Evidence for an Age Cutoff Greater Than 365 Days for Neuroblastoma Risk Group Stratification in the Children's Oncology Group
http://www.100md.com
《临床肿瘤学》
the University of Florida and Children's Oncology Group Department of Statistics, Gainesville, FL
University of Alabama at Birmingham, The Children's Hospital, Birmingham, AL
University of California at San Francisco School of Medicine, San Francisco
Children's Hospital of Los Angeles and The University of Southern California Keck School of Medicine, Los Angeles, CA
Harvard University, The Dana-Farber Cancer Institute, Boston, MA
Hospital for Sick Children, Toronto, Ontario, Canada
Children's Hospital of Philadelphia and University of Pennsylvania, Philadelphia, PA
Northwestern University's Feinberg School of Medicine, Chicago, IL
ABSTRACT
PURPOSE: In the Children's Oncology Group, risk group assignment for neuroblastoma is critical for therapeutic decisions, and patients are stratified by International Neuroblastoma Staging System stage, MYCN status, ploidy, Shimada histopathology, and diagnosis age. Age less than 365 days has been associated with favorable outcome, but recent studies suggest that older age cutoff may improve prognostic precision.
METHODS: To identify the optimal age cutoff, we retrospectively analyzed data from the Pediatric Oncology Group biology study 9047 and Children's Cancer Group studies 321p1-p4, 3881, 3891, and B973 on 3,666 patients (1986 to 2001) with documented ages and follow-up data. Twenty-seven separate analyses, one for each different age cutoff (adjusting for MYCN and stage), tested age influence on outcome. The cutoff that maximized outcome difference between younger and older patients was selected.
RESULTS: Thirty-seven percent of patients were younger than 365 days, and 64% were 365 days old (4-year event-free survival [EFS] rate ± SE: 83% ± 1% [n = 1,339] and 45% ± 1% [n = 2,327], respectively; P < .0001). Graphical analyses revealed the continuous nature of the prognostic contribution of age to outcome. The optimal 460-day cutoff we selected maximized the outcome difference between younger and older patients. Forty-three percent were younger than 460 days, and 57% were 460 days old (4-year EFS rate ± SE: 82% ± 1% [n = 1,589] and 42% ± 1% [n = 2,077], respectively; P < .0001). Using a 460-day cutoff (assuming stage 4, MYCN-amplified patients remain high-risk), 5% of patients (365 to 460 days: 4-year EFS 92% ± 3%; n = 135) fell into a lower risk group.
CONCLUSION: The prognostic contribution of age to outcome is continuous in nature. Within clinically relevant risk stratification, statistical support exists for an age cutoff of 460 days.
INTRODUCTION
Neuroblastoma patients in the Children's Oncology Group (COG) are currently stratified into low-, intermediate-, and high-risk groups based on age at diagnosis (< 365 days v 365 days),1 International Neuroblastoma Staging System (INSS) stage (1, 2, 3, 4, 4s),2,3 MYCN status (genomic amplification v nonamplified),4,5 Shimada histopathology (favorable v unfavorable),6,7 and tumor cell ploidy (DNA index = 1.0 v > 1.0).8 Treatment is tailored according to risk. Survival rates for patients with low- and intermediate-risk disease are excellent, and ongoing clinical trials are focused on reducing therapy and decreasing long-term treatment-related sequelae.9-12 Most patients with low-risk disease are treated with surgery alone, whereas children with intermediate-risk neuroblastoma receive four to eight cycles of chemotherapy and surgery. In contrast, long-term outcome of children with high-risk disease remains poor13 if treated with nonmyeloablative chemotherapy, but is improved for those treated with intensive induction chemotherapy, surgery, radiation, myeloablative therapy, and 13-cis-retinoic acid for minimal residual disease.14
Because treatment strategies and inherent toxicities differ significantly between high-risk and low- and intermediate-risk patients, it is important to optimize treatment stratification criteria. The currently accepted age cutoff of 365 days was based on observations by Breslow and McCann more than 30 years ago that older age was associated with worse outcome.1 Using logistic regression to examine the relationship of age and Evans stage15 to outcome in 246 children, they found that survival probabilities decreased with increasing age during the first year or two, but age effect tapered off thereafter. Although these results suggest that days of age should be used for risk stratification as a continuous variable, this is not clinically practical. Instead, age has been used as a prognostic variable with the convenient cutoff of 365 days, and several studies have demonstrated the significant clinical impact of this age cutoff.7,16-18 However, excellent outcome also has been observed recently in a subset of toddlers (12 to 18 months of age) with stage 4 disease who were treated with multiagent chemotherapy, suggesting that 365 days may not be the best age cutoff for risk stratification.19,20 To identify the statistically optimal age cutoff, we analyzed the influence of age on outcome in 3,666 patients after adjusting for known prognostic factors INSS stage and MYCN status.
METHODS
Patients
We enrolled 3,666 patients in Pediatric Oncology Group (POG) and Children's Cancer Group (CCG) studies from 1986 to 2001 (both groups are now part of the Children's Oncology Group [COG]). All POG patients were enrolled in biology study 9047,21 and the primary POG therapeutic trials were 8742,22 9243, 9340,23 9341, and 9342. CCG patients were enrolled on therapeutic trials 321p1, 321p2, 321p3, 321p4,24 3881, 3891,14 or biology study B973. Patients had to meet eligibility requirements of their biology or therapeutic study, at a minimum: (1) diagnosis of neuroblastoma confirmed by central pathology review; (2) age 21 years; and (3) written informed consent. Data for age and stage were collected on all studies, and tumor tissue for biologic studies was required or strongly encouraged. MYCN status was determined by Southern blotting (before July 1993) or fluorescence in situ hybridization (in POG/COG after July 1993)5,25-27 or tissue expression with immunoperoxidase staining for MYCN protein (in CCG after 1993). For CCG patients and some POG patients, central pathology review to determine Shimada histopathology was performed.6,7 For POG patients, ploidy (DNA index [DI] of tumor cells determined by flow cytometry) was used.8 The patient's age was determined on the date of diagnosis, where the date of diagnosis was the date of the pathology review diagnosis of neuroblastoma from the first surgery, biopsy, or bone marrow biopsy.
Statistical Considerations
For clinically relevant analyses, the cohort was divided into 20 age groups, each with a sufficient number of patients for descriptive purposes. Because of preponderance of younger patients, groups spanned narrow age ranges at younger ages and wider ranges at older ages. For inferential analyses to identify an optimal cutoff, the cohort was repeatedly divided into two groups. In addition to the 19 cutoffs that created the 20 groups, a concentration of eight more cutoffs were tested between 12 and 24 months to provide a sufficient level of detail within the age range of clinical interest. Twenty-seven paired groups were studied, each separated by a different cutoff. The outcome difference between each paired group was quantified by the P value and the hazard ratio (henceforth referred to as relative risk [RR]) from a multivariable Cox proportional hazards model.28
To ensure that age was prognostic and not a spurious statistically significant result due to large sample size, a two-fold cross-validation methodology was employed.29,30 The cohort was divided at random into set 1 and set 2. Within each set, univariate (log-rank) and multivariable (Cox proportional hazards) testing for significance of age was performed.28 For multivariate analysis, stepwise backward model building identified statistically significant (P < .05) factors and reflected inter-relationships.31 Once age was shown to be prognostic within each subset, set 1 and set 2 were combined, and the age cutoff was determined in the overall cohort.29,30 Two approaches using a Cox model were used to select the cutoff: (a) a minimum corrected P value approach31; and (b) a maximum hazard rate (RR) approach. Survival tree regression was performed32 to demonstrate the influence age cutoff has on the entire structure of risk stratification. To identify groups of patients with differing prognoses, all possible trees for all possible age cutoffs were formed using INSS stage, MYCN status, ploidy, Shimada histopathology, and age.
We checked for sufficient power to support the number of covariates in the Cox model for the 719 events in this cohort.33,34 For event-free survival (EFS), events considered were relapse, disease progression, secondary malignancy, and death, whichever occurred first. Survival rates are quoted as the rate ± SE and are calculated using the methods of Kaplan and Meier35 and Peto.36
RESULTS
Clinical and Biologic Characteristics of Study Cohort
Median age was 573 days; the number of patients decreased exponentially with increasing age (Fig 1). Sixty-four percent of patients (n = 2,327) were 365 days old, 44% had INSS stage 4 disease (n = 1,522), 18% had MYCN-amplified tumors (n = 520), 29% had diploid tumors (n = 483), and 43% had tumors with unfavorable histopathology (n = 665). Median follow-up time was 5.8 years. Overall 8-year EFS and overall survival rates were 55% ± 2% and 61% ± 2%, respectively (n = 3,666). The EFS rate of patients 365 days old was lower than that of patients less than 365 days old (4-year EFS rates: 83% ± 1% [n = 1,339] and 45% ± 1% [n = 2,327], respectively; P < .0001; Table 1). The long-term EFS rate decreased with increasing age (Fig 2).
Relationship of Age With MYCN Status, Stage, Ploidy, and Histopathology
When age was plotted against the proportion of patients with specific clinical or biologic features, the curves showed the relative distribution of patients with and without particular features (Fig 3). In general, before the age at which a pair of curves crossed, the cohort was dominated by patients whose prognoses were favorable based on the given feature. After the age at which the curves crossed, most patients had the less favorable feature. A crossing or confluence of curves occurred at a point beyond 365 days for MYCN status (12.3 months), stage (20 months), ploidy (14 months), and histopathology (17 months; Fig 3).
Outcome by Age Groups
In a univariate analysis (n = 2,825 with complete data for INSS stage, MYCN status, and follow-up), the risk of an event for older age groups was roughly 3 to 4 times that of the reference group, patients age 0 to 21 days (Fig 4A). Sixty-five percent of patients with MYCN-amplified tumors were diagnosed at 11.5 to 40 months old, and most stage 4 patients were diagnosed at older than 30 months (Figs 3A-B). Many events of older patients may be associated with advanced stage or MYCN amplification and not necessarily older age. To adjust for this, we included age group (P < .0001), stage (P < .0001), and MYCN status (P < .0001) in a multivariable Cox model, resulting in an arguably more clinically useful model than with age alone. This adjustment accounted for events due to MYCN status and stage, and what remained was a 50% to 70% increased risk of an event for older patients (Fig 4B). Before 600 days, there was decreased risk of an event, and after 600 days there was increased risk.
Determination of Optimal Age Cutoff
To identify an age cutoff that may be preferable to the 365-day cutoff, 27 potential cutoffs were evaluated, each in a separate analysis. Each analysis divided the entire cohort into two groups—one that included patients older than the age cutoff and one that included those younger. We used two measures to quantify the difference in EFS between the younger and the older group—the P value and RR for the effect of age.
Each of the 27 separate analyses of age cutoff adjusted for stage and MYCN status (n = 2,825). A 573-day cutoff had the lowest P value (corrected P = 9 x 10–32; Fig 5A). Due to the fact that under certain circumstances, the greatest statistical power is reached when a cohort is divided into two groups of equal sample size, it is perhaps not a coincidence that 573 days was also the median age. The P values for cutoffs from 15 to 22 months were all extremely small (P < 1 x 10–30), and any cutoff in this range could be considered for use in risk stratification. An age cutoff of 460 days had the largest RR for an event (RR = 2.598); however, any cutoff from 14 to 19 months has large enough RR to be considered for use in risk stratification (Fig 5B). As presented in Table 1, the 4-year EFS ± SE of patients aged 365 to 460 days is 73% ± 3% (n = 250). If INSS stage 4 patients with MYCN-amplified tumors were excluded, the 4-year EFS of patients 365 to 460 days old was excellent: 92% ± 3% (n = 135).
In the survival tree regression, any cutoff from 570 to 670 days resulted in a stratification that optimized the prognostic contribution of age and other risk group factors. Other cutoffs resulted in different stratification trees, though none resulted in a stratification tree in which age was as highly statistically significant.
DISCUSSION
In neuroblastoma, age can be thought of as a surrogate risk factor for genetic or biologic markers of risk that are yet undiscovered or unproven. Age will remain important in determination of risk group and treatment until molecular markers are identified that provide similar prognostic information. This analysis of a large cohort will improve the use of age as a prognostic factor.
