|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 26, No 3 (January 20), 2008: pp. 380-385 © 2008 American Society of Clinical Oncology. DOI: 10.1200/JCO.2007.14.1291 Individualized Prediction of Colon Cancer Recurrence Using a Nomogram
From the Departments of Surgery, Pathology, Epidemiology-Biostatistics, and Medical Oncology, Memorial Sloan-Kettering Cancer Center, New York, NY Corresponding author: Martin R. Weiser, MD, Colorectal Service, Department of Surgery, Memorial Sloan-Kettering Cancer Center, Room C-1075, 1275 York Ave, New York, NY 10021; e-mail: weiser1{at}mskcc.org
Purpose Estimates of recurrence after curative colon cancer surgery are integral to patient care, forming the basis of cancer staging and treatment planning. The categoric staging system of the American Joint Committee on Cancer (AJCC) is commonly used to convey risk by grouping patients based on anatomic elements. Although easy to implement, there remains significant heterogeneity within each stage grouping. In the era of multimodality treatment, a more refined tool is needed to predict recurrence. Methods An institutional database of 1,320 patients with nonmetastatic colon cancer was used to develop a nomogram to estimate recurrence after curative surgery. Prognostic factors were assessed with multivariable analysis using Cox regression, whereas nonlinear continuous variables were modeled with cubic splines. The model was internally validated with bootstrapping, and performance was assessed by concordance index and a calibration curve. Results The colon cancer recurrence nomogram predicted relapse with a concordance index of 0.77, improving on the stratification provided by either the AJCC fifth or sixth staging scheme. Factors in the model included patient age, tumor location, preoperative carcinoembryonic antigen, T stage, numbers of positive and negative lymph nodes, lymphovascular invasion, perineural invasion, and use of postoperative chemotherapy. Conclusion Using common clinicopