Purpose: Proliferative activity defined by Ki-67 staining index (SI) has been correlated with progression and prognosis in a number of malignant tumors including prostate cancer. However, few studies have examined Ki-67 SI in pretreatment diagnostic material from patients treated with definitive radiotherapy. In a prior study, we found that a Ki-67 SI of >3.5% was associated with poorer patient outcome. The goals of this analysis were to validate the prognostic value of Ki-67 SI and this cut point.
Experimental Design: Of 456 assessable patients in Radiation Therapy Oncology Group Protocol 86-10, diagnostic material from 108 patients was available for Ki-67 analysis using MIB-1 antibody. Sixty patients were treated with external beam radiotherapy (EBRT) alone, and 48 patients were treated with short-term androgen deprivation + EBRT. Median follow-up was 9 years for those living. The relationship of Ki-67 with distant metastasis (DM), disease-specific survival (DSS), and overall survival (OS) was examined.
Results: The median Ki-67 SI was 7.1% (range, 0.2–45.5%). The 7.1% cut point was associated with DM and DSS; however, the 3.5% cut point was as strong a determinant and was the focus of this analysis. In Cox proportional hazards regression, Ki-67 SI was independently associated with DM and DSS. When the Ki-67 SI was ≤3.5% and >3.5%, the 5-year risk of DM was 13.5% and 50.8% (P = 0.0005), respectively, and the 5-year risk of DSS was 97.3% and 67.7% (P = 0.0039), respectively. No association of Ki-67 SI with OS was observed.
Conclusions: Higher Ki-67 SI was significantly associated with a greater risk of DM and DSS in locally advanced prostate cancer after definitive EBRT or AD + EBRT.
Quantification of the proportion of cells with nuclear Ki-67 antigen expression is a measure of growth fraction (1, 2, 3, 4) and hence biological aggressiveness in malignancy. This immunohistochemically detected marker has shown potential as an independent correlate of outcome for prostate cancer patients treated with radical prostatectomy in the vast majority of reports (5, 6, 7, 8, 9) . More recently, Ki-67 staining has shown promise for patients treated definitively with external beam radiotherapy (EBRT; Refs. 10, 11, 12 ). Using MIB-1 monoclonal antibody for the determination of Ki-67 activity in archival, formalin-fixed diagnostic material, Cowen et al. (12) found that when the proportion of tumor cells staining for Ki-67 [Ki-67 staining index (SI)] was >3.5%, biochemical and disease failure was significantly higher than when Ki-67 SI was ≤3.5%. Because a number of Ki-67 SI cut points have been described in different study populations, the applicability of one particular cut point and, consequently, the results of immunohistochemical staining of Ki-67 in clinical practice remain questionable.
In this report, the value of the 3.5% Ki-67 SI cut point was evaluated in a cohort of locally advanced patients treated in Radiation Therapy Oncology Group (RTOG) Protocol 86-10 (13, 14, 15) . In this randomized trial of 456 assessable patients, half the patients received EBRT alone, and the other half received short-term neoadjuvant androgen deprivation (AD) + EBRT. The AD was started 2 months before and continued for the duration of EBRT. There were 108 cases available for the Ki-67 SI analyses described here.
MATERIALS AND METHODS
The details of RTOG Protocol 86-10 have been described previously (13) . Of the 108 patients for whom pretreatment diagnostic material was available for Ki-67 SI analysis, 72 samples were from needle biopsies, and 36 samples were from transurethral resections. Confirmation of the pathological diagnosis and Gleason grading were done by the study pathologist (D. J. G.). A Gleason score of 7–10 was observed in 80 cases, and a Gleason score of <7 was observed in 27 cases (Gleason score was not available in 1 case). Clinical category T2 was present in 31 patients, and T3 was present in 77 patients. Radiotherapy alone (EBRT) was administered to 60 patients, and 48 cases received AD + EBRT (see Table 1⇓ ).
Ki-67 Immunohistochemical Staining Procedure and Quantification.
Paraffin-embedded archival sections (4 μm) were deparaffinized in xylene and then rehydrated in graded ethanol. Slides were subsequently heated in 10 mm citrate buffer (pH 6.0) for 50 min using a pressure cooker (BioCare Medical, Walnut Creek, CA). The slides were treated with 0.3% hydrogen peroxide for 5 min and then incubated with the monoclonal antibody MIB-1 (DAKO Corp., Carpinteria, CA) for 10 min. Secondary biotinylated antibody was then applied for 10 min, followed a 10-min incubation with streptavidin peroxidase (DAKO Corp.). After rinsing, slides were stained with diaminobenzidine chromogen solution (ResGen Invitrogen Corp., Carlsbad, CA) and counterstained with routine hematoxylin. Staining was accomplished in three batches, using a DAKO Autostainer (DAKO Corp.). Negative controls for the staining were biopsy slides stained with omission of the primary antibody. Positive controls were normal tonsil sections that had previously been studied by flow cytometry to determine the percentage of proliferating cells.
Nuclear staining of Ki-67 was considered positive. The Ki-67 SI was defined as the percentage of positive nuclei of a total of 2000 tumor cells counted using an eyepiece grid. The positive nuclei were counted by one investigator (R. L.) without prior knowledge of the patient prognosis-related information.
Definition of End Points.
The end points investigated in the analyses were local failure, distant metastasis (DM), overall survival (OS), and disease-specific survival (DSS). The parameters of local failure, DM, and OS are as described in the initial reports (14, 15, 16) . Because no relationship was observed between Ki-67 SI and local failure, the focus of this report was on the other end points. For DSS, failure was defined as death due to prostate cancer, protocol treatment toxicity, or unknown cause of death with active disease. DM was defined as clinical or radiological evidence of disease outside the pelvis. OS and DSS were measured from the date of randomization to the date of death or last follow-up date, if the patient did not fail. Time to DM was measured from the date of randomization to the date of failure, death, or the last follow-up date, if the patient did not fail.
This analysis was performed using the 456 eligible and analyzable patients from RTOG Protocol 86-10. Of these, there were 108 patients who had KI-67 SI determinations available. As of June 30, 2000, the median follow-up in the study cohort was 6 years (range, 5 months to 11.8 years) for all patients and 9 years (range, 6–11.8 years) for living patients.
Based on Ki-67 SI expression, the data were dichotomized at the median value of 7.1% and at 3.5%; the latter was based on prior results (12) . Analysis of the distribution of patients by Ki-67 SI and various potential prognostic factors was conducted by the Pearson χ2 test with the Yates correction factor. Estimates of OS were derived using the Kaplan-Meier method (17) . The cumulative incidence approach was used to estimate DM and DSS (18) . This method adjusts for competing risks, such as dying without the recurrence of prostate cancer. Statistical comparisons of OS, DSS, and DM were made using the log-rank test (19) .
Multivariate Cox proportional hazard models were applied to each of the end points (20) . The initial multivariate analyses were restricted to patients for whom Ki-67 SI was available. The prognostic importance of Ki-67 SI was appraised after adjusting for treatment assignment, clinical stage, age (for the OS model), and Gleason score as fixed covariates (20) . All factors were considered as dichotomous variables and coded as follows: treatment assignment, 0 (radiotherapy alone) versus 1 (radiotherapy + AD); clinical stage, 0 (T2) versus 1 (T3); age, 0 (<75 years) versus 1 (≥75 years); grouped Gleason sums, 0 (sums of 2–6) versus 1 (sums of 7–10); p53, 0 (negative) versus 1 (positive); and Ki-67 SI, 0 (≤3.5%) versus 1 (>3.5%). In addition, a multivariate analysis was performed that considered an interaction between Ki-67 and treatment assignment for each end point. The fitted parameter from the Cox model is used to estimate the relative risk associated with each prognostic variable and corresponding 95% confidence interval. A ratio of 1 would indicate no difference between the two subgroups. The greater the difference is from 1, the greater the difference in the failure rates between the two subgroups. The treatment effect was modeled in such a way that a value < 1 favored the addition of hormones. Ki-67 SI was modeled in a way that a value > 1 indicates a greater risk of failure for patients with Ki-67 SI > 3.5%. All of the statistical comparisons were made with two-tailed tests.
A second type of multivariate analysis adjusted for two additional factors, p53 and missing tumor determinations. Of the 456 assessable patients, 108 (24%) had Ki-67 SI determinations, and 129 (28%) had p53 determinations. Both were available in 79 (17%) patients. There are potential analytical problems due to the missing values. Selection bias may occur, wherein the patients in whom the assays were done do not constitute a random sample from the whole study. As a consequence, the study cohort may have a better or worse outcome than the parent cohort. Moreover, when cases with missing values are excluded from the analysis, the number of patients to be analyzed may be relatively small, compromis-ing the statistical power needed to detect clinically meaningful differences.
To adjust for the problem of missing values in the second multivariate analysis, two variables (instead of one) were used to evaluate each marker. For Ki-67 SI, patients were classed into three categories: determination not done; Ki-67 SI ≤ 3.5%; and Ki-67 SI > 3.5%. For p53, patients were classed into three categories: determination not done; negative; and positive. The first variable for Ki-67 SI would then be 0 (Ki-67 SI ≤ 3.5%/not done) versus 1 (Ki-67 SI > 3.5%), and the second variable would be 0 (Ki-67 SI > 3.5%/not done) versus 1 (Ki-67 SI ≤ 3.5%). The estimated relative risk of Ki-67 SI was figuratively obtained by subtracting out the two variables. The 27 patients without centrally reviewed Gleason scores were excluded, leaving 429 patients for the analysis.