We have shown biologic and statistical support of a choice of age cutoff more than 365 days. (1) Curves of the proportion of patients by risk factor crossed or came together between 12.2 to 20 months. (2) Adjusted for stage and MYCN status, there was decreased risk of an event at less than 600 days (19.7 months), and increased risk of an event at 600 days. (3) The minimum P value for age cutoff occurred at 573 days (18.8 months). (4) The maximum RR for age cutoff occurred at 460 days (15.1 months). (5) The stratification tree that made optimal use of the prognostic contribution of age had a cutoff between 570 and 670 days (18.8 to 22.0 months). All of these choices for age cutoff fall within a range of 15 to 20 months, and any choice in this range would be statistically valid and well supported by this large data cohort.
Besides evidence of a relationship between age and outcome, there is biologic evidence of a relationship between age and MYCN status, stage, ploidy, and histopathology. We propose that these relationships provide empirical biologic evidence that the age cutoff of 365 days is too low, and that the time point beyond which unfavorable risk factors begin to dominate is higher than 365 days in each case.
Our findings indicate that the age cutoff should be increased for COG and for international risk group stratification systems for neuroblastoma. If the cutoff were increased to more than 365 days, there is a subset of patients who would be classified in a lower risk group than if an age cutoff of 365 days had been used. Because these patients would be considered to have lower risk for disease recurrence under the newer paradigm, they would receive less therapy than they would with use of a cutoff of 365 days. If a 460-day cutoff were selected, approximately 5% of patients would shift to a lower risk group, or 11% for a 573-day cutoff. This calculation excludes patients older 365 days old at diagnosis who are INSS stage 4 with MYCN-amplified tumors, who are still considered to be in a higher risk group.
Although it is possible that prognostic factor analyses could provide misleading findings,37 a strength of this analysis was the large sample size. Also, patients in this cohort met eligibility requirements of their biology or therapeutic studies; however, some bias could have been introduced due to missing data. For most patients, tumor specimen submission was an eligibility requirement and was met in most cases. Specimens were analyzed for MYCN status, ploidy (POG), and histopathology (CCG). Adjustment for INSS stage and MYCN status was performed in the analyses, but there were 841 patients (23%) for whom MYCN status and/or stage data were unavailable. These patients had to be excluded from analyses used to determine the optimal age cutoff. It is possible that data for these patients were not missing at random. Tumor specimens of sufficient size and quality tend to be less readily available in patients with INSS stage 3 to 4 tumors, and stage 3 to 4 patients may have been under-represented because of missing MYCN status.
A potential confounding issue was the lack of homogeneity of patients' treatments. Treatment is known for most CCG patients because of their enrollment in therapeutic studies, but many POG patients were enrolled only in biology study 9047. However, it is reasonable to assume that patients were treated in accordance with the standard of care for the known risk factors at that time. Due to the wide variety of treatments, and because treatment was unknown for some patients, it was not possible to adjust for treatment in the models.
A paradigm whereby older patients receive more therapy than younger patients has been used in neuroblastoma treatment for several decades, long before the first patients in this cohort were diagnosed. One could postulate that the good outcome of patients slightly older than 365 days was simply a result of more therapy for patients older than 365 days. The report of Schmidt et al19 on study CCG-3891 provided evidence that this difference was an age effect and not a treatment effect. The EFS rate for patients 12 to 18 months old (6-year EFS: 87% ± 8.8%) was significantly better than the EFS rate for patients 18 to 24 months (6-year EFS: 36% ± 14.5%) or for patients older than than 24 months (6-year EFS: 23% ± 3.3%) who received the same therapy.19
With a change in age cutoff, the risk/benefit of an associated change in therapy must also be considered.38 Before a decision is made regarding a new age cutoff, consideration must be given to the number of patients who will be permitted to relapse under the lesser therapy in order to allow a greater number of nonrelapsing patients to avoid the toxicity and long-term side effects of the more intensive therapy.
In conclusion, the prognostic contribution of age to outcome in neuroblastoma is continuous in its nature. The worsening of outcome with increasing age is gradual, with no clear delineation of an age cutoff. There exists strong statistical, and perhaps biologic evidence in this cohort for an age cutoff greater than 365 days. The closer the new cutoff is to 365 days, the more conservative the choice, because fewer patients would be shifted to a lower risk group. There is statistical support for any choice of age cutoff between 15 and 19 months for use in risk stratification of neuroblastoma patients. A new cutoff of 460 days (15.1 months) at this time would be conservative, yet allow decreased therapy in an appropriate cohort of patients (approximately 5%).
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
Acknowledgment
We acknowledge the Pediatric Oncology Group and Children's Cancer Group for supplying tumor samples. We thank Dina Ingle for data management support, Christina Rigby for logistical support, and Sue Rowe for technical laboratory assistance.
