A Population-Based Validation of the American Joint Committee on Cancer Melanoma Staging System
http://www.100md.com
《临床肿瘤学》
The Melanoma Program of the Abramson Cancer Center of the University of Pennsylvania
Department of Biostatistics and Epidemiology
Department of Medicine
Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine, Philadelphia, PA
Comprehensive Cancer Center, University of Alabama at Birmingham, AL
ABSTRACT
PURPOSE: A major revision of the American Joint Committee on Cancer (AJCC) stages for melanoma was implemented in 2002 after its validation in multinational cohorts including patients from cancer centers and cooperative groups. This staging system has not been validated in a US population-based cohort.
PATIENTS AND METHODS: We used 41,417 patients with primary invasive cutaneous melanoma diagnosed between 1988 and 2001 from the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) cancer registry to validate the revised AJCC staging system. Survival rates computed from stage-specific Kaplan-Meier curves (time to melanoma-specific death) were compared with the survival rates from 17,600 patients in the original AJCC validation study.
RESULTS: In the SEER cohort, 65% of reported melanomas were 1.00 mm in thickness and 8.7% were more than 4.00 mm compared with 39% and 10% in the AJCC cohort (P < .001), respectively. AJCC stages were able to discriminate among SEER patient groups with different prognosis. However, SEER survival rates were significantly higher than those in the AJCC study and notably so in patients with T1a lesions ( 1 mm without ulceration). This population-specific effect remained significant after controlling for lesion thickness in all substages except stage IIA.
CONCLUSION: Although this national population-based study validates the most recent revision of AJCC stages for melanoma, it emphasizes that survival rates are population specific and found them to be generally higher for SEER compared with AJCC patients. Population-specific survival rates should be used in study designs and decisions about patient-specific interventions.
INTRODUCTION
In 2001 the American Joint Commission on Cancer (AJCC) published the most recent revision of the staging system for cutaneous melanoma.1 This revision is now codified in the sixth edition of the AJCC Cancer Staging Handbook.2 Before its publication, this staging system was validated using a large, international cohort of 17,600 patients3 from 13 cancer centers and cooperative groups wherein data were available on the prognostic factors used in the proposed revision. As part of the validation, clinical and pathologic attributes were evaluated to identify those that were independent prognostic factors. The attributes tested were those identified from previous studies to be important, reproducible, and readily ascertained. Tumor thickness and ulceration were found to be the most important prognostic factors in patients with stage I and II primary lesions. In those with stage III disease, the number of involved regional nodes, the quality of involvement (microscopic at sentinel node biopsy or macroscopic when clinically and histologically positive), and primary tumor ulceration were the most important adverse predictors. In stage IV disease, the site(s) of distant metastasis and serum lactate dehydrogenase level were the two prognostic factors selected.
The AJCC staging and the TNM classification systems have played important roles in clinical research and patient care by identifying factors that are associated with prognosis and providing estimates of survival probabilities for patients with varying prognosis. This information is used by clinical trialists to stratify patients and by clinicians to inform clinical decision making and address prognosis with individual patients. Because the accuracy of this information is of great importance, it is necessary to investigate the generalizability of these classification systems in different populations. The authors of the AJCC staging system noted that their cohort was likely not representative of the general population; they commented that "varying referral patterns and clinical trial entry criteria to these tertiary medical centers and cooperative groups might have skewed the patient distribution overall compared to the general population."3
We have conducted an independent validation study using data from the Surveillance, Epidemiology, and End Results (SEER) registry supported by the National Cancer Institute to examine the transportability of the AJCC staging system to a population-based cohort.4 The cohort included all patients (n = 41,417) with primary invasive cutaneous melanoma with complete data on prognostic factors required to stage patients according to the AJCC staging criteria who were diagnosed between 1988 and 2001 in 11 SEER regions within the United States. In this validation study, we investigate the ability of the TNM and AJCC staging classifications for melanoma to discriminate among patients based on their survival distributions and to predict survival rates for the SEER patients.
PATIENTS AND METHODS
Data for SEER patients were obtained from the SEER Cancer Incidence Public-Use Database, November 2003 Submission. Data for the AJCC patients were obtained from the electronic database used for the AJCC validation study.1,3 These cohorts are referred to as the SEER and AJCC cohorts, respectively. The registries, centers, and organizations that have contributed patients to the SEER and the AJCC cohorts are listed in Table 1. The potential overlap between these cohorts is limited to the University of Washington (Seattle), Eastern Cooperative Oncology Group, and Southwest Oncology Group, and this overlap represents at most 5% of the AJCC stage I to III patients.
Patients in the SEER cohort were diagnosed with invasive cutaneous melanoma between 1988 and 2001 in the following SEER regions: San Francisco-Oakland standard metropolitan statistical area, Connecticut, Metropolitan Detroit, Hawaii, Iowa, New Mexico, Seattle (Puget Sound), Utah, Metropolitan Atlanta, San Jose-Monterey, and Los Angeles. During this period, there were 57,230 individuals who had a diagnosis of primary melanoma (International Classification of Diseases for Oncology morphology codes between 872.0 and 879.9) of a lesion on the lip, eyelid, external ear, face, scalp, neck, trunk, limbs shoulder, hip, or skin not otherwise specified. Patients were excluded if they did not receive definitive treatment for their melanoma (n = 2,818; 4.9%). In addition, 11,239 patients were excluded because of incomplete data on one or more of these four variables: nonspecific or unknown SEER extent of disease classification (3.1%), unknown tumor thickness (11.9%), missing information on examination of lymph nodes (0.5%), and unknown vital status and/or survival time (2.2%). Finally, patients were excluded when thickness and Clark’s level were inconsistent (n = 1,756; 3.1%). After all exclusions, 41,417 patients (72.4%) were available for this study. Three variables (sex, age, and anatomic site) were complete for all 57,230 patients and these variables were used to compare those patients included in and excluded from the analysis. There were no significant differences between these two groups on these three characteristics (data not presented). Patients identified for the AJCC cohort included 30,450 patients seen between 1940 and 1999. Patients with missing data on variables required for staging were excluded and the remaining 17,600 patients (58%) were included in this study.3
The AJCC staging system is based on the TNM classification of melanomas. The T classification requires information on thickness, ulceration, and Clark’s level and all three variables were available in the SEER data. The N classification requires information on the number of regional nodes containing metastases as well as the presence or absence of in-transit metastasis and/or satellites. The SEER data include the number of regional lymph nodes removed for staging purposes as well as the number of positive lymph nodes. An indication of clinically evident in-transit metastasis or satellites was also available from the SEER description of the clinical extent of the lesion. However, information on the clinical status of the regional nodes was not available. We assumed that patients who had no nodes examined had no clinical evidence of nodal involvement at diagnosis. The M classification requires information about the sites of distant metastasis and serum lactate dehydrogenase levels and these data are not available in SEER. Consequently, for this study all patients with a distant metastasis are classified as AJCC stage IV without additional subclassification.
For all patients, definitive surgery is classified as biopsy of the primary tumor and excision (less than a wide excision and < 1 cm margin), wide excision, radical excision, major amputation, and surgery not otherwise specified. For all patients diagnosed after 1997, information on the scope of regional lymph node surgery was added to the SEER database to accommodate information from those who had a sentinel lymph node procedure, which was followed-up when positive by node dissection. The total number of positive regional lymph nodes was recorded.
Several other important prognostic factors are available in the SEER database. Patients were characterized by sex, age at diagnosis, and race, whereas anatomic site and histogenetic type further characterize their melanomas. Age and anatomic site were recoded as was done in the original AJCC validation study. Age groups were defined using deciles (0 = birth to 9 years; 1 = 10 to 19 years, and so on). Lesions were characterized as either axial or extremity.
Survival time was defined as the time between diagnosis and melanoma-specific death. For those alive at last follow-up, the survival time was censored on that date. For those who died as a result of other causes or who were lost to follow-up, survival times were censored at the date of death or date of last contact, respectively. In contrast to the SEER cohort, in which all patients were diagnosed between 1988 and 2001, 71% of the AJCC patients were diagnosed with melanoma between 1988 and 1999, and the remaining 29% were diagnosed between 1940 and 1987. Consequently, the lengths of follow-up differed between the two cohorts. Among SEER patients alive at the time of last follow-up (n = 33,728), the average follow-up time was 61.7 months, and 87.6% (n = 29,537) had at least 1 year of follow-up, 76.0% (n = 25,622) had at least 2 years of follow-up, 46.2% (n = 15,513) had at least 5 years of follow-up, and 12.7% (n = 4,266) had at least 10 years of follow-up. For those who died as a result of melanoma, the average time to death was 30.2 months. For AJCC patients alive at the time of last follow-up (n = 13,530), the average follow-up time was 61.9 months, and 78.5% (n = 10,615) had at least 1 year of follow-up, 65.1% (n = 8,814) had at least 2 years of follow-up, 40.2% (n = 5,434) had at least 5 years of follow-up, and 16.3% (n = 2,200) had at least 10 years of follow-up. For those who died as a result of melanoma, the average time to death was 38.1 months.
We used three quantitative methods to compare prognostic discrimination in the two cohorts. The first addressed prognostic separation at the landmark times of 1, 2, 5, and 10 years, computed as the difference between the best and worst survival rate at a specific time.5 The second used the variability of stage-specific survival rates relative to the overall survival rate at each of the landmark times, computed as the average squared deviations of the four survival rates from the overall survival rate at a particular time. The third used the AJCC stage-specific hazard ratios (HRs) from the Cox model that characterized the risk of melanoma-specific death for stage II, III, and IV compared with stage I. For all three methods, a larger value for the statistic means greater discrimination among AJCC stages (each using a different standard).
All survival distributions were estimated using the Kaplan-Meier procedure. Survival rates were obtained from these distributions, and the CIs were computed using Greenwood’s formula. Risk ratios for poor prognostic factors (male, axial lesions, thicker lesions, older patients, and higher Clark level) were used to compare the SEER cohort with the AJCC cohort. Unadjusted and adjusted hazard ratios were obtained from the univariate and multivariate Cox proportional hazards models. All survival analyses were done with Proc PHREG in SAS software (SAS Institute, Cary, NC).6
RESULTS
Discrimination Among SEER Patients by AJCC Staging
To assess the ability of the AJCC stages to discriminate among patients with different prognoses in the US population-based cohort, SEER patients were classified according to the AJCC staging criteria; the stage-specific survival curves and the overall survival curve were computed and compared with those of the AJCC patients (Fig 1). We observed the same relationship between stage and prognosis in both cohorts: stage I defined those with the best prognosis, stage IV defined those with worst prognosis; stage II defined those with a better prognosis than stage III.
Generally, at each landmark time the prognostic separation (the difference between the best and worst survival rates) in SEER was similar to the corresponding AJCC separation (Table 2). For example, 10-year survival rates ranged from 14.1% to 94.3% (separation, 80.2%) compared with 6.8% to 83.6% (separation, 76.8%) for SEER and AJCC, respectively. There was more heterogeneity among the SEER stage-specific survival rates, particularly for 10-year rates, where the variability was 1.7-fold higher compared with those for AJCC. The stage-specific HRs for stages II to IV compared with stage I were computed (Table 3). The HRs increased with increasing stage and were significantly higher for each subsequent stage. Of note, the stage-specific HRs were 2.3- to three-fold higher for the SEER cohort compared with the AJCC cohort. On the basis of each criterion, the AJCC stages identified groups of SEER patients with differential prognosis at least as well as, and in some cases better than, in the AJCC cohort. A similar pattern was observed with the survival curves for subgroups of patients classified by pathologic stage within each AJCC stage (Fig 2).
