Comparison of the effect of different techniques for measurement of Ki67 proliferation on reproducibility and prognosis prediction accuracy in breast cancer

Abstract

The proliferation factor Ki67 is prognostic in breast cancer and included in international therapy guidelines, but measurement procedures differ between laboratories. We compared the reproducibility and prognostic value of different Ki67 sampling and measurement methods. In 237 T(1,2) N(0) M(0) breast cancers without adjuvant systemic treatment, strictly standardized section thickness, automated antigen retrieval and immunohistochemistry were used. The percentages of Ki67-positive nuclei were assessed using (i) a 'quick-scan rapid estimate', (ii) ocular-square-guided counts by independent pathologists, (iii) computerized point-grid-sampling interactive morphometry (CIM) and (iv) automated digital image analysis (DIA). Quick-scan rapid estimates were poorly reproducible. The optimal prognostic thresholds of Ki67 counts by two pathologists differed greatly (4%, 14%; kappa: 0.36), with many therapeutic differences. CIM-Ki67 and DIA-Ki67 were strongly prognostic (P<0.0001) and reproducible. DIA-Ki67 (threshold: 6.5%) was the strongest and most robust prognosticator (the threshold could vary from 4 to 15% without significant prognostic loss). Ki67 was prognostically strongest in the periphery of the tumour. In node-negative breast cancer without adjuvant systemic treatment, Ki67% by DIA, but not subjective counts, is reproducible and prognostically strong. This casts serious doubt on therapeutic guidelines using subjective counts of Ki67.