The positive staining of Ki-67 was distinct, and the Ki-67 SI varied significantly from case to case. The Ki-67 SI ranged from 0.2% to 45.5%, with a median of 7.1% and a mean of 8.9%. Table 1⇓ shows that 51% of cases had a Ki-67 SI of ≤7.1%, whereas 34% had a Ki-67 SI of ≤3.5%. By the χ2 test, Ki-67 SI > 7.1% and Ki-67 SI > 3.5% were associated with higher combined Gleason score and younger age (borderline for the 3.5% cut point), but not with clinical T category or assigned treatment (Table 1)⇓ . The focus of the univariate and multivariate analyses described was using the end points of DM, DSS, and OS. No relationship was found between local failure and Ki-67 SI, and these data are not described.
Tables 2⇓ and 3⇓ display the univariate survival results. Both Ki-67 SI >7.1% and Ki-67 SI >3.5% were significantly associated with higher rates of DM and lower rates of DSS. However, Ki-67 SI was not significantly associated with OS. Based on the univariate results, the 3.5% cut point appears to be as good as the median value (7.1%) for the locally advanced study population here. Moreover, the 3.5% cut point is more relevant for the analysis of the typical contemporary patient with earlier disease (12) . Because a major emphasis of this investigation was to validate the prior results showing that the Ki-67 SI 3.5% cut point is predictive of patient outcome, the remaining analyses center on this categorization. Fig. 1⇓ displays the cumulative incidence curves demonstrating the effect of Ki-67 SI on DM and DSS. Also shown are the Kaplan-Meier curves for OS. No relationship was observed between Ki-67 SI status and the study randomization of EBRT versus AD + EBRT, although the statistical power to detect a difference is low.
Multivariate analyses restricted to patients with Ki-67 data showed that patients with a Ki-67 SI > 3.5% had a higher risk of DM and a higher risk of death due to prostate cancer, but Ki-67 SI > 3.5% was not associated with a higher risk of overall death (Table 4)⇓ . No other factor was found to be associated with DM or DSS.
RTOG Protocol 86-10 included 456 assessable patients. Thus, there were 348 cases for whom Ki-67 SI data were missing. A separate set of analyses was performed to determine whether the subgroup analyzed for Ki-67 SI was representative of the patient cohort. Table 5⇓ displays the distribution of pretreatment characteristics of patients by the Ki-67 SI subgroup and the remaining patients. No statistically significant differences were seen. However, patients with Ki-67 data had a borderline increased risk of DM and had significantly lower DSS and OS rates than those with missing Ki-67 data (Table 6)⇓ .
In a previously reported analysis, p53 status was found to be a significant prognostic variable for survival, but it was only available for 79 patients with a Ki-67 determination. As a consequence, a multivariate analysis using Cox proportional hazards regression that adjusted for this population selection effect was done using 429 patients (see “Materials and Methods”). Table 7⇓ shows that after adjusting for population effects and p53 status, in addition to Gleason score, clinical stage, and assigned treatment, Ki-67 was an independent prognostic factor for DM and DSS. Gleason score and p53 were associated with DM and DSS, whereas clinical stage and treatment regimen (EBRT alone versus AD + EBRT) were not. Patient age, clinical T category, Gleason score, and p53 were associated with OS.
Ki-67 expression has been related to patient outcome for a number of tumor types, including prostate cancer. Although there are several reports of the correlation of Ki-67 immunohistochemical staining status and prostate cancer progression, few have examined the predictive value of this biomarker in patients treated with radiotherapy (10, 11, 12) . In addition, the investigation of Ki-67 SI as a potential predictor of prostate cancer patient outcome has previously been done in mainly small, single-institution studies. To our knowledge this is the first study of the expression of this marker in a well-defined group treated in a large multi-institutional randomized trial. As such, there is knowledge of the patient characteristics and outcome of those in whom the marker was not available. Table 6⇓ shows that there were differences in DSS and OS between the patients in whom Ki-67 SI was determined and those in whom it was missing. The cause of this difference is unclear. It is a rare biomarker study in which the features and outcome of those excluded are known and the potential selection bias is acknowledged. Tissue might not have been available for Ki-67 SI analysis because the original archival material was not accessible or because there was insufficient tumor tissue in the retrieved blocks. As has been described previously (15) , it is possible to adjust for the selection bias by inclusion of the patients with missing Ki-67 SI data in multivariate analysis. When this was done (Table 7)⇓ , Ki-67 SI still significantly predicted for DM and DSS.