One major change in the revision of the AJCC staging system was the inclusion of the presence or absence of ulceration of the primary tumor as a prognostic factor. Figure 3 depicts SEER and AJCC survival curves for patients stratified by thickness ( 1.00, 1.01 to 2.00, 2.01 to 4.00, and > 4.0 mm from top to bottom, respectively) and ulceration (present and not present). The pattern of upstaging by ulceration in the survival curves originally observed in the AJCC cohort (Fig 3B) was also seen for the SEER patients (Fig 3A): survival for patients with ulcerated melanomas was diminished such that these patients had rates equivalent to thicker melanomas without ulceration. For SEER patients, the HR for a specific thickness group with ulceration did not differ from that of the next higher thickness group without ulceration (Table 4)
SEER and AJCC Survival Rates
Survival rates were computed using the SEER data for each AJCC stage to assess the degree to which the AJCC survival rates predict survival rates in the SEER cohort. To facilitate comparison of survival rates from the SEER cohort and AJCC cohort, the 1-, 2-, 5-, and 10-year survival rates and their SEs (as tabulated in Balch et al1) are presented in Table 5 by pathologic stage. The majority of the SEER survival rates exceed the corresponding AJCC survival rate and differences were not specific to any one landmark time (see Fig 4).
Prognostic Factors and Patient Mix in the SEER and AJCC Cohorts
The characteristics of the SEER and AJCC patients and their lesions were examined to identify any differences in patient mix that potentially could explain the observed differences in the predicted survival rates. Tables 6, 7, and 8 present the characteristics of SEER patients by AJCC stages. The differences between the two cohorts were summarized using a risk ratio for each prognostic factor, for which the risk ratio was defined as the proportion of patients in the SEER cohort with the category associated with poor prognosis divided by the corresponding proportion in the AJCC cohort (Fig 5). For three of the five prognostic factors (age, Clark level, and axial lesions), SEER patients were generally more likely to be in the poor-risk category. For example, the risk ratios for age were all 1.2 (and as high as 1.8), characterizing the greater proportion of patients 60 years of age or older in SEER compared with AJCC patients. Generally, the risk ratios for sex were approximately equal to 1 (ie, the proportions of male patients were approximately equal for the two cohorts). For thickness, however, AJCC patients generally were more likely to be in the poor-risk category. In particular, the risk ratios were 0.5 and 0.8 for T1a and T1b (thin lesions), respectively, and the proportion of patients with thicker lesions (> 0.75 mm) within the thin range were two-fold (T1a) and 1.3-fold (T1b) higher for the AJCC cohort compared with the SEER cohort, respectively.
Thickness and Prognosis
Thickness is considered to be the strongest univariate predictor of disease-specific survival in patients with melanoma. The distributions of lesion thickness for SEER and AJCC are presented in Figure 6 and these distributions differ particularly with respect to the thinner lesions. As is apparent in Figure 7, for melanomas 5 mm, 10-year survival rates decrease linearly with thickness with similar slopes. Nevertheless, for patients with lesions of the same thickness, SEER survival rates are higher than the AJCC rates, consistent with the pattern in the stage-specific rates. Cox regression models including cohort and thickness were used to investigate whether thicker lesions in the AJCC cohort could explain the higher mortality risk observed within each AJCC substage (data not shown). With the exception of stage IIA, there was a significant difference in mortality risk between the SEER and AJCC cohorts after controlling for lesion thickness. Stage IA lesions had the largest adjusted cohort effect; AJCC stage IA patients had a four-fold higher mortality risk compared with the SEER stage IA patients.
DISCUSSION
The identification of prognostic factors and their use in estimating patients’ outcomes are important tools in clinical research and practice.7 Most published prognostic studies present exploratory studies wherein prognostic factors are assessed using patient populations that are available for study. Their implementation in practice awaits additional validation and/or replication. The two major criteria for validation of a prognostic model are accuracy and generalizability.8 A prognostic model, such as the AJCC staging system, is accurate when the model is able to separate patients’ outcomes (discrimination) and when the predicted outcomes are similar to observed outcomes (calibration).9 An accurate model is generalizable if the model’s accuracy can be replicated in a similar independent sample (reproducible) or is shown to be accurate in a related population sample (transportable).
In 2002, a revised version of the AJCC staging system for cutaneous melanoma was published that included stage-specific survival rates.1 It included a multicenter validation study that demonstrated the significance of the prognostic factors used in the AJCC staging system.3 This study demonstrated a statistically significant difference among the stage-specific and substage-specific survival curves. The 10-year survival rates for TNM classes ranged between 87.9% for T1a patients and 15.0% for N2b patients.3 Thickness and ulceration were shown to be statistically significant independent prognostic factors. The AJCC study established that thickness, ulceration, age, site, and level were statistically significant and independent prognostic factors for those melanoma patients without evidence of regional or disseminated metastatic disease. Nodal status, thickness, ulceration, site, and age were statistically significant and independent prognostic factors for melanoma patients without clinical evidence of nodal metastasis whose regional lymph nodes were pathologically staged after sentinel or elective lymphadenectomy.
In this article we have presented an evaluation of the accuracy and generalizability of the AJCC staging system using population-based data on melanoma patients collected in 11 US sites as part of the National Cancer Institute’s SEER program. The significant differences seen among the AJCC stage-specific survival curves were also seen in the SEER stage-specific survival curves. Hence, AJCC stages are discriminative. However, survival rates were generally better for SEER patients than for AJCC patients in all stages, notably among those with thinner lesions, in whom we observed a 10-year survival rate that was 10% higher for those with T1a lesions (approximately 60% of all SEER melanoma patients). The observed differences in the survival rates indicate that the AJCC staging system is imperfectly calibrated with respect to the general US population as reflected in SEER.
There are a number of potential reasons for the difference in survival rates in the AJCC and SEER populations. As we have noted, the SEER cohort differs from the AJCC cohort in several ways. The SEER cohort is population based, given that all SEER sites obtain data on all patients diagnosed within their region. The SEER cohort is also more contemporary; patients included in the present study were diagnosed after 1987, when thickness became available in the SEER data set, whereas the AJCC cohort had patients diagnosed as early as 1940. However, when we compared the stage-specific survival rates of AJCC patients diagnosed between 1985 and 1999 with those of the SEER patients, the same differential survival rates were seen (data not shown). Differences in the length of follow-up are not likely to explain differential survival. Although the SEER cohort has a lower proportion of people with 10 years of follow-up, the number of patients with at least 10 years of follow-up is greater (4,509 and 2,200 for SEER and AJCC, respectively). We observed that SEER patients had a mix of unfavorable and favorable prognostic factors that distinguished them from the AJCC patients. On the one hand, SEER patients were older and their lesions were more likely to be axial and Clark level IV or V—all of which are associated with poorer survival. On the other hand, they were more likely to have thinner lesions, which are prognostic of better survival. Because lesional thinness is a strong, favorable prognostic factor, its higher prevalence in SEER melanomas may contribute to better outcome among low-stage SEER patients. However, whether we addressed outcome by analyzing subsets of thickness within the AJCC thickness ranges or by using an alternative classification strategy based on a prognostic tree (data not shown), we observed better prognosis among SEER patients compared with AJCC patients in the same risk group.
