Analytical Comparison of Commonly Used Laboratory-Developed Tests for the Assessment of Ki-67 in Breast Carcinoma With a Food and Drug Administration-Approved Benchmark.

Ki-67 immunohistochemistry (IHC), a commonly used assay for breast cancer risk prognostication, has significant interlaboratory heterogeneity. This study assessed the impact of antibody clones by comparing the Ki-67 IHC MIB-1 pharmDx assay (Dako Omnis; Agilent Technologies) with clones MIB-1 (Dako Autostainer Link 48 platform), K2 (Leica BOND-III platform), and 30-9 (Ventana BenchMark ULTRA platform) used in Ki-67 laboratory-developed IHC tests. Breast cancer tissue microarrays were processed and stained in 2 central laboratories per the manufacturer's instructions. Digitized images were assessed by 5 pathologists (L.K.C., X.L., M.F.P., S.S., K.C.S.) before and after training on the reference scoring algorithm. Results were compared with results obtained by an artificial intelligence software for biomarker assessment (Mindpeak Breast Ki-67). Positive percent agreement (sensitivity), negative percent agreement (specificity), and overall percent agreement were calculated against the reference assay for the 20% cutoff, with exploratory analysis conducted for 5%, 10%, and 30% cutoffs. At the 20% cutoff, none of the laboratory-developed IHC tests achieved high overall agreement with the reference assay (predetermined as ≥85%). Both clones MIB-1 on Dako Autostainer Link 48 (sensitivity = 24.8% [95% CI, 20.2-29.9]; specificity = 99.5% [95% CI, 98.6-99.8]) and K2 on Leica BOND-III (sensitivity = 25.1% [95% CI, 20.5-30.3]; specificity = 100% [95% CI, 99.4-100]) had specificity comparable to that of the reference test, whereas clone 30-9 (Ventana BenchMark ULTRA) differed substantively in specificity from the reference assay in these measures (sensitivity = 99.3% [95% CI, 97.6-99.8]; specificity = 53.6% [95% CI, 49.6-57.5]). Intraclass correlation coefficient (ICC) for each pathologist ranged from 0.6 to 0.8, indicating good consistency across pathologists with little variability (variance component = 7.1 [95% CI, 2.1-29.7]). Training did not substantively alter within-assay or within-pathologist agreement. Ki-67 artificial intelligence analysis (ICC, 0.7; 95% CI, 0.4-0.9) was comparable to pathologists' assessment (ICC, 0.6-0.8). Commonly used IHC assays for Ki-67 assessment in breast cancer can significantly vary. Pathologists should be aware of variables that may impact Ki-67 interpretation and look to standardize biomarker assessments for early breast cancer patient care.

Duke Scholars

Author Kyle C. Strickland Pathology