Assessment of the American Joint Committee on Cancer Staging (sixth and seventh editions) for clinically localized prostate cancer treated with external beam radiotherapy and comparison with the National Comprehensive Cancer Network risk-stratification method†‡
Nicholas G. Zaorsky MD,
Department of Radiation Oncology, Fox Chase Cancer Center, Philadelphia, Pennsylvania
Presented at the 2012 American Society of Clinical Oncology Genitourinary Cancer Symposium; February 2-4, 2012; San Francisco, CA.
The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute, the National Institutes of Health, or Varian Medical Systems.
The objective of this study was to compare the prognostic value of the sixth and seventh editions of the American Joint Cancer Committee (AJCC) Cancer Staging Manual and the risk-stratification model of the National Comprehensive Cancer Network (NCCN).
Two-thousand four hundred twenty-nine men who received definitive radiotherapy with or without androgen-deprivation therapy (median follow-up, 74 months) were analyzed.
There was a migration of stage II patients to stage I with AJCC seventh edition (stage I increased from 1% to 38%, and stage II decreased from 91% to 55%). One pair-wise comparison (4%) of Kaplan-Meier estimates of biochemical failure, distant metastasis, prostate cancer-specific survival, and overall survival between stages was statistically significant for the AJCC sixth edition. Conversely, 16 of 24 comparisons (67%) were significant for the AJCC seventh edition. With the NCCN risk-stratification model, 9 of 12 comparisons (75%) were significant. Concordance probability estimate (CPE) and standard error (SE) analysis indicated uniform and significant improvement in the predictive power of the AJCC seventh edition versus the sixth edition for all outcomes. CPE ± SE values for the AJCC seventh edition versus the sixth edition were 0.51 ± 0.009 versus 0.59 ± 0.02, respectively, for biochemical failure; 0.54 ± 0.02 versus 0.70 ± 0.05, respectively, for distant metastasis; 0.57 ± 0.009 versus 0.76 ± 0.007, respectively, for prostate cancer-specific survival; and 0.52 ± 0.006 versus 0.57 ± 0.01, respectively, for overall survival. CPE ± SE values for the NCCN model were 0.59 ± 0.02 for biochemical failure, 0.72 ± 0.05 for distant metastasis, 0.80 ± 0.01 for prostate cancer-specific survival, and 0.57 ± 0.01 for overall survival.
The American Joint Committee on Cancer (AJCC) published the first edition of the AJCC Cancer Staging Manual in 1977. It used classifications for T (tumor extent), N (lymph node invasion), and M (presence or absence of metastasis) to stage prostate cancer. Staging is important to: 1) categorize the severity of disease, 2) estimate prognosis, 3) recommend treatment, and 4) aid health care providers and researchers in exchanging information about patients.1 The manual has been revised several times with the advent of new diagnostic tools and treatments.
There are several important changes that have been implemented in seventh edition (AJCC v7).2 For the first time, serum prostate-specific antigen (PSA),3 a predictor of survival, is recognized in prostate cancer staging. Gleason score (GS) is now recognized as the preferred tumor grading system. And perhaps most important, PSA and GS are included in the stage grouping (ie, stages I, II, III, and IV). Together, T-classification, PSA, and GS have been used commonly in various risk-group stratification models4-7 that help physicians guide patients in terms of treatment and risks of recurrence, but they have not appeared together in the AJCC staging until now.
It is unclear, however, whether, these changes have significantly improved the staging system's ability to group patients according to prognosis. The purpose of this investigation was to compare the ability of the sixth and seventh editions (AJCC v68 and AJCC v72) to predict biochemical failure (BF), distant metastasis (DM), prostate cancer-specific survival (PCSS), and overall survival (OS) in men with prostate cancer who received radiotherapy (RT) with or without androgen-deprivation therapy (ADT).9
MATERIALS AND METHODS
Between 1989 and 2006, 2469 men with clinical stage T1-T4, N0/NX-N1, M0 adenocarcinoma of the prostate received definitive RT with or without adjuvant ADT. Three-dimensional, conformal RT (3D-CRT) (July 1992 to June 2001) was received by 1465 men (59%); and intensity-modulated RT (IMRT) (July 2001 to July 2004) was received by 1004 men (41%). The mean follow-up from the completion of RT was 74 months (range, 1-213 months); 70 months (range, 2-108) for the IMRT group and 86 months (range, 1-213 months) for the 3D-CRT group. The techniques used for 3D-CRT and IMRT have been reported previously.10, 11
ADT generally was used at the discretion of the treating physician and consisted of oral antiandrogen given for 1 to 4 months and a luteinizing hormone-releasing hormone agonist administered as depot injections.
Patient Evaluation and Staging
All patients had a history and physical examination, including digital rectal examination, initial serum PSA, and histologic confirmation of adenocarcinoma with a GS. Thirty-one percent of patients had a computed tomography scan of the pelvis, and 88% had a bone scan. T-classification was established by palpation findings only, without upstaging using pathologic or radiographic information. Patients were staged using the AJCC v68 and v72 guidelines.
Univariate analysis was performed using the chi-square test for discrete variables, and the Wilcoxon test was used for continuous variables. The Kaplan-Meier estimation method and a Cox proportional hazards model were used to evaluate the ability of each clinical staging system to predict the time to BF, DM, PCSS, and OS. In the multivariate analysis, the effects of treatment (RT vs RT+ADT) and RT dose (<75.6 grays [Gy] vs ≥75.6 Gy) were accounted for as covariates. The Benjamini and Hochberg adjustment12 was used to produce a series of adjusted P values, and adjusted P value <.05 was considered significant. BF was defined as the PSA nadir plus 2 ng/mL.13 The concordance probability estimate (CPE)14, 15 and its standard error (SE) were used in assessing the predictive accuracy of the model for various endpoints. The closer the CPE is to 0.5, the lower the predictive accuracy of the model; the closer the CPE is to 1, the better the predictive accuracy.
Various patient-related and treatment-related characteristics are listed in Table 1. On the basis of NCCN criteria, the proportion of patients with low-risk disease was 38%, the proportion with intermediate-risk disease was 42%, and the proportion with high-risk disease was 21%. The majority of patients (79%) received RT alone (no ADT) at a median RT dose of 76 Gy. For the men who received ADT, the median duration was 11.2 months (range, 0.4-124 months).
Table 1. Various Patient and Treatment-Related Characteristics, n = 2469
No. of Patients (%)
Abbreviations: ADT, androgen-deprivation therapy; Gy, grays; NCCN, National Comprehensive Cancer Network; PSA, prostate-specific antigen.
Age: Mean [range], y
Radiation dose: Median [range], Gy
NCCN risk group
Patient staging statistics according to AJCC v6 and v7 are listed in Table 2. When transitioning from AJCC v6 to AJCC v7, the proportion of men in stage I increased (from 1% to 38%), the proportion in stage II decreased (from 91% to 55%), and stages III and IV remained the same (6% and 1%, respectively). Regarding the subdivision of stage II into the IIA and IIB subcategories according to AJCC v7, 35% of all patients had IIA disease, and 20% had IIB disease.
Table 2. Staging Statistics According to American Joint Committee on Cancer (AJCC) Cancer Staging Manual Sixth and Seventh Editions
Abbreviations: GS, Gleason score; M, metastasis N, lymph node; PSA, prostate specific antigen; T, tumor; X, not assessable.
Grade 1, well differentiated (GS 2-4), grades 2-4, moderately differentiated (GS 5-6), grades 5 and 6, poorly differentiated (GS 7-10).