That we were unable to explain the differences in survival rates between the two cohorts by differences in lesion thickness, temporal trends, or any of the other AJCC prognostic factors indicates that there are likely differences between the two cohorts on unmeasured factors. In addition, there remains heterogeneity among patients within the AJCC stages that can be resolved with a more informative set of biomarkers. We and others have evaluated a number of biomarkers that are not currently used in AJCC staging, and are not available in SEER, which may be both more biologically plausible and predictive.10-19 These include variables that should be considered in the next revision of the AJCC staging system, and that are readily ascertained in routine histology (eg, growth phase, dermal mitotic count, and the presence, location, and density of tumor-infiltrating lymphocytes) or by current immunohistologic methods.20
We conclude that the AJCC staging system discriminates well among melanoma patients with different prognoses and includes prognostic factors that are significant predictors of disease-specific survival. It provides a classification scheme that may be adequate to stratify for clinical trials patients referred to cancer centers and comparable institutions. However, the survival rates predicted by AJCC staging are specific for its population, underestimate survival in the US population as reflected in SEER, and may not reflect accurately the survival rates in other populations.12,21 The SEER survival rates for the AJCC stages that we have presented in this article (Table 5) provide data that potentially can be used in trial design, clinical management, and patient counseling for a population thought to reflect the general US population of melanoma patients.
Authors’ Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
Acknowledgment
We thank April Fritz, Manager, Data Quality, SEER Program of the Division of Cancer Control and Population Sciences of the National Cancer Institute, for help with the interpretation of the SEER data; and all of the institutions, organizations, and cooperative groups that contributed patients to the AJCC database used in developing the AJCC staging system for melanoma.
NOTES
Supported in part by Grant No. CA-093372 (M. Herlyn, principal investigator), SPORE on Skin Cancer.
Presented in part at the 40th Annual Meeting of the American Society of Clinical Oncology, New Orleans, LA, June 5-8, 2004.
Authors’ disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Balch CM, Buzaid AC, Soong SJ, et al: Final version of the American Joint Committee on caner staging system for cutaneous melanoma. J Clin Oncol 19:3635-3648, 2001
Greene F, Page D, Fleming I, et al (eds): AJCC Cancer Staging Handbook (ed 6). New York, NY, Springer, 2002
Balch CM, Soong SJ, Gershenwald JE, et al: Prognostic factors analysis of 17,600 melanoma patients: Validation of the American Joint Committee on Cancer melanoma staging system. J Clin Oncol 19:3622-3634, 2001
Surveillance, Epidemiology, and End Results (SEER) Program Public-Use Data (1973-2001), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch, released April 2004, based on the November 2003 submission. http://www.seer.cancer.gov/publicdata/
Altman DG, Royston P: What do we mean by validating a prognostic model? Stat Med 19:453-473, 2000
SAS Institute Inc: SAS/STAT User’s Guide, Version 8. Cary, NC, SAS Institute Inc, 1999
Mackillop WJ: The role of prognosis in cancer medicine, in Gospodarowicz MK, Henson DE, Hutter RVP, et al (eds): Prognostic Factors in Cancer, (ed 2). New York, NY, Wiley-Liss, 2001
Justice AC, Covinsky KE, Berlin JA: Assessing the generalizability of prognostic information. Ann Intern Med 130:515-524, 1999
Harrell FE: Regression Modeling Strategies. New York, NY, Springer-Verlag, 2001
Clark WH, Elder DE, Guerry D, et al: Model predicting survival in stage I melanoma based on tumor progression. J Natl Cancer Inst 81:1893-1904, 1989
Gimotty PA, Guerry D, Ming M, et al: Thin primary cutaneous malignant melanoma: A prognostic tree for 10-year metastasis is more accurate than American Joint Committee on Cancer Staging. J Clin Oncol 22:3668-3676, 2004
McKinnon JG, Yu XQ, McCarthy WH, et al: Prognosis for patients with thin cutaneous melanoma. Cancer 98:1223-1231, 2003
Mansson-Brahme E, Carstensen J, Erhardt K, et al: Prognostic factors in thin cutaneous malignant melanoma. Cancer 73:2324-2332, 1994
Finley JW, Gibbs JF, Rodriguez LM, et al: Pathologic and clinical features influencing outcome of thin cutaneous melanoma: Correlation with newly proposed staging system. Am Surg 66:527-531, 2000
Oliveira-Filho RS, Ferreira LM, Biasi LJ, et al: Vertical growth phase and positive sentinel node in thin melanoma. Braz J Med Biol Res 36:347-350, 2003
Gromet MA, Epstein WL, Blois MS: The regressing thin malignant melanoma: A distinctive lesion with metastatic potential. Cancer 42:2282-2292, 1978
Cook MG, Spatz A, Brocker E, et al: Identification of histological features associated with metastatic potential of thin (< 1.0 mm) cutaneous melanoma with metastases: A study on behalf of the EORTC Melanoma Group. J Pathol 197:188-193, 2002
Lefevre M, Vergier B, Balme B, et al: Relevance of vertical growth pattern in thin level II cutaneous superficial spreading melanomas. Am J Surg Pathol 27:717-724, 2003
Massi D, Franchi A, Borgognoni L, et al: Thin cutaneous malignant melanomas (< or =1.5 mm): identification of risk factors indicative of progression. Cancer 85:1067-1076, 1999
Alonso SR, Ortiz P, Pollan M, et al: Progression in cutaneous malignant melanoma is associated wsith distinct expression profiles: A tissue microarray-based study. Am J Pathol 164:193-203, 2004
Eigentler TK, Buettner PG, Leiter U, et al: Impact of ulceration in stages I to III cutaneous melanoma as staged by the American Joint Committee on Cancer Staging System: An analysis of the German Central Malignant Melanoma Registry. J Clin Oncol 22:4376-4383, 2004(Phyllis A. Gimotty, Jeffr)
Department of Biostatistics and Epidemiology
Department of Medicine
Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine, Philadelphia, PA
Comprehensive Cancer Center, University of Alabama at Birmingham, AL
ABSTRACT
PURPOSE: A major revision of the American Joint Committee on Cancer (AJCC) stages for melanoma was implemented in 2002 after its validation in multinational cohorts including patients from cancer centers and cooperative groups. This staging system has not been validated in a US population-based cohort.
