BACKGROUND: Ki-67 immunohistochemical expression is a prognostic and predictive marker in many breast cancer studies. Instead of the conventional time-consuming score of Ki-67 single stains associated with low reproducibility, Ki-67/KL1 double stains may facilitate fast, repeatable quantification by digital image analysis. This study aims to detect the difference in accuracy and precision between manual indices of single and double stains, to develop an automated quantification of double stains, and to explore the relation between automated indices and tumor characteristics when quantified in different regions: hot spots, global tumor areas, and invasive fronts. MATERIALS AND METHODS: Paraffin-embedded, formalin-fixed tissue from 100 consecutive patients with invasive breast cancer was immunohistochemically stained for Ki-67 and Ki-67/KL1. Ki-67 was manually scored in different regions by 2 observers and automated image analysis. RESULTS: Indices were predominantly higher for single stains than double stains (P≤0.002), yet the difference between observers was statistically significant (P<0.001) for both stains. The Pearson correlation coefficient for manual and automated indices ranged from 0.69 to 0.85 (P<0.001). Hot spots were slightly superior to other regions when correlating automated indices with tumor characteristics, for example, tumor size (P<0.001), grade (P=0.009), and estrogen receptor status (P=0.04). CONCLUSIONS: Although precision was unsatisfactory for manual indices of both stains, Ki-67 should be quantified on double stains to reach a higher accuracy. Automated indices correlated well with manual estimates and tumor characteristics, and they are thus possibly valuable tools in future exploration of Ki-67 in breast cancer.
Applied Immunohistochemistry and Molecular Morphology, 2014, Vol 22, Issue 8, p. 568-76