T1aN0M0, grade 1
T1a-cN0M0, PSA <10 ng/mL, GS ≤6
T2aN0M0, PSA <10 ng/mL, GS ≤6
T1-T2aN0M0, PSAX, GSX
T1aN0M0, grade 2-4
T1bN0M0, any grade
T1a-cN0M0, PSA <20 ng/mL, GS 7
T1cN0M0, any grade
T1a-cN0M0, PSA, ≥10 & <20 ng/mL, GS ≤6
T1N0M0, any grade
T2aN0M0, PSA ≥10 & <20 ng/mL, GS ≤6
T2N0M0, any grade
T2aN0M0, PSA <20 ng/mL, GS 7
T2bN0M0, PSA <20 ng/mL, GS ≤7
T2bN0M0, PSAX, GSX
T2cN0M0, any PSA, any GS
T1-2N0M0, PSA ≥20 ng/mL, any GS
T1-2N0M0, any PSA, GS ≥8
T3N0M0, any grade
T3a-bN0M0, any PSA, any GS
T4N0M0, any grade
T4N0M0, any PSA, any GS
Any T, N1, M0, any grade
Any T, N1, M0, any PSA, any GS
Any T4, any N, M1, any grade
Any T4, any N, M1, any PSA, any GS
Figure 1 illustrates Kaplan-Meier estimates of BF (Fig. 1A), DM (Fig. 1B), PCSS (Fig. 1C), and OS (Fig. 1D) for AJCC v6 (Fig. 1, left column) and AJCC v7 (Fig. 1, right column). Table 3 provides paired P value comparisons between any 2 Kaplan-Meier survival curves from Figure 1. The significant (ie, <.05) paired P values are indicated. In the AJCC v6 staging, overall differences in DM, PCSS, and OS were not significant, with the exception of outcome prediction of PCSS for patients with stage II and III disease. Conversely, in the AJCC v7 staging, paired P value comparisons were more likely to be significant. This is a marked improvement compared with AJCC v6. Figure 2 illustrates Kaplan-Meier estimates of BF (Fig. 2A), DM (Fig. 2B), PCSS (Fig. 2C), and OS (Fig. 2D) for AJCC v7 stage IIA versus stage IIB. Substratification between stages IIA and IIB did not provide more prognostic information. Figure 3 illustrates Kaplan-Meier estimates of BF (Fig. 3A), DM (Fig. 3B), PCSS (Fig. 3C), and OS (Fig. 3D) based on NCCN risk group. Table 4 provides paired P value comparisons between any 2 Kaplan-Meier curves from Figure 3. NCCN risk-group stratification was significant for most of the Kaplan-Meier curves in the outcome measures studies, particularly for OS.
Table 3. Paired P Value Comparison of Disease Stage According to American Joint Committee on Cancer (AJCC) Guidelines: AJCC Cancer Staging Manual Sixth and Seventh Editions
Table 5 lists the CPEs and SEs in evaluating the ability of AJCC v6, AJCC v7, and the NCCN risk groups to prognosticate BF, DM, PCSS, and OS after controlling for treatment and dose effects. The results indicate uniform improvement in the predictive power of the model based on AJCC v7 compared with AJCC v6 for all endpoints, particularly for DM and PCSS. NCCN risk-group stratification was equally predictive of BF and OS, and it was more predictive of DM and PCSS. Figure 4 illustrates the data presented in Table 5.
Table 5. Concordance Probability Estimates and Their Respective Standard Errors in the American Joint Committee on Cancer (AJCC) Cancer Staging Manual Sixth and Seventh Editions and National Comprehensive Cancer Network Assessmenta
AJCC 6th Edition
AJCC 7th Edition
Abbreviations: CPE, concordance probability estimate; NCCN, National Comprehensive Cancer Network; SE, standard error.
The closer the CPE to 1, the better the predictive power of the model; the closer the CPE to 0.5, the worse the predictive power.
Prostate cancer-specific survival
Staging is important for radiotherapy, because it guides treatment recommendations, such as the use ADT, RT dose, and treatment volume (ie, prostate only vs prostate and pelvis). Although several relatively common models that subdivide patients into low-risk, medium-risk, and high-risk groups have been proposed,16, 17 there also are several models that are more sophisticated but less commonly used.5, 18-22 The results from those studies have been integrated in designing AJCC guidelines. In the current study, we compared the ability of AJCC v68 and v72 to predict BF, DM, PCSS, and OS in men who received RT with or without ADT.
The first major improvement with AJCC v7 is better prognostication between stages I and II. According to AJCC v6, 2263 men (91%) in our cohort had stage II disease, and the great majority had T1c tumors. Nine-hundred eight patients (40%) with stage II disease migrated from AJCC v6 stage II to AJCC v7 stage I. The remaining 1355 men (60%) with AJCC v6 stage II disease remained in stage II. This migration positively influenced the prognostic value of AJCC v7. Patients with stage I disease had significantly longer times until BF, DM, and OS compared with patients with AJCC v7 stage II disease (Table 3, Fig. 1).
The second major advancement associated with AJCC v7 is the improved prognostic value and ability to predict for BF, DM, PCSS, and OS (Tables 3 and 5). The greatest gains were observed for the DM and PCSS endpoints. The CPE increased from 0.54 to 0.70 for DM and from 0.57 to 0.76 for PCSS. When considering that a CPE of 0.5 is associated with purely random ranking, these improvements are a remarkable accomplishment that should be applauded.