PATIENTS AND METHODS: We used 41,417 patients with primary invasive cutaneous melanoma diagnosed between 1988 and 2001 from the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) cancer registry to validate the revised AJCC staging system. Survival rates computed from stage-specific Kaplan-Meier curves (time to melanoma-specific death) were compared with the survival rates from 17,600 patients in the original AJCC validation study.
RESULTS: In the SEER cohort, 65% of reported melanomas were 1.00 mm in thickness and 8.7% were more than 4.00 mm compared with 39% and 10% in the AJCC cohort (P < .001), respectively. AJCC stages were able to discriminate among SEER patient groups with different prognosis. However, SEER survival rates were significantly higher than those in the AJCC study and notably so in patients with T1a lesions ( 1 mm without ulceration). This population-specific effect remained significant after controlling for lesion thickness in all substages except stage IIA.
CONCLUSION: Although this national population-based study validates the most recent revision of AJCC stages for melanoma, it emphasizes that survival rates are population specific and found them to be generally higher for SEER compared with AJCC patients. Population-specific survival rates should be used in study designs and decisions about patient-specific interventions.
INTRODUCTION
In 2001 the American Joint Commission on Cancer (AJCC) published the most recent revision of the staging system for cutaneous melanoma.1 This revision is now codified in the sixth edition of the AJCC Cancer Staging Handbook.2 Before its publication, this staging system was validated using a large, international cohort of 17,600 patients3 from 13 cancer centers and cooperative groups wherein data were available on the prognostic factors used in the proposed revision. As part of the validation, clinical and pathologic attributes were evaluated to identify those that were independent prognostic factors. The attributes tested were those identified from previous studies to be important, reproducible, and readily ascertained. Tumor thickness and ulceration were found to be the most important prognostic factors in patients with stage I and II primary lesions. In those with stage III disease, the number of involved regional nodes, the quality of involvement (microscopic at sentinel node biopsy or macroscopic when clinically and histologically positive), and primary tumor ulceration were the most important adverse predictors. In stage IV disease, the site(s) of distant metastasis and serum lactate dehydrogenase level were the two prognostic factors selected.
The AJCC staging and the TNM classification systems have played important roles in clinical research and patient care by identifying factors that are associated with prognosis and providing estimates of survival probabilities for patients with varying prognosis. This information is used by clinical trialists to stratify patients and by clinicians to inform clinical decision making and address prognosis with individual patients. Because the accuracy of this information is of great importance, it is necessary to investigate the generalizability of these classification systems in different populations. The authors of the AJCC staging system noted that their cohort was likely not representative of the general population; they commented that "varying referral patterns and clinical trial entry criteria to these tertiary medical centers and cooperative groups might have skewed the patient distribution overall compared to the general population."3
We have conducted an independent validation study using data from the Surveillance, Epidemiology, and End Results (SEER) registry supported by the National Cancer Institute to examine the transportability of the AJCC staging system to a population-based cohort.4 The cohort included all patients (n = 41,417) with primary invasive cutaneous melanoma with complete data on prognostic factors required to stage patients according to the AJCC staging criteria who were diagnosed between 1988 and 2001 in 11 SEER regions within the United States. In this validation study, we investigate the ability of the TNM and AJCC staging classifications for melanoma to discriminate among patients based on their survival distributions and to predict survival rates for the SEER patients.
PATIENTS AND METHODS
Data for SEER patients were obtained from the SEER Cancer Incidence Public-Use Database, November 2003 Submission. Data for the AJCC patients were obtained from the electronic database used for the AJCC validation study.1,3 These cohorts are referred to as the SEER and AJCC cohorts, respectively. The registries, centers, and organizations that have contributed patients to the SEER and the AJCC cohorts are listed in Table 1. The potential overlap between these cohorts is limited to the University of Washington (Seattle), Eastern Cooperative Oncology Group, and Southwest Oncology Group, and this overlap represents at most 5% of the AJCC stage I to III patients.
Patients in the SEER cohort were diagnosed with invasive cutaneous melanoma between 1988 and 2001 in the following SEER regions: San Francisco-Oakland standard metropolitan statistical area, Connecticut, Metropolitan Detroit, Hawaii, Iowa, New Mexico, Seattle (Puget Sound), Utah, Metropolitan Atlanta, San Jose-Monterey, and Los Angeles. During this period, there were 57,230 individuals who had a diagnosis of primary melanoma (International Classification of Diseases for Oncology morphology codes between 872.0 and 879.9) of a lesion on the lip, eyelid, external ear, face, scalp, neck, trunk, limbs shoulder, hip, or skin not otherwise specified. Patients were excluded if they did not receive definitive treatment for their melanoma (n = 2,818; 4.9%). In addition, 11,239 patients were excluded because of incomplete data on one or more of these four variables: nonspecific or unknown SEER extent of disease classification (3.1%), unknown tumor thickness (11.9%), missing information on examination of lymph nodes (0.5%), and unknown vital status and/or survival time (2.2%). Finally, patients were excluded when thickness and Clark’s level were inconsistent (n = 1,756; 3.1%). After all exclusions, 41,417 patients (72.4%) were available for this study. Three variables (sex, age, and anatomic site) were complete for all 57,230 patients and these variables were used to compare those patients included in and excluded from the analysis. There were no significant differences between these two groups on these three characteristics (data not presented). Patients identified for the AJCC cohort included 30,450 patients seen between 1940 and 1999. Patients with missing data on variables required for staging were excluded and the remaining 17,600 patients (58%) were included in this study.3
The AJCC staging system is based on the TNM classification of melanomas. The T classification requires information on thickness, ulceration, and Clark’s level and all three variables were available in the SEER data. The N classification requires information on the number of regional nodes containing metastases as well as the presence or absence of in-transit metastasis and/or satellites. The SEER data include the number of regional lymph nodes removed for staging purposes as well as the number of positive lymph nodes. An indication of clinically evident in-transit metastasis or satellites was also available from the SEER description of the clinical extent of the lesion. However, information on the clinical status of the regional nodes was not available. We assumed that patients who had no nodes examined had no clinical evidence of nodal involvement at diagnosis. The M classification requires information about the sites of distant metastasis and serum lactate dehydrogenase levels and these data are not available in SEER. Consequently, for this study all patients with a distant metastasis are classified as AJCC stage IV without additional subclassification.
