Analytical Variables That Affect the Prognostic Utility of Ki67 in Invasive Breast Cancer
Ying Wu, Brian Li, Greg Pond, Anna Marie Mulligan, Mark N Levine, Tim Whelan, Anita Bane. McMaster University, Hamilton, ON, Canada; University of Toronto, Toronto, ON, Canada
Background: Ki67 has been demonstrated to be a prognostic marker in invasive breast cancer; a high Ki67 is associated with poorer disease free survival (DFS) and overall survival (OS). There are however several analytical variables; including scoring methods and appropriate cut points for assigning a tumor as Ki67 high or low, that remain to be established. The goals of this study were to objectively determine the number of tumor nuclei requiring evaluation to reliably assign a tumor as Ki67 high or low and to determine an appropriate cut point.
Design: 300 full sections of T1/T2 lymph node negative, invasive breast cancers with 12 yr DFS and OS available were stained for Ki67 using the SP6 clone. 200 tumor nuclei were evaluated in each of four randomly selected regions of interest (ROI) and one hotspot ROI for every tumor. The Ki67 labelling index (LI); defined as the % of tumor nuclei positively stained for Ki67 was calculated for each ROI separately and in combination. Tumors were classified as Ki67 high or low based on cut points ranging from 5% to 35%. Likelihood ratio statistics were calculated using Cox proportional hazards regression models, with DFS and OS as the outcome and Ki67 status based on each cut point as the prognostic factor, to define a potentially optimal cut point based on discriminatory ability.
Results: The Ki67 LI was strongly correlated between different ROI (Spearman's r>0.78 for all comparisons), regardless of whether the ROI was randomly selected or a hotspot. Despite this, discrepancies in classification of Ki67 status were common; amongst randomly selected ROI, ∼15% of tumors would be classified differently when LI was calculated from 200 nuclei compared with 600 nuclei. Patients with tumors classified as Ki67 high had a worse DFS and OS than those classified as Ki67 low. Greatest discrimination after combining all randomly-selected ROI (800 nuclei) was observed at the 14% cut point; HR=1.70 (95% CI: 1.10-2.64), p=0.017 for DFS and HR=1.73 (95% CI: 1.05-2.84), p=0.031 for OS. When the hotspot evaluation was analysed alone or in combination with randomly selected ROI it was not associated with DFS or OS, regardless of the cut point applied.
Conclusions: In this series of breast cancers evaluation of at least 800 tumor nuclei across multiple ROI for Ki67 is required for prognostic utility. The hotspot Ki67 LI does not appear to be valuable. A Ki67 LI cut point of 14% had the strongest discriminatory ability for DFS and OS in this data set.
Monday, March 4, 2013 11:15 AM
Proffered Papers: Section B, Monday Morning