NOTES
Supported in part by Pediatric Oncology Group Statistics and Data Center grant U10 CA29139, Children's Cancer Group Statistics and Data Center grant U10 CA13539, and Children's Oncology Group Statistics and Data Center grant U10 CA98413-01.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Breslow N, McCann B: Statistical estimation of prognosis for children with neuroblastoma. Cancer Res 31:2098-2103, 1971
Smith EI, Haase GM, Seeger RC, et al: A surgical perspective on the current staging in neuroblastoma: The International Neuroblastoma Staging System proposal. J Pediatr Surg 24:386-390, 1989
Brodeur GM, Pritchard J, Berthold F, et al: Revisions of the international criteria for neuroblastoma diagnosis, staging, and response to treatment. J Clin Oncol 11:1466-1477, 1993
Brodeur GM, Seeger RC, Schwab M, et al: Amplification of N-myc in untreated human neuroblastomas correlates with advanced disease stage. Science 224:1121-1124, 1984
Seeger RC, Brodeur GM, Sather H, et al: Association of multiple copies of the N-myc oncogene with rapid progression of neuroblastomas. N Engl J Med 313:1111-1116, 1985
Shimada H, Chatten J, Newton WA Jr, et al: Histopathologic prognostic factors in neuroblastic tumors: Definition of subtypes of ganglioneuroblastoma and an age-linked classification of neuroblastomas. J Natl Cancer Inst 73:405-416, 1984
Joshi VV, Cantor AB, Altshuler G, et al: Age-linked prognostic categorization based on a new histologic grading system of neuroblastomas: A clinicopathologic study of 211 cases from the Pediatric Oncology Group. Cancer 69:2197-2211, 1992
Look AT, Hayes FA, Nitschke R, et al: Cellular DNA content as a predictor of response to chemotherapy in infants with unresectable neuroblastoma. N Engl J Med 311:231-235, 1984
Perez CA, Matthay KK, Atkinson JB, et al: Biologic variables in the outcome of stages I and II neuroblastoma treated with surgery as primary therapy: A Children's Cancer Group study. J Clin Oncol 18:18-26, 2000
Schmidt ML, Lukens JN, Seeger RC, et al: Biologic factors determine prognosis in infants with stage IV neuroblastoma: A prospective Children's Cancer Group study. J Clin Oncol 18:1260-1268, 2000
Nickerson HJ, Matthay KK, Seeger RC, et al: Favorable biology and outcome of stage IV-S neuroblastoma with supportive care or minimal therapy: A Children's Cancer Group study. J Clin Oncol 18:477-486, 2000
Matthay KK, Perez C, Seeger RC, et al: Successful treatment of stage III neuroblastoma based on prospective biologic staging: A Children's Cancer Group study. J Clin Oncol 16:1256-1264, 1998
Brodeur GM, Maris JM: Neuroblastoma, in Pizzo PA, Poplack DG (eds): Principles and Practice of Pediatric Oncology (ed 4). Philadelphia, PA, Lippincott-Raven, 2001, pp 895-937
Matthay KM, Villablanca JG, Seeger RC, et al: Treatment of high-risk neuroblastoma with intensive chemotherapy, radiotherapy, autologous bone marrow transplantation, and 13-cis-retinoic acid. N Engl J Med 341:1165-1173, 1999
Evans AE: Staging and treatment of neuroblastoma. Cancer 45:1799-1802, 1980
Evans AE, D'Angio GJ, Sather HN, et al: A comparison of four staging systems for localized and regional neuroblastoma: A report from the Children's Cancer Study Group. J Clin Oncol 8:678-688, 1990
Berthold F, Kassenbohmer R, Zieschang J: Multivariate evaluation of prognostic factors in localized neuroblastoma. Am J Pediatr Hematol Oncol 16:107-115, 1994
Saito T, Tsunematsu Y, Saeki M, et al: Trends of survival in neuroblastoma and independent risk factors for survival at a single institution. Med Pediatr Oncol 29:197-205, 1997
Schmidt ML, Lal A, Seeger R, et al: Favorable prognosis for patients ages 12-18 months with stage 4 MYCN-nonamplified neuroblastoma. Proc Am Soc Clin Oncol 22:800, 2003 (abstr 3214)
George R, London WB, Maris JM, et al: Age as a continuous variable in predicting outcome for neuroblastoma patients with metastatic disease: Impact of tumor biologic features. Proc Am Soc Clin Oncol 22:799, 2003 (abstr 3213)
Alvarado CS, London WB, Look AT, et al: Natural history and biology of stage A neuroblastoma: A Pediatric Oncology Group study. J Pediatr Hematol Oncol 22:197-205, 2000
Strother D, van Hoff J, Rao PV, et al: Event-free survival of children with biologically favourable neuroblastoma based on the degree of initial tumour resection: Results from the Pediatric Oncology Group. Eur J Cancer 33:2121-2125, 1997
Kretschmar CS, Kletzel M, Murray K, et al: Response to paclitaxel, topotecan, and topotecan-cyclophosphamide in children with untreated disseminated neuroblastoma treated in an upfront phase II investigational window: A Pediatric Oncology Group study. J Clin Oncol 22:4119-4126, 2004
Matthay KK, O'Leary MC, Ramsay NK, et al: Role of myeloablative therapy in improved outcome for high risk neuroblastoma: Review of recent Children's Cancer Group results. Eur J Cancer 31A:572-575, 1995
Schwab M, Alitalo K, Klempnauer KH, et al: Amplified DNA with limited homology to myc cellular oncogene is shared by human neuroblastoma cell lines and a neuroblastoma tumour. Nature 305:245-248, 1983
Shapiro DN, Valentine MB, Rowe ST, et al: Detection of N-myc gene amplification by fluorescence in situ hybridization: Diagnostic utility for neuroblastoma. Am J Pathol 142:1339-1346, 1993
Mathew P, Valentine MB, Bowman LC, et al: Detection of MYCN gene amplification in neuroblastoma by fluorescence in situ hybridization: A Pediatric Oncology Group study. Neoplasia 3:105-109, 2001
Cox DR: Regression models and life tables. J R Stat Soc B 34:187-220, 1972
Mazumdar M, Smith A, Bacik J: Methods for categorizing a prognostic variable in a multivariable setting. Stat Med 22:559-571, 2003
Faraggi D, Simon R: A simulation study of cross-validation for selecting an optimal cutpoint in univariate survival analysis. Stat Med 15:2203-2213, 1996
Altman DG, Lausen B, Sauerbrei W, et al: Dangers of using "optimal" cutpoints in the evaluation of prognostic factors. J Natl Cancer Inst 86:829-835, 1994
Segal MR: Regression trees for censored data. Biometrics 44:35-47, 1988
Harrell FE Jr, Lee KL, Califf RM, et al: Regression modelling strategies for improved prognostic prediction. Stat Med 3:143-152, 1984
Feinstein AR. Multivariable Analysis: An Introduction. New Haven, CT, Yale University Press, 1996, pp 184-187, pp 578-582
Kaplan EL, Meier P: Nonparametric estimation from incomplete observations. J Am Stat Assoc 53:457-481, 1958
Peto R, Peto J: Asymptotically efficient rank invariant test procedures. J R Stat Soc A 135:185-198, 1972
Simon R, Altman DG: Statistical aspects of prognostic factor studies in oncology. Br J Cancer 69:979-985, 1994
Grosfeld JL: Risk-based management: Current concepts of treating malignant solid tumors of childhood. J Am Coll Surg 189:407-425, 1999(W.B. London, R.P. Castleb)
University of Alabama at Birmingham, The Children's Hospital, Birmingham, AL
University of California at San Francisco School of Medicine, San Francisco
Children's Hospital of Los Angeles and The University of Southern California Keck School of Medicine, Los Angeles, CA
Harvard University, The Dana-Farber Cancer Institute, Boston, MA
Hospital for Sick Children, Toronto, Ontario, Canada
Children's Hospital of Philadelphia and University of Pennsylvania, Philadelphia, PA
Northwestern University's Feinberg School of Medicine, Chicago, IL
ABSTRACT
PURPOSE: In the Children's Oncology Group, risk group assignment for neuroblastoma is critical for therapeutic decisions, and patients are stratified by International Neuroblastoma Staging System stage, MYCN status, ploidy, Shimada histopathology, and diagnosis age. Age less than 365 days has been associated with favorable outcome, but recent studies suggest that older age cutoff may improve prognostic precision.
