reviewed by three pathologists and classified as squamous intra-epithelial lesion (SIL) or negative for SIL (NoSIL) for each level. Morphological identification of SIL in at least one of the levels by at least one of the reviewers, plus the positive result of at least two out of three immunohistochemical biomarkers (p16INK4a, MIB-1 and ProExTMC) were employed to establish a final diagnosis. Morphologic diagnosis in each re-cut level was compared to the morphological/immunohistochemical final diagnosis. Kappa statistics and receiver operating characteristic analysis were used to assess agreement and test performance.
Results: Agreement on the diagnosis of SIL by H&E alone across all three pathologists was good to very good (kappa=0.41 to 0.58). The proportions of correctly diagnosed SIL and NoSIL respectively increased and decreased after the multi-level section analysis for all reviewers, as well as for the consensus morphological diagnosis (SIL: 77.5% for L1, 80.0% for L1+L2, 83.3% for L1+L2+L3 and 87.0% for L1+L2+L3+L4; NoSIL: 92.3% for L1, 84.6% for L1+L2, 81.0% for L1+L2+L3 and 77.8% for L1+L2+L3+L4). Diagnostic accuracy to detect either SIL or NoSIL for H&E diagnosis by level were 0.85 (95%CI 0.77-0.93) for L1, 0.82 (95%CI 0.73-0.92) for L1+L2, 0.82 (95%CI 0.71-0.93) for L1+L2+L3, 0.82 (95%CI 0.70-0.95) for L1+L2+L3+L4.
Conclusions: Whereas the use of multi-level sectioning improved SIL detection sensitivity, this occurred at the expense of specificity. Although the use of additional H&E slide sections can help identify otherwise undetected SIL cases, this also may lead to over treatment in some patients with SIL mimics.
Monday, March 9, 2009 1:00 PM
Poster Session II # 132, Monday Afternoon