These improvements attributable to several important changes that distinguish AJCC v7 from AJCC v6. First, PSA is recognized in TNM staging. Serum PSA is an important prognosticator in patients who have a benign prostate examination,23 and lower PSAs have been associated with more favorable pathologic findings in clinical stage T1c cancers.24 Second, the AJCC v7 TNM system obligates preferential use of the GS in assessing histopathologic tumor grade. The GS is an important determinant of outcome and has been identified as superior to other histopathologic reports.7, 25, 26
Problems still exist with AJCC v7 staging. First, there are still too many “intermediate-risk” patients (ie, stage II [55%]), and there is a lack of substratification of these patients. The clustering of patients in stages I and II probably is secondary to the majority of patients presenting with localized or impalpable disease but an increased PSA.27 Moreover, patients with stage IIA disease do not have a statistically significant difference in the rate of BF, DM, PCSS, or OS compared with patients who have stage IIB disease (Fig. 2). Second, AJCC v7 does not alter the number of patients in stages III and IV, a finding demonstrated in other patient populations.28 The proportions of patients were 6% in stage III and 1% and in stage IV, which were small compared with the proportions in stages I and II. Third, AJCC v7 does not have a high predictive accuracy for BF or OS (CPE, 0.59 and 0.57, respectively) (Table 5). This is especially true for stratifying and predicting the outcomes of patients with stage II and III disease (Fig. 1, right column). Fourth, the NCCN risk groups have statistically greater predictive accuracy compared with AJCC v7 for PCSS (0.8 vs 0.76, respectively) (Table 5). Thus, the NCCN model is superior to the AJCC v7 and remains the preferred method for risk-based clinical management of prostate cancer with RT.
Clinical T-classification, GS, and initial PSA have been identified as independent predictors of BF and are used commonly in staging systems.16 The TNM staging system is a surgical staging system and does not include many other important pretreatment characteristics. Future models will use more advanced data analysis in predicting outcome. For example, to improve the TNM staging system, upstaging may be considered for patients who have low-risk or favorable intermediate-risk disease and >50% positive core biopsies.28-30 Second, staging systems may include the number of GS 4 or 5 in biopsy specimens, because it has been demonstrated that this value is a predictor of BF and DM after RT, independent of GS.31 Third, patients with a PSA velocity >2 ng/mL in the year before radical prostatectomy or RT may be upstaged, because it has been demonstrated that these men have poorer PCSS.32, 33
Future systems also may integrate patient-specific biomarkers into prognostication. Patient-specific and disease-specific treatment is demanded by the US government's impetus for expanding comparative effectiveness research (CER).34 CER focuses on developing personalized medicine by examining the racial, ethnic, socioeconomic, and geographic variations in care that affect health outcomes.35, 36 Prostate cancer is a high-impact site of CER because of its high prevalence, the many treatment options available, and the emerging biomarkers used in its staging and treatment.37
The future of CER for prostate cancer therapy may integrate specific cancer-specific and patient-specific biomarkers into staging, including B-cell leukemia/lymphoma 2 (Bcl-2),38-40 Bcl-2-associated X protein (Bax),38, 39 the Bcl-2/Bax ratio,39, 41 CD44,42 e-cadherin,42 p53 (tumor suppressor),42 p21/waf1 (cyclin-dependent kinase inhibitor),41 cytochrome c oxidase subunit II (COX-2),43, 44 p53b E3 ubiquitin protein ligase homolog (MDM2),45 Ki-67 (proliferation marker),46, 47 p120 (catenin-associated protein),46 proliferating cell nuclear antigen (PCNA),46 DNA microarrays,48 and cytochrome p450 polymorphisms.49 Thus, although the TNM staging system may be used to describe almost any patient with prostate cancer, its generalizability precludes its specificity and prognostic capability. Future biomarker-based staging systems, although not universal, may be more specific and may have greater predictive accuracy.
In conclusion, AJCC v7 is a major improvement over AJCC v6, because it distributes patients better among the stages and is more prognostic. Further improvements are needed, because the majority of men (55%) have stage II disease, and the substratification into stages IIA and IIB was not prognostic. The NCCN model is superior to the AJCC v7 and remains the preferred method for risk-based clinical management of prostate cancer with RT.
We thank Dr. Gerald Hanks for his leadership in the establishment of the Fox Chase Cancer Center database for the treatment of prostate cancer.
This publication was supported by grant P30 CA006927 from the National Cancer Institute/National Institutes of Health and by a departmental grant from Varian Medical Systems.