Ki67 and breast cancer mortality in women with invasive breast cancer

Abstract Background The percentage of cells staining positive for Ki67 is sometimes used for decision-making in patients with early invasive breast cancer (IBC). However, there is uncertainty regarding the most appropriate Ki67 cut points and the influence of interlaboratory measurement variability. We examined the relationship between breast cancer mortality and Ki67 both before and after accounting for interlaboratory variability and 8 patient and tumor characteristics. Methods A multicenter cohort study of women with early IBC diagnosed during 2009-2016 in more than 20 NHS hospitals in England and followed until December 31, 2020. Results Ki67 was strongly prognostic of breast cancer mortality in 8212 women with estrogen receptor (ER)–positive, human epidermal growth factor receptor 2 (HER2)–negative early IBC (Ptrend < .001). This relationship remained strong after adjustment for patient and tumor characteristics (Ptrend < .001). Standardization for interlaboratory variability did little to alter these results. For women with Ki67 scores of 0%-5%, 6%-10%, 11%-19%, and 20%-29% the corresponding 8-year adjusted cumulative breast cancer mortality risks were 3.3% (95% confidence interval [CI] = 2.8% to 4.0%), 3.7% (95% CI = 3.0% to 4.4%), 3.4% (95% CI = 2.8% to 4.1%), and 3.4% (95% CI = 2.8% to 4.1%), whereas for women with Ki67 scores of 30%-39% and 40%-100%, these risks were higher, at 5.1% (95% CI = 4.3% to 6.2%) and 7.7% (95% CI = 6.6% to 9.1) (Ptrend < .001). Similar results were obtained when the adjusted analysis was repeated with omission of pathological information about tumor size and nodal involvement, which would not be available preoperatively for patients being considered for neoadjuvant therapy. Conclusion Our findings confirm the prognostic value of Ki67 scores of 30% or more in women with ER-positive, HER2-negative early IBC, irrespective of interlaboratory variability. These results also suggest that Ki67 may be useful to aid decision-making in the neoadjuvant setting.


A. Tabulation of person-years at risk & observed events
For each woman who was eligible for the study, the length of time from 3 months after her cancer diagnosis until the earliest of death, emigration, or 31st December 2020 was calculated.These lengths of time were added together to form the person-years at risk.The number of women whose contribution to the personyears was terminated by death from breast cancer was also obtained, while women whose contribution to the person-years was terminated by emigration or the end of follow-up were censored, as were women who died from a cause other than breast cancer (apart from the analyses that considered deaths from all causes).The numbers of person-years were then tabulated simultaneously according to all the factors shown in Table 1, as were the numbers of deaths.

B. Annual breast cancer mortality rates and rate ratios
Crude annual breast cancer mortality rates were calculated by dividing the number of deaths observed in a particular group by the number of person-years at risk, and their associated confidence intervals were calculated by assuming that the number of deaths observed had a Poisson distribution.Crude rate ratios were estimated using Poisson regression with the numbers of deaths as the dependent variable, the numbers of personyears (which were assumed to be fixed) as the exposure, and the variable of interest (e.g.Ki67) included as a categorical factor.Time since diagnosis was also included as a categorical factor by classifying both the personyears and the numbers of deaths into the following categories: 0.25-, 1-, 2-, ...7-, 8+ years.Adjusted rate ratios were estimated by including these factors and also all the other characteristics listed in Table 1 in the model simultaneously, using the categories displayed in Table 1.For each factor, missing values were assigned to a separate category.Significance tests were carried out using the likelihood ratio omitting the categories for missing values.Group-specific 95% confidence intervals were calculated for the rate ratios for each category of each factor (Plummer M. Improved estimates of floating absolute risk.Stat.Med. 2004;23:93-104).
To obtain adjusted annual mortality rates (rather than rate ratios), a Poisson regression model with main effects for Ki67 and all the factors to be included in the adjustment was fitted.Then, for each category of a particular characteristic, a weighted average of the estimates involving that category was calculated, with weights proportional to the person-years.

C. Adjusted cumulative risk of breast cancer mortality
If denotes the estimated adjusted annual breast cancer mortality rate for time-interval (as calculated in section B above), and denotes the variance of , where is the length of the time interval, then the cumulative rate, and its corresponding variance, were calculated as The cumulative rates (point estimates and 95% confidence limits) were then transformed into the cumulative risk using the formula, = 1 − exp( − Λ ).