METHODS: To identify the optimal age cutoff, we retrospectively analyzed data from the Pediatric Oncology Group biology study 9047 and Children's Cancer Group studies 321p1-p4, 3881, 3891, and B973 on 3,666 patients (1986 to 2001) with documented ages and follow-up data. Twenty-seven separate analyses, one for each different age cutoff (adjusting for MYCN and stage), tested age influence on outcome. The cutoff that maximized outcome difference between younger and older patients was selected.
RESULTS: Thirty-seven percent of patients were younger than 365 days, and 64% were 365 days old (4-year event-free survival [EFS] rate ± SE: 83% ± 1% [n = 1,339] and 45% ± 1% [n = 2,327], respectively; P < .0001). Graphical analyses revealed the continuous nature of the prognostic contribution of age to outcome. The optimal 460-day cutoff we selected maximized the outcome difference between younger and older patients. Forty-three percent were younger than 460 days, and 57% were 460 days old (4-year EFS rate ± SE: 82% ± 1% [n = 1,589] and 42% ± 1% [n = 2,077], respectively; P < .0001). Using a 460-day cutoff (assuming stage 4, MYCN-amplified patients remain high-risk), 5% of patients (365 to 460 days: 4-year EFS 92% ± 3%; n = 135) fell into a lower risk group.
CONCLUSION: The prognostic contribution of age to outcome is continuous in nature. Within clinically relevant risk stratification, statistical support exists for an age cutoff of 460 days.
INTRODUCTION
Neuroblastoma patients in the Children's Oncology Group (COG) are currently stratified into low-, intermediate-, and high-risk groups based on age at diagnosis (< 365 days v 365 days),1 International Neuroblastoma Staging System (INSS) stage (1, 2, 3, 4, 4s),2,3 MYCN status (genomic amplification v nonamplified),4,5 Shimada histopathology (favorable v unfavorable),6,7 and tumor cell ploidy (DNA index = 1.0 v > 1.0).8 Treatment is tailored according to risk. Survival rates for patients with low- and intermediate-risk disease are excellent, and ongoing clinical trials are focused on reducing therapy and decreasing long-term treatment-related sequelae.9-12 Most patients with low-risk disease are treated with surgery alone, whereas children with intermediate-risk neuroblastoma receive four to eight cycles of chemotherapy and surgery. In contrast, long-term outcome of children with high-risk disease remains poor13 if treated with nonmyeloablative chemotherapy, but is improved for those treated with intensive induction chemotherapy, surgery, radiation, myeloablative therapy, and 13-cis-retinoic acid for minimal residual disease.14
Because treatment strategies and inherent toxicities differ significantly between high-risk and low- and intermediate-risk patients, it is important to optimize treatment stratification criteria. The currently accepted age cutoff of 365 days was based on observations by Breslow and McCann more than 30 years ago that older age was associated with worse outcome.1 Using logistic regression to examine the relationship of age and Evans stage15 to outcome in 246 children, they found that survival probabilities decreased with increasing age during the first year or two, but age effect tapered off thereafter. Although these results suggest that days of age should be used for risk stratification as a continuous variable, this is not clinically practical. Instead, age has been used as a prognostic variable with the convenient cutoff of 365 days, and several studies have demonstrated the significant clinical impact of this age cutoff.7,16-18 However, excellent outcome also has been observed recently in a subset of toddlers (12 to 18 months of age) with stage 4 disease who were treated with multiagent chemotherapy, suggesting that 365 days may not be the best age cutoff for risk stratification.19,20 To identify the statistically optimal age cutoff, we analyzed the influence of age on outcome in 3,666 patients after adjusting for known prognostic factors INSS stage and MYCN status.
METHODS
Patients
We enrolled 3,666 patients in Pediatric Oncology Group (POG) and Children's Cancer Group (CCG) studies from 1986 to 2001 (both groups are now part of the Children's Oncology Group [COG]). All POG patients were enrolled in biology study 9047,21 and the primary POG therapeutic trials were 8742,22 9243, 9340,23 9341, and 9342. CCG patients were enrolled on therapeutic trials 321p1, 321p2, 321p3, 321p4,24 3881, 3891,14 or biology study B973. Patients had to meet eligibility requirements of their biology or therapeutic study, at a minimum: (1) diagnosis of neuroblastoma confirmed by central pathology review; (2) age 21 years; and (3) written informed consent. Data for age and stage were collected on all studies, and tumor tissue for biologic studies was required or strongly encouraged. MYCN status was determined by Southern blotting (before July 1993) or fluorescence in situ hybridization (in POG/COG after July 1993)5,25-27 or tissue expression with immunoperoxidase staining for MYCN protein (in CCG after 1993). For CCG patients and some POG patients, central pathology review to determine Shimada histopathology was performed.6,7 For POG patients, ploidy (DNA index [DI] of tumor cells determined by flow cytometry) was used.8 The patient's age was determined on the date of diagnosis, where the date of diagnosis was the date of the pathology review diagnosis of neuroblastoma from the first surgery, biopsy, or bone marrow biopsy.