For all patients, definitive surgery is classified as biopsy of the primary tumor and excision (less than a wide excision and < 1 cm margin), wide excision, radical excision, major amputation, and surgery not otherwise specified. For all patients diagnosed after 1997, information on the scope of regional lymph node surgery was added to the SEER database to accommodate information from those who had a sentinel lymph node procedure, which was followed-up when positive by node dissection. The total number of positive regional lymph nodes was recorded.
Several other important prognostic factors are available in the SEER database. Patients were characterized by sex, age at diagnosis, and race, whereas anatomic site and histogenetic type further characterize their melanomas. Age and anatomic site were recoded as was done in the original AJCC validation study. Age groups were defined using deciles (0 = birth to 9 years; 1 = 10 to 19 years, and so on). Lesions were characterized as either axial or extremity.
Survival time was defined as the time between diagnosis and melanoma-specific death. For those alive at last follow-up, the survival time was censored on that date. For those who died as a result of other causes or who were lost to follow-up, survival times were censored at the date of death or date of last contact, respectively. In contrast to the SEER cohort, in which all patients were diagnosed between 1988 and 2001, 71% of the AJCC patients were diagnosed with melanoma between 1988 and 1999, and the remaining 29% were diagnosed between 1940 and 1987. Consequently, the lengths of follow-up differed between the two cohorts. Among SEER patients alive at the time of last follow-up (n = 33,728), the average follow-up time was 61.7 months, and 87.6% (n = 29,537) had at least 1 year of follow-up, 76.0% (n = 25,622) had at least 2 years of follow-up, 46.2% (n = 15,513) had at least 5 years of follow-up, and 12.7% (n = 4,266) had at least 10 years of follow-up. For those who died as a result of melanoma, the average time to death was 30.2 months. For AJCC patients alive at the time of last follow-up (n = 13,530), the average follow-up time was 61.9 months, and 78.5% (n = 10,615) had at least 1 year of follow-up, 65.1% (n = 8,814) had at least 2 years of follow-up, 40.2% (n = 5,434) had at least 5 years of follow-up, and 16.3% (n = 2,200) had at least 10 years of follow-up. For those who died as a result of melanoma, the average time to death was 38.1 months.
We used three quantitative methods to compare prognostic discrimination in the two cohorts. The first addressed prognostic separation at the landmark times of 1, 2, 5, and 10 years, computed as the difference between the best and worst survival rate at a specific time.5 The second used the variability of stage-specific survival rates relative to the overall survival rate at each of the landmark times, computed as the average squared deviations of the four survival rates from the overall survival rate at a particular time. The third used the AJCC stage-specific hazard ratios (HRs) from the Cox model that characterized the risk of melanoma-specific death for stage II, III, and IV compared with stage I. For all three methods, a larger value for the statistic means greater discrimination among AJCC stages (each using a different standard).
All survival distributions were estimated using the Kaplan-Meier procedure. Survival rates were obtained from these distributions, and the CIs were computed using Greenwood’s formula. Risk ratios for poor prognostic factors (male, axial lesions, thicker lesions, older patients, and higher Clark level) were used to compare the SEER cohort with the AJCC cohort. Unadjusted and adjusted hazard ratios were obtained from the univariate and multivariate Cox proportional hazards models. All survival analyses were done with Proc PHREG in SAS software (SAS Institute, Cary, NC).6
RESULTS
Discrimination Among SEER Patients by AJCC Staging
To assess the ability of the AJCC stages to discriminate among patients with different prognoses in the US population-based cohort, SEER patients were classified according to the AJCC staging criteria; the stage-specific survival curves and the overall survival curve were computed and compared with those of the AJCC patients (Fig 1). We observed the same relationship between stage and prognosis in both cohorts: stage I defined those with the best prognosis, stage IV defined those with worst prognosis; stage II defined those with a better prognosis than stage III.
Generally, at each landmark time the prognostic separation (the difference between the best and worst survival rates) in SEER was similar to the corresponding AJCC separation (Table 2). For example, 10-year survival rates ranged from 14.1% to 94.3% (separation, 80.2%) compared with 6.8% to 83.6% (separation, 76.8%) for SEER and AJCC, respectively. There was more heterogeneity among the SEER stage-specific survival rates, particularly for 10-year rates, where the variability was 1.7-fold higher compared with those for AJCC. The stage-specific HRs for stages II to IV compared with stage I were computed (Table 3). The HRs increased with increasing stage and were significantly higher for each subsequent stage. Of note, the stage-specific HRs were 2.3- to three-fold higher for the SEER cohort compared with the AJCC cohort. On the basis of each criterion, the AJCC stages identified groups of SEER patients with differential prognosis at least as well as, and in some cases better than, in the AJCC cohort. A similar pattern was observed with the survival curves for subgroups of patients classified by pathologic stage within each AJCC stage (Fig 2).
One major change in the revision of the AJCC staging system was the inclusion of the presence or absence of ulceration of the primary tumor as a prognostic factor. Figure 3 depicts SEER and AJCC survival curves for patients stratified by thickness ( 1.00, 1.01 to 2.00, 2.01 to 4.00, and > 4.0 mm from top to bottom, respectively) and ulceration (present and not present). The pattern of upstaging by ulceration in the survival curves originally observed in the AJCC cohort (Fig 3B) was also seen for the SEER patients (Fig 3A): survival for patients with ulcerated melanomas was diminished such that these patients had rates equivalent to thicker melanomas without ulceration. For SEER patients, the HR for a specific thickness group with ulceration did not differ from that of the next higher thickness group without ulceration (Table 4)
SEER and AJCC Survival Rates
Survival rates were computed using the SEER data for each AJCC stage to assess the degree to which the AJCC survival rates predict survival rates in the SEER cohort. To facilitate comparison of survival rates from the SEER cohort and AJCC cohort, the 1-, 2-, 5-, and 10-year survival rates and their SEs (as tabulated in Balch et al1) are presented in Table 5 by pathologic stage. The majority of the SEER survival rates exceed the corresponding AJCC survival rate and differences were not specific to any one landmark time (see Fig 4).
