Abstract P1-01-02: Ki-67 proliferation index supported by digital quantitation in breast cancer: A comparative study

Abstract

Abstract Introduction: In cases of breast cancer, in addition to hormone receptor and Her2 status, proliferation markers (mitotic index, Ki-67 proliferation index = KIPI) also have therapeutic implications. The 2013 St. Gallen consensus guideline includes 14% cut-off point (20% by many experts) for KIPI to distinguish luminal A-like and luminal B-like subtypes, that might be associated with remarkable intra-/interobsever variability applied in daily pathological routine utilizing semiquantitative (SQ) "eye-balling" method. Objective: The comparison of conventional SQ method and digital image-analysis (DIA) processes for the detection of KIPI. Methods: Three hundred and forty-seven breast cancer patients' samples with 99.24 months median follow-up data were included in our study (ethical approval: IKEB #7/2008 and 7-1/2008). Tissue microarrays (TMA) were prepared from the representative paraffin-embedded tumor blocks. After performing Ki-67 (MIB1 clone, #0505 by iOT on Ventana Benchmark XT autostainer by Roche) immunoreaction, conventional evaluation of KIPI was performed by 3 breast pathologists independently (SQ1-3). Digital image analysis was supported by PatternQuant (Pannoramic Viewer v15.3 and QuantCenter 2.0, 3DHistech Ltd.) applying a fully automatic tumor tissue recognition module with KIPI detection (DIA-1), and an adjustable module (DIA-2) with the possibility of manual corrections to exclude false detections. Interobserver variability was estimated with intra-class correlation coefficient (ICC). Digital pathological methods were compared to the - currently gold standard - SQ determination of KIPI using SPSS 22 statistical program. Results: The three pathologists' SQ evaluations demonstrated a remarkable concordance (ICC=0.889; 95% CI= 0.834-0.922). A reference KIPI value (KIPI-RV) was derived from mean values of SQ2 and SQ3, since no significant difference was found between them (p=0.617). KIPI-RV and DIA-2 showed no significant difference (p=0.754), and excellent concordance (ICC=0.979; 95% CI=0.975-0.982). Significant difference has occurred between KIPI-RV and results of DIA-1 (p=0.001). Upon dichotomizing KIPI value at 14%, no significant difference was found between KIPI-RV and DIA-2 (p=0.262), while KIPI-RV and DIA-1 differed (p=0.006). For prognosis prediction, all three methods were able to perform statistically significant division of our patients into 2 cohorts with distinct DFS at 14% (p<0.017-0.038). At 20% threshold of KIPI, DIA-1 failed (p=0.053), while KIPI-RV and DIA-2 were able to separate good and unfavorable prognosis patients' cohorts (p=0.01; p=0.004). Conclusion: The DIA processes are objective methods in the evaluation of KIPI. The fully automated DIA-1 method differed most from SQ results. Digital image analysis adjusted by a pathologist (our DIA-2 method) reached high concordance with results of SQ. Further refinement and validation are needed to verify applicability of automatic tumor pattern recognition software in diagnostic practice. Our results confirm that SQ evaluation of KIPI is reliable. This study was supported by the research grant from Hungarian Society of Medical Oncology 2014 and research grant from Doctoral School of Ph.D. Studies, Semmelweis University 2014. Citation Format: Acs B, Madaras L, Kovacs KA, Tokes A-M, Kulka J, Szasz AM. Ki-67 proliferation index supported by digital quantitation in breast cancer: A comparative study. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P1-01-02.