Quantifying Ki67 for characterising molecular subtypes in breast cancer: a comparative study of visual assessment and digital image analysis

Aims Ki67 index assessment is crucial for breast cancer classification and prognosis. We compared Ki67 by immunohistochemistry (IHC) using visual counting-eyeballing (VA-E), manual counting on images (VA-M) and digital image analysis (DIA) to evaluate agreement levels and the impact when using IHC as a surrogate for molecular classification. Methods The study included needle biopsies of invasive breast cancers. The Ki67 index was re-estimated by DIA and VA-M. The re-estimated values were compared with previously reported VA-E values. Agreement was assessed by the kappa (κ) coefficient, and Bland-Altman plots were used to analyse the mean difference in the index and its impact on molecular classification. Results The study included 389 invasive ductal carcinomas of no special type. The majority were grade 2 (n=205, 52.7%) and 44% (n=171) were hormone receptor-positive (HR+). Grade 1 and 2 tumours with a Ki67 index of <20% showed significantly lower estimates by VA-E than DIA. Bland-Altman analysis revealed mean differences of 1.8 (95% CI −0.24 to 3.4) for all IDC and 4.2 (95% CI −1.91 to 6.49) for HR+ luminal cancers. Re-evaluation of the Ki67 index by DIA reclassified 28/62 luminal-A-like tumours to luminal-B-like and 16/109 luminal-B-like tumours to luminal-A-like, with an absolute discordance of 25.7%. VA-M estimates correlated more closely with DIA than VA-E. Conclusions Lower Ki67 was observed with VA-E compared with DIA. There was discordance in the classification of luminal-like tumours, highlighting potential clinical implications when applying Ki67 cut-offs. VA-M demonstrated greater concordance with DIA than VA-E. For facilities lacking DIA, VA-M may be performed for Ki67, particularly in low-grade HR+ tumours.