Statistical Considerations
For clinically relevant analyses, the cohort was divided into 20 age groups, each with a sufficient number of patients for descriptive purposes. Because of preponderance of younger patients, groups spanned narrow age ranges at younger ages and wider ranges at older ages. For inferential analyses to identify an optimal cutoff, the cohort was repeatedly divided into two groups. In addition to the 19 cutoffs that created the 20 groups, a concentration of eight more cutoffs were tested between 12 and 24 months to provide a sufficient level of detail within the age range of clinical interest. Twenty-seven paired groups were studied, each separated by a different cutoff. The outcome difference between each paired group was quantified by the P value and the hazard ratio (henceforth referred to as relative risk [RR]) from a multivariable Cox proportional hazards model.28
To ensure that age was prognostic and not a spurious statistically significant result due to large sample size, a two-fold cross-validation methodology was employed.29,30 The cohort was divided at random into set 1 and set 2. Within each set, univariate (log-rank) and multivariable (Cox proportional hazards) testing for significance of age was performed.28 For multivariate analysis, stepwise backward model building identified statistically significant (P < .05) factors and reflected inter-relationships.31 Once age was shown to be prognostic within each subset, set 1 and set 2 were combined, and the age cutoff was determined in the overall cohort.29,30 Two approaches using a Cox model were used to select the cutoff: (a) a minimum corrected P value approach31; and (b) a maximum hazard rate (RR) approach. Survival tree regression was performed32 to demonstrate the influence age cutoff has on the entire structure of risk stratification. To identify groups of patients with differing prognoses, all possible trees for all possible age cutoffs were formed using INSS stage, MYCN status, ploidy, Shimada histopathology, and age.
We checked for sufficient power to support the number of covariates in the Cox model for the 719 events in this cohort.33,34 For event-free survival (EFS), events considered were relapse, disease progression, secondary malignancy, and death, whichever occurred first. Survival rates are quoted as the rate ± SE and are calculated using the methods of Kaplan and Meier35 and Peto.36
RESULTS
Clinical and Biologic Characteristics of Study Cohort
Median age was 573 days; the number of patients decreased exponentially with increasing age (Fig 1). Sixty-four percent of patients (n = 2,327) were 365 days old, 44% had INSS stage 4 disease (n = 1,522), 18% had MYCN-amplified tumors (n = 520), 29% had diploid tumors (n = 483), and 43% had tumors with unfavorable histopathology (n = 665). Median follow-up time was 5.8 years. Overall 8-year EFS and overall survival rates were 55% ± 2% and 61% ± 2%, respectively (n = 3,666). The EFS rate of patients 365 days old was lower than that of patients less than 365 days old (4-year EFS rates: 83% ± 1% [n = 1,339] and 45% ± 1% [n = 2,327], respectively; P < .0001; Table 1). The long-term EFS rate decreased with increasing age (Fig 2).
Relationship of Age With MYCN Status, Stage, Ploidy, and Histopathology
When age was plotted against the proportion of patients with specific clinical or biologic features, the curves showed the relative distribution of patients with and without particular features (Fig 3). In general, before the age at which a pair of curves crossed, the cohort was dominated by patients whose prognoses were favorable based on the given feature. After the age at which the curves crossed, most patients had the less favorable feature. A crossing or confluence of curves occurred at a point beyond 365 days for MYCN status (12.3 months), stage (20 months), ploidy (14 months), and histopathology (17 months; Fig 3).
Outcome by Age Groups
In a univariate analysis (n = 2,825 with complete data for INSS stage, MYCN status, and follow-up), the risk of an event for older age groups was roughly 3 to 4 times that of the reference group, patients age 0 to 21 days (Fig 4A). Sixty-five percent of patients with MYCN-amplified tumors were diagnosed at 11.5 to 40 months old, and most stage 4 patients were diagnosed at older than 30 months (Figs 3A-B). Many events of older patients may be associated with advanced stage or MYCN amplification and not necessarily older age. To adjust for this, we included age group (P < .0001), stage (P < .0001), and MYCN status (P < .0001) in a multivariable Cox model, resulting in an arguably more clinically useful model than with age alone. This adjustment accounted for events due to MYCN status and stage, and what remained was a 50% to 70% increased risk of an event for older patients (Fig 4B). Before 600 days, there was decreased risk of an event, and after 600 days there was increased risk.
Determination of Optimal Age Cutoff
To identify an age cutoff that may be preferable to the 365-day cutoff, 27 potential cutoffs were evaluated, each in a separate analysis. Each analysis divided the entire cohort into two groups—one that included patients older than the age cutoff and one that included those younger. We used two measures to quantify the difference in EFS between the younger and the older group—the P value and RR for the effect of age.
Each of the 27 separate analyses of age cutoff adjusted for stage and MYCN status (n = 2,825). A 573-day cutoff had the lowest P value (corrected P = 9 x 10–32; Fig 5A). Due to the fact that under certain circumstances, the greatest statistical power is reached when a cohort is divided into two groups of equal sample size, it is perhaps not a coincidence that 573 days was also the median age. The P values for cutoffs from 15 to 22 months were all extremely small (P < 1 x 10–30), and any cutoff in this range could be considered for use in risk stratification. An age cutoff of 460 days had the largest RR for an event (RR = 2.598); however, any cutoff from 14 to 19 months has large enough RR to be considered for use in risk stratification (Fig 5B). As presented in Table 1, the 4-year EFS ± SE of patients aged 365 to 460 days is 73% ± 3% (n = 250). If INSS stage 4 patients with MYCN-amplified tumors were excluded, the 4-year EFS of patients 365 to 460 days old was excellent: 92% ± 3% (n = 135).
In the survival tree regression, any cutoff from 570 to 670 days resulted in a stratification that optimized the prognostic contribution of age and other risk group factors. Other cutoffs resulted in different stratification trees, though none resulted in a stratification tree in which age was as highly statistically significant.
DISCUSSION
In neuroblastoma, age can be thought of as a surrogate risk factor for genetic or biologic markers of risk that are yet undiscovered or unproven. Age will remain important in determination of risk group and treatment until molecular markers are identified that provide similar prognostic information. This analysis of a large cohort will improve the use of age as a prognostic factor.