Prognostic Factors and Patient Mix in the SEER and AJCC Cohorts
The characteristics of the SEER and AJCC patients and their lesions were examined to identify any differences in patient mix that potentially could explain the observed differences in the predicted survival rates. Tables 6, 7, and 8 present the characteristics of SEER patients by AJCC stages. The differences between the two cohorts were summarized using a risk ratio for each prognostic factor, for which the risk ratio was defined as the proportion of patients in the SEER cohort with the category associated with poor prognosis divided by the corresponding proportion in the AJCC cohort (Fig 5). For three of the five prognostic factors (age, Clark level, and axial lesions), SEER patients were generally more likely to be in the poor-risk category. For example, the risk ratios for age were all 1.2 (and as high as 1.8), characterizing the greater proportion of patients 60 years of age or older in SEER compared with AJCC patients. Generally, the risk ratios for sex were approximately equal to 1 (ie, the proportions of male patients were approximately equal for the two cohorts). For thickness, however, AJCC patients generally were more likely to be in the poor-risk category. In particular, the risk ratios were 0.5 and 0.8 for T1a and T1b (thin lesions), respectively, and the proportion of patients with thicker lesions (> 0.75 mm) within the thin range were two-fold (T1a) and 1.3-fold (T1b) higher for the AJCC cohort compared with the SEER cohort, respectively.
Thickness and Prognosis
Thickness is considered to be the strongest univariate predictor of disease-specific survival in patients with melanoma. The distributions of lesion thickness for SEER and AJCC are presented in Figure 6 and these distributions differ particularly with respect to the thinner lesions. As is apparent in Figure 7, for melanomas 5 mm, 10-year survival rates decrease linearly with thickness with similar slopes. Nevertheless, for patients with lesions of the same thickness, SEER survival rates are higher than the AJCC rates, consistent with the pattern in the stage-specific rates. Cox regression models including cohort and thickness were used to investigate whether thicker lesions in the AJCC cohort could explain the higher mortality risk observed within each AJCC substage (data not shown). With the exception of stage IIA, there was a significant difference in mortality risk between the SEER and AJCC cohorts after controlling for lesion thickness. Stage IA lesions had the largest adjusted cohort effect; AJCC stage IA patients had a four-fold higher mortality risk compared with the SEER stage IA patients.
DISCUSSION
The identification of prognostic factors and their use in estimating patients’ outcomes are important tools in clinical research and practice.7 Most published prognostic studies present exploratory studies wherein prognostic factors are assessed using patient populations that are available for study. Their implementation in practice awaits additional validation and/or replication. The two major criteria for validation of a prognostic model are accuracy and generalizability.8 A prognostic model, such as the AJCC staging system, is accurate when the model is able to separate patients’ outcomes (discrimination) and when the predicted outcomes are similar to observed outcomes (calibration).9 An accurate model is generalizable if the model’s accuracy can be replicated in a similar independent sample (reproducible) or is shown to be accurate in a related population sample (transportable).
In 2002, a revised version of the AJCC staging system for cutaneous melanoma was published that included stage-specific survival rates.1 It included a multicenter validation study that demonstrated the significance of the prognostic factors used in the AJCC staging system.3 This study demonstrated a statistically significant difference among the stage-specific and substage-specific survival curves. The 10-year survival rates for TNM classes ranged between 87.9% for T1a patients and 15.0% for N2b patients.3 Thickness and ulceration were shown to be statistically significant independent prognostic factors. The AJCC study established that thickness, ulceration, age, site, and level were statistically significant and independent prognostic factors for those melanoma patients without evidence of regional or disseminated metastatic disease. Nodal status, thickness, ulceration, site, and age were statistically significant and independent prognostic factors for melanoma patients without clinical evidence of nodal metastasis whose regional lymph nodes were pathologically staged after sentinel or elective lymphadenectomy.
In this article we have presented an evaluation of the accuracy and generalizability of the AJCC staging system using population-based data on melanoma patients collected in 11 US sites as part of the National Cancer Institute’s SEER program. The significant differences seen among the AJCC stage-specific survival curves were also seen in the SEER stage-specific survival curves. Hence, AJCC stages are discriminative. However, survival rates were generally better for SEER patients than for AJCC patients in all stages, notably among those with thinner lesions, in whom we observed a 10-year survival rate that was 10% higher for those with T1a lesions (approximately 60% of all SEER melanoma patients). The observed differences in the survival rates indicate that the AJCC staging system is imperfectly calibrated with respect to the general US population as reflected in SEER.
There are a number of potential reasons for the difference in survival rates in the AJCC and SEER populations. As we have noted, the SEER cohort differs from the AJCC cohort in several ways. The SEER cohort is population based, given that all SEER sites obtain data on all patients diagnosed within their region. The SEER cohort is also more contemporary; patients included in the present study were diagnosed after 1987, when thickness became available in the SEER data set, whereas the AJCC cohort had patients diagnosed as early as 1940. However, when we compared the stage-specific survival rates of AJCC patients diagnosed between 1985 and 1999 with those of the SEER patients, the same differential survival rates were seen (data not shown). Differences in the length of follow-up are not likely to explain differential survival. Although the SEER cohort has a lower proportion of people with 10 years of follow-up, the number of patients with at least 10 years of follow-up is greater (4,509 and 2,200 for SEER and AJCC, respectively). We observed that SEER patients had a mix of unfavorable and favorable prognostic factors that distinguished them from the AJCC patients. On the one hand, SEER patients were older and their lesions were more likely to be axial and Clark level IV or V—all of which are associated with poorer survival. On the other hand, they were more likely to have thinner lesions, which are prognostic of better survival. Because lesional thinness is a strong, favorable prognostic factor, its higher prevalence in SEER melanomas may contribute to better outcome among low-stage SEER patients. However, whether we addressed outcome by analyzing subsets of thickness within the AJCC thickness ranges or by using an alternative classification strategy based on a prognostic tree (data not shown), we observed better prognosis among SEER patients compared with AJCC patients in the same risk group.