D. Standardization between laboratories
For the women with ER-positive and HER2-negative disease, an investigation of the distribution of Ki67 measurements from each laboratory was conducted to assess whether there was any evidence of systematic differences between them.A box and whisker plot showed that the median of the Ki67 measurements varied between laboratories by a factor of more than 4 (from 5% to 22%) (Figure 5A), and a Kruskal-Wallis rank test (allowing for ties) also rejected the hypothesis that the distribution of the Ki67 measurements was identical in the different laboratories (p<0.001).
The distributions of Ki67 measurements within each laboratory were also highly skewed with a heavy upper tail.Therefore, before adjusting for the difference in median values between laboratories, it was desirable to find a transformation that removed the skewness.This was done with the aid of quantile-quantile plots.In these plots the Ki67 scores in each laboratory were sorted into rank order (x(1), x(2), x(3), ..., x(N)), where N is the number of observations in the laboratory, and the ranked values were plotted against a scaled inverse normal distribution.To obtain the appropriate scaling for each laboratory, the mean (μ) and variance (σ 2 ) of the Ki67 scores in the laboratory were calculated.Then, if Φ(i) is the inverse of the standard normal distribution corresponding to x(i) (i.e.(i/N+1) ), the scaled value of x(i was The quantile-quantile plots for the raw Ki67 scores are shown in Supplementary Figure 2A.A number of different transformations were investigated and it was found that transforming the Ki67 measurements by taking natural logarithms (plus one to avoid a singularity with measurements of zero) resulted in distributions that were approximately symmetric (Figure 5B) and also approximately normally distributed, see Supplementary Figure 2B.
After transformation, a one-way analysis of variance showed that there were still highly significant differences between the means of the transformed measurements from the laboratories (F(24,8187)=49.66,p<0.0001).These differences were removed by subtracting from each transformed measurement the mean value of the laboratory-specific transformed measurements, and then adding to it the mean value over all the transformed measurements.In these standardized values, the laboratory-specific median values varied by only a factor of 1.2 and there was little evidence of skewness (Figure 5C).In addition, a Kruskal-Wallis rank test (allowing for ties) provided no evidence to reject the hypothesis that the distributions of the transformed measurements were identical in the different laboratories (p=0.96).The standardized scores were backtransformed to the percentage scale by exponentiating and subtracting one (3 women had scores below 0 and 33 above 100).These back-transformed scores were rounded to the nearest integer value before being grouped using the same cut-points as in the unstandardized analyses.The upper bounds of the highest Ki67 groups was left unrestricted, because 31 women had rounded standardized scores above 100.

E. Digit preference in Ki67 recording
The recorded Ki67 scores exhibited considerable digit preference, with the majority of scores reported as 5, 10, 15, etc. (Figure 1).This suggests that, for example, many women for whom a score of 5 is reported, may actually have a score of 3, 4, 6, or 7. Therefore, to provide a potentially more robust and separation of the women into groups with different risks, grouped the Ki67 scores into categories that did not use the preferred digits as cut-points: 0-7%, 8-17%, 18-27%, 28-37%, 38-57%, and 58-100%.Sensitivity analyses were carried out to establish if the digit preference has altered our results.

Table 1i. Distribution of Ki67 samples by source of Ki67 score, according to ER and HER2 status and calendar period of diagnosis. Ki67
sample date linked to nearest pathology investigation type within ±1 week.

Table 5 . Cohort studies reporting the association with Ki67 score and outcomes in women with early breast cancer
]When adjustment is made for tumor grade only, the rate ratio is reduced by a larger amount than when adjustment is made for any other single variable.This shows that tumor grade is the strongest confounding variable.The median follow-up period started after recurrence, as the aim of this study was to determine the impact of subtype and the year of recurrence on the survival times of recurrent breast cancer.Among the 4 centers in this study, 3 of them used 14% as the cut-off value, while 1 center used 25% as the cut-off value.HR was not explicitly stated however OS was 34.82 months (±0.30) for patients with low ki67 vs 32.41 months (±0..43) for high ki67 and DFS was 37.029 months (±0.476) for patients with low vs 32.242 months (±0.792) for high Ki67 scores significantly lower in patients with high Ki67 scores Women with metastatic disease were identified as follows: no surgery recorded (N=1898) or with a record of metastatic disease within 3 months of breast cancer diagnosis, or a drug usually given for metastatic disease, or palliative radiotherapy within a year of breast cancer diagnosis (N=1226) b Women recorded as receiving neoadjuvant therapy (chemotherapy, endocrine therapy, targeted therapy or radiotherapy) were excluded because comparable staging information was unavailable for them.
q a Supplementary Figure 1.Derivation of study population