To Count or Not To Count? Scoring Ki67 in Breast Cancers According to International Working Group Recommendations
Kathi H Adamson, Eric Q Konnick, Suzanne M Dintzis, Mara H Rendi, Hannah M Linden, Kimberly H Allison. University of Washington, Seattle, WA
Background: Proliferation related markers can be used to both identify breast cancers that may respond to a therapy and provide a measurement of therapy response. An International Ki67 in Breast Cancer Working (IKBCWG) published Ki67 assessment recommendations, which proposed 1000 cell counts per case. We tested the practical application of their recommendations and compared results between observers as well as with routinely reported results.
Design: 43 cases of ER positive breast cancer were counted independently by two pathologists using the IKBCWG recommendations with 1,000 cells counted per case. Percent Ki67 positive cells (Ki67 index) were compared both between observers and routinely reported Ki67 results (did not require counting cells). A Ki67 index threshold of 14% was used to classify cases into low verses high. Lastly, scoring was analyzed by 100-cell count increments to determine accuracy per 100-cell count vs final 1,000-cell count scores for all observations. Deming regression and descriptive statistics were used to analyze correlation between observers and reported estimates.
Results: Good correlation between two observers was found when 1000 cells were counted to ascertain Ki67 index (y = 1.08x + 0.08, R2= 0.82). The standard deviation (SD) per read ranged from 0-9.9%, with the highest coefficent of variation (CV) at low Ki67 index values. The mean 1000 cell count scores had good correlation with the clinically reported estimates (y = 0.89x - 2.97, R2= 0.86), with a slight negative bias to the reported estimate. Using a 14% threshold for low vs high Ki67, 100% of cases classified as low by 1000-cell count were also classified as low by routine estimation. However, only 75% of cases classified as high by 1000-cell count were high by routine estimation. When counts were analyzed in 100-count increments, 200-cell counts were found to be within 5% of the 1000-cell count method, but 800-cell counts were required to be within 2.5% of the 1000-cell method, with 95% frequency in each case.
Conclusions: Routine Ki67 estimation underestimates the Ki67 index when compared to 1,000-cell count methods. There is good correlation between pathologists when 1000-cell counting methods are used, however, there is inter-observer variation, which is increased at low Ki67 index values. Counting fewer than 1000 cells can produce results similar to 1000-cell counts and may be a more practical in clinical practice.
Monday, March 4, 2013 11:30 AM
Proffered Papers: Section B, Monday Morning