That we were unable to explain the differences in survival rates between the two cohorts by differences in lesion thickness, temporal trends, or any of the other AJCC prognostic factors indicates that there are likely differences between the two cohorts on unmeasured factors. In addition, there remains heterogeneity among patients within the AJCC stages that can be resolved with a more informative set of biomarkers. We and others have evaluated a number of biomarkers that are not currently used in AJCC staging, and are not available in SEER, which may be both more biologically plausible and predictive.10-19 These include variables that should be considered in the next revision of the AJCC staging system, and that are readily ascertained in routine histology (eg, growth phase, dermal mitotic count, and the presence, location, and density of tumor-infiltrating lymphocytes) or by current immunohistologic methods.20
We conclude that the AJCC staging system discriminates well among melanoma patients with different prognoses and includes prognostic factors that are significant predictors of disease-specific survival. It provides a classification scheme that may be adequate to stratify for clinical trials patients referred to cancer centers and comparable institutions. However, the survival rates predicted by AJCC staging are specific for its population, underestimate survival in the US population as reflected in SEER, and may not reflect accurately the survival rates in other populations.12,21 The SEER survival rates for the AJCC stages that we have presented in this article (Table 5) provide data that potentially can be used in trial design, clinical management, and patient counseling for a population thought to reflect the general US population of melanoma patients.
Authors’ Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
Acknowledgment
We thank April Fritz, Manager, Data Quality, SEER Program of the Division of Cancer Control and Population Sciences of the National Cancer Institute, for help with the interpretation of the SEER data; and all of the institutions, organizations, and cooperative groups that contributed patients to the AJCC database used in developing the AJCC staging system for melanoma.
NOTES
Supported in part by Grant No. CA-093372 (M. Herlyn, principal investigator), SPORE on Skin Cancer.
Presented in part at the 40th Annual Meeting of the American Society of Clinical Oncology, New Orleans, LA, June 5-8, 2004.
Authors’ disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Balch CM, Buzaid AC, Soong SJ, et al: Final version of the American Joint Committee on caner staging system for cutaneous melanoma. J Clin Oncol 19:3635-3648, 2001
Greene F, Page D, Fleming I, et al (eds): AJCC Cancer Staging Handbook (ed 6). New York, NY, Springer, 2002
Balch CM, Soong SJ, Gershenwald JE, et al: Prognostic factors analysis of 17,600 melanoma patients: Validation of the American Joint Committee on Cancer melanoma staging system. J Clin Oncol 19:3622-3634, 2001
Surveillance, Epidemiology, and End Results (SEER) Program Public-Use Data (1973-2001), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch, released April 2004, based on the November 2003 submission. http://www.seer.cancer.gov/publicdata/
Altman DG, Royston P: What do we mean by validating a prognostic model? Stat Med 19:453-473, 2000
SAS Institute Inc: SAS/STAT User’s Guide, Version 8. Cary, NC, SAS Institute Inc, 1999
Mackillop WJ: The role of prognosis in cancer medicine, in Gospodarowicz MK, Henson DE, Hutter RVP, et al (eds): Prognostic Factors in Cancer, (ed 2). New York, NY, Wiley-Liss, 2001
Justice AC, Covinsky KE, Berlin JA: Assessing the generalizability of prognostic information. Ann Intern Med 130:515-524, 1999
Harrell FE: Regression Modeling Strategies. New York, NY, Springer-Verlag, 2001
Clark WH, Elder DE, Guerry D, et al: Model predicting survival in stage I melanoma based on tumor progression. J Natl Cancer Inst 81:1893-1904, 1989
Gimotty PA, Guerry D, Ming M, et al: Thin primary cutaneous malignant melanoma: A prognostic tree for 10-year metastasis is more accurate than American Joint Committee on Cancer Staging. J Clin Oncol 22:3668-3676, 2004
McKinnon JG, Yu XQ, McCarthy WH, et al: Prognosis for patients with thin cutaneous melanoma. Cancer 98:1223-1231, 2003
Mansson-Brahme E, Carstensen J, Erhardt K, et al: Prognostic factors in thin cutaneous malignant melanoma. Cancer 73:2324-2332, 1994
Finley JW, Gibbs JF, Rodriguez LM, et al: Pathologic and clinical features influencing outcome of thin cutaneous melanoma: Correlation with newly proposed staging system. Am Surg 66:527-531, 2000
Oliveira-Filho RS, Ferreira LM, Biasi LJ, et al: Vertical growth phase and positive sentinel node in thin melanoma. Braz J Med Biol Res 36:347-350, 2003
Gromet MA, Epstein WL, Blois MS: The regressing thin malignant melanoma: A distinctive lesion with metastatic potential. Cancer 42:2282-2292, 1978
Cook MG, Spatz A, Brocker E, et al: Identification of histological features associated with metastatic potential of thin (< 1.0 mm) cutaneous melanoma with metastases: A study on behalf of the EORTC Melanoma Group. J Pathol 197:188-193, 2002
Lefevre M, Vergier B, Balme B, et al: Relevance of vertical growth pattern in thin level II cutaneous superficial spreading melanomas. Am J Surg Pathol 27:717-724, 2003
Massi D, Franchi A, Borgognoni L, et al: Thin cutaneous malignant melanomas (< or =1.5 mm): identification of risk factors indicative of progression. Cancer 85:1067-1076, 1999
Alonso SR, Ortiz P, Pollan M, et al: Progression in cutaneous malignant melanoma is associated wsith distinct expression profiles: A tissue microarray-based study. Am J Pathol 164:193-203, 2004
Eigentler TK, Buettner PG, Leiter U, et al: Impact of ulceration in stages I to III cutaneous melanoma as staged by the American Joint Committee on Cancer Staging System: An analysis of the German Central Malignant Melanoma Registry. J Clin Oncol 22:4376-4383, 2004(Phyllis A. Gimotty, Jeffr)