One of the inherent problems with immunohistochemical analyses is the wide variation in cut points used to define associations with patient outcome. This consideration is particularly apropos for Ki-67 SI analyses. As we have pointed out previously (12) , a number of Ki-67 SI cut points (1–25%) have been found to be predictive of freedom from biochemical or clinical failure in men with prostate cancer. These cut points largely were chosen around the median Ki-67 SI for the cohort studied. The differences in median Ki-67 SI are probably related primarily to the aggressiveness of the tumors but also could be influenced by staining technique. A critical step is antigen retrieval, which could be affected by the original fixation parameters before embedding in paraffin, length of heating, and method of heating.
There are several unique features of the current Ki-67 SI study. First, the tissue was collected from around the country in the setting of a multi-institutional randomized trial. The handling and fixation of the biopsy material were not controlled and therefore reflect current practices in the United States. Second, the staining was performed in an autostainer in three batches. Positive controls for the batches showed identical proportions of proliferating cells. Third, the cohort studied was treated rather uniformly with EBRT ± AD on protocol using conventional techniques. For these reasons, the associations described should be generally applicable in the community.
The other important aspect of the RTOG Protocol 86-10 population is that it was composed of cases with very high-risk characteristics. Consequently, the median Ki-67 SI was 7.1%, which is substantially greater than the 2.3% observed in the patients with favorable- to intermediate-risk features comprising the cohort from M. D. Anderson Cancer Center (MDACC) used in our prior analysis (12) . In the prior analysis, Ki-67 SI was a significant predictor of biochemical failure in both univariate and multivariate Cox proportional hazards regression analyses. Those with a Ki-67 SI of >3.5% were identified as having a particularly poor prognosis, with a freedom from biochemical failure rate of 33%, versus a rate of 76% for those with a Ki-67 SI of ≤3.5%. Of the patients investigated by prostate biopsy for rising prostate-specific antigen (PSA), 65% had documented local persistence of disease. The number of patients with DM in the MDACC study was too small (n = 4) to assess for any correlation with Ki-67 SI.
Despite the inherent differences in the MDACC (12) and RTOG Protocol 86-10 cohorts, there were some parallels in the relationships found. Whereas biochemical failure was not investigated in the RTOG group because this was a pre-PSA era study, Ki-67 SI was significantly associated with DM and DSS. The strong association of Ki-67 at the 3.5% cut point to biochemical failure (relative risk = 2.8) in the MDACC series was likewise robust for the DM and DSS end points (relative risk = 3.41 and 4.24, respectively) in the current study. Thus, although there were pronounced differences in the characteristics of the MDACC and RTOG patient populations, the 3.5% Ki-67 SI cut point was validated. We also found that the median value of 7.1% in the RTOG Protocol 86-10 group was predictive of DM and DSS.
Another finding in the current analysis was that there was no association between Ki-67 SI and local failure. Some possible reasons include the following: (a) Ki-67 SI is a much stronger determinant of DM than it is a predictor of radiation response; (b) the RTOG Protocol 86-10 patient population was very advanced, and in the setting of DM, the documentation of local disease was probably underdocumented; and (c) in the absence of PSA as an early indication of treatment failure in this pre-PSA study, the impetus for prostate biopsies in the absence of obvious clinical local progression would not have been present. Further investigation of Ki-67 SI in patients treated with radiotherapy in the PSA era would resolve these questions.
Also examined herein was the association of Ki-67 SI to OS. Despite the strong relationship of a high Ki-67 SI to the development of DM and death due to prostate cancer, no relation to OS was seen. The data suggest that this may be due to dilution of disease-specific deaths by deaths from intercurrent causes.
In summary, Ki-67 SI at the 3.5% cut point was validated as a significant correlate of outcome for prostate cancer patients treated with EBRT or AD + EBRT. The relationship to DM and DSS described appeared to be independent of whether AD was given and was independent of p53 immunohistochemical staining status. The main drawback of the study is that only 108 cases were available. Further testing in a larger cohort should be undertaken before incorporating Ki-67 SI in future clinical trials.
Grant support: Department of Defense, United States Army Medical Research Grant PC020427, and the Tobacco Grant from the Pennsylvania Department of Health.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Requests for reprints: Alan Pollack, Department of Radiation Oncology, Fox Chase Cancer Center, 333 Cottman Avenue, Philadelphia, PA 19111. Phone: (215) 728-2940; Fax: (215) 728-2868; E-mail:
- Received July 17, 2003.
- Revision received January 21, 2004.
- Accepted March 31, 2004.