We have shown biologic and statistical support of a choice of age cutoff more than 365 days. (1) Curves of the proportion of patients by risk factor crossed or came together between 12.2 to 20 months. (2) Adjusted for stage and MYCN status, there was decreased risk of an event at less than 600 days (19.7 months), and increased risk of an event at 600 days. (3) The minimum P value for age cutoff occurred at 573 days (18.8 months). (4) The maximum RR for age cutoff occurred at 460 days (15.1 months). (5) The stratification tree that made optimal use of the prognostic contribution of age had a cutoff between 570 and 670 days (18.8 to 22.0 months). All of these choices for age cutoff fall within a range of 15 to 20 months, and any choice in this range would be statistically valid and well supported by this large data cohort.
Besides evidence of a relationship between age and outcome, there is biologic evidence of a relationship between age and MYCN status, stage, ploidy, and histopathology. We propose that these relationships provide empirical biologic evidence that the age cutoff of 365 days is too low, and that the time point beyond which unfavorable risk factors begin to dominate is higher than 365 days in each case.
Our findings indicate that the age cutoff should be increased for COG and for international risk group stratification systems for neuroblastoma. If the cutoff were increased to more than 365 days, there is a subset of patients who would be classified in a lower risk group than if an age cutoff of 365 days had been used. Because these patients would be considered to have lower risk for disease recurrence under the newer paradigm, they would receive less therapy than they would with use of a cutoff of 365 days. If a 460-day cutoff were selected, approximately 5% of patients would shift to a lower risk group, or 11% for a 573-day cutoff. This calculation excludes patients older 365 days old at diagnosis who are INSS stage 4 with MYCN-amplified tumors, who are still considered to be in a higher risk group.
Although it is possible that prognostic factor analyses could provide misleading findings,37 a strength of this analysis was the large sample size. Also, patients in this cohort met eligibility requirements of their biology or therapeutic studies; however, some bias could have been introduced due to missing data. For most patients, tumor specimen submission was an eligibility requirement and was met in most cases. Specimens were analyzed for MYCN status, ploidy (POG), and histopathology (CCG). Adjustment for INSS stage and MYCN status was performed in the analyses, but there were 841 patients (23%) for whom MYCN status and/or stage data were unavailable. These patients had to be excluded from analyses used to determine the optimal age cutoff. It is possible that data for these patients were not missing at random. Tumor specimens of sufficient size and quality tend to be less readily available in patients with INSS stage 3 to 4 tumors, and stage 3 to 4 patients may have been under-represented because of missing MYCN status.
A potential confounding issue was the lack of homogeneity of patients' treatments. Treatment is known for most CCG patients because of their enrollment in therapeutic studies, but many POG patients were enrolled only in biology study 9047. However, it is reasonable to assume that patients were treated in accordance with the standard of care for the known risk factors at that time. Due to the wide variety of treatments, and because treatment was unknown for some patients, it was not possible to adjust for treatment in the models.
A paradigm whereby older patients receive more therapy than younger patients has been used in neuroblastoma treatment for several decades, long before the first patients in this cohort were diagnosed. One could postulate that the good outcome of patients slightly older than 365 days was simply a result of more therapy for patients older than 365 days. The report of Schmidt et al19 on study CCG-3891 provided evidence that this difference was an age effect and not a treatment effect. The EFS rate for patients 12 to 18 months old (6-year EFS: 87% ± 8.8%) was significantly better than the EFS rate for patients 18 to 24 months (6-year EFS: 36% ± 14.5%) or for patients older than than 24 months (6-year EFS: 23% ± 3.3%) who received the same therapy.19
With a change in age cutoff, the risk/benefit of an associated change in therapy must also be considered.38 Before a decision is made regarding a new age cutoff, consideration must be given to the number of patients who will be permitted to relapse under the lesser therapy in order to allow a greater number of nonrelapsing patients to avoid the toxicity and long-term side effects of the more intensive therapy.
In conclusion, the prognostic contribution of age to outcome in neuroblastoma is continuous in its nature. The worsening of outcome with increasing age is gradual, with no clear delineation of an age cutoff. There exists strong statistical, and perhaps biologic evidence in this cohort for an age cutoff greater than 365 days. The closer the new cutoff is to 365 days, the more conservative the choice, because fewer patients would be shifted to a lower risk group. There is statistical support for any choice of age cutoff between 15 and 19 months for use in risk stratification of neuroblastoma patients. A new cutoff of 460 days (15.1 months) at this time would be conservative, yet allow decreased therapy in an appropriate cohort of patients (approximately 5%).
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
Acknowledgment
We acknowledge the Pediatric Oncology Group and Children's Cancer Group for supplying tumor samples. We thank Dina Ingle for data management support, Christina Rigby for logistical support, and Sue Rowe for technical laboratory assistance.
NOTES
Supported in part by Pediatric Oncology Group Statistics and Data Center grant U10 CA29139, Children's Cancer Group Statistics and Data Center grant U10 CA13539, and Children's Oncology Group Statistics and Data Center grant U10 CA98413-01.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Breslow N, McCann B: Statistical estimation of prognosis for children with neuroblastoma. Cancer Res 31:2098-2103, 1971
Smith EI, Haase GM, Seeger RC, et al: A surgical perspective on the current staging in neuroblastoma: The International Neuroblastoma Staging System proposal. J Pediatr Surg 24:386-390, 1989
Brodeur GM, Pritchard J, Berthold F, et al: Revisions of the international criteria for neuroblastoma diagnosis, staging, and response to treatment. J Clin Oncol 11:1466-1477, 1993
Brodeur GM, Seeger RC, Schwab M, et al: Amplification of N-myc in untreated human neuroblastomas correlates with advanced disease stage. Science 224:1121-1124, 1984
Seeger RC, Brodeur GM, Sather H, et al: Association of multiple copies of the N-myc oncogene with rapid progression of neuroblastomas. N Engl J Med 313:1111-1116, 1985
Shimada H, Chatten J, Newton WA Jr, et al: Histopathologic prognostic factors in neuroblastic tumors: Definition of subtypes of ganglioneuroblastoma and an age-linked classification of neuroblastomas. J Natl Cancer Inst 73:405-416, 1984
Joshi VV, Cantor AB, Altshuler G, et al: Age-linked prognostic categorization based on a new histologic grading system of neuroblastomas: A clinicopathologic study of 211 cases from the Pediatric Oncology Group. Cancer 69:2197-2211, 1992
Look AT, Hayes FA, Nitschke R, et al: Cellular DNA content as a predictor of response to chemotherapy in infants with unresectable neuroblastoma. N Engl J Med 311:231-235, 1984
Perez CA, Matthay KK, Atkinson JB, et al: Biologic variables in the outcome of stages I and II neuroblastoma treated with surgery as primary therapy: A Children's Cancer Group study. J Clin Oncol 18:18-26, 2000
Schmidt ML, Lukens JN, Seeger RC, et al: Biologic factors determine prognosis in infants with stage IV neuroblastoma: A prospective Children's Cancer Group study. J Clin Oncol 18:1260-1268, 2000
Nickerson HJ, Matthay KK, Seeger RC, et al: Favorable biology and outcome of stage IV-S neuroblastoma with supportive care or minimal therapy: A Children's Cancer Group study. J Clin Oncol 18:477-486, 2000
Matthay KK, Perez C, Seeger RC, et al: Successful treatment of stage III neuroblastoma based on prospective biologic staging: A Children's Cancer Group study. J Clin Oncol 16:1256-1264, 1998
Brodeur GM, Maris JM: Neuroblastoma, in Pizzo PA, Poplack DG (eds): Principles and Practice of Pediatric Oncology (ed 4). Philadelphia, PA, Lippincott-Raven, 2001, pp 895-937
Matthay KM, Villablanca JG, Seeger RC, et al: Treatment of high-risk neuroblastoma with intensive chemotherapy, radiotherapy, autologous bone marrow transplantation, and 13-cis-retinoic acid. N Engl J Med 341:1165-1173, 1999
Evans AE: Staging and treatment of neuroblastoma. Cancer 45:1799-1802, 1980
Evans AE, D'Angio GJ, Sather HN, et al: A comparison of four staging systems for localized and regional neuroblastoma: A report from the Children's Cancer Study Group. J Clin Oncol 8:678-688, 1990
Berthold F, Kassenbohmer R, Zieschang J: Multivariate evaluation of prognostic factors in localized neuroblastoma. Am J Pediatr Hematol Oncol 16:107-115, 1994
Saito T, Tsunematsu Y, Saeki M, et al: Trends of survival in neuroblastoma and independent risk factors for survival at a single institution. Med Pediatr Oncol 29:197-205, 1997
Schmidt ML, Lal A, Seeger R, et al: Favorable prognosis for patients ages 12-18 months with stage 4 MYCN-nonamplified neuroblastoma. Proc Am Soc Clin Oncol 22:800, 2003 (abstr 3214)
George R, London WB, Maris JM, et al: Age as a continuous variable in predicting outcome for neuroblastoma patients with metastatic disease: Impact of tumor biologic features. Proc Am Soc Clin Oncol 22:799, 2003 (abstr 3213)
Alvarado CS, London WB, Look AT, et al: Natural history and biology of stage A neuroblastoma: A Pediatric Oncology Group study. J Pediatr Hematol Oncol 22:197-205, 2000
Strother D, van Hoff J, Rao PV, et al: Event-free survival of children with biologically favourable neuroblastoma based on the degree of initial tumour resection: Results from the Pediatric Oncology Group. Eur J Cancer 33:2121-2125, 1997
Kretschmar CS, Kletzel M, Murray K, et al: Response to paclitaxel, topotecan, and topotecan-cyclophosphamide in children with untreated disseminated neuroblastoma treated in an upfront phase II investigational window: A Pediatric Oncology Group study. J Clin Oncol 22:4119-4126, 2004
Matthay KK, O'Leary MC, Ramsay NK, et al: Role of myeloablative therapy in improved outcome for high risk neuroblastoma: Review of recent Children's Cancer Group results. Eur J Cancer 31A:572-575, 1995
Schwab M, Alitalo K, Klempnauer KH, et al: Amplified DNA with limited homology to myc cellular oncogene is shared by human neuroblastoma cell lines and a neuroblastoma tumour. Nature 305:245-248, 1983
Shapiro DN, Valentine MB, Rowe ST, et al: Detection of N-myc gene amplification by fluorescence in situ hybridization: Diagnostic utility for neuroblastoma. Am J Pathol 142:1339-1346, 1993
Mathew P, Valentine MB, Bowman LC, et al: Detection of MYCN gene amplification in neuroblastoma by fluorescence in situ hybridization: A Pediatric Oncology Group study. Neoplasia 3:105-109, 2001
Cox DR: Regression models and life tables. J R Stat Soc B 34:187-220, 1972
Mazumdar M, Smith A, Bacik J: Methods for categorizing a prognostic variable in a multivariable setting. Stat Med 22:559-571, 2003
Faraggi D, Simon R: A simulation study of cross-validation for selecting an optimal cutpoint in univariate survival analysis. Stat Med 15:2203-2213, 1996
Altman DG, Lausen B, Sauerbrei W, et al: Dangers of using "optimal" cutpoints in the evaluation of prognostic factors. J Natl Cancer Inst 86:829-835, 1994
Segal MR: Regression trees for censored data. Biometrics 44:35-47, 1988
Harrell FE Jr, Lee KL, Califf RM, et al: Regression modelling strategies for improved prognostic prediction. Stat Med 3:143-152, 1984
Feinstein AR. Multivariable Analysis: An Introduction. New Haven, CT, Yale University Press, 1996, pp 184-187, pp 578-582
Kaplan EL, Meier P: Nonparametric estimation from incomplete observations. J Am Stat Assoc 53:457-481, 1958
Peto R, Peto J: Asymptotically efficient rank invariant test procedures. J R Stat Soc A 135:185-198, 1972
Simon R, Altman DG: Statistical aspects of prognostic factor studies in oncology. Br J Cancer 69:979-985, 1994
Grosfeld JL: Risk-based management: Current concepts of treating malignant solid tumors of childhood. J Am Coll Surg 189:407-425, 1999(W.B. London, R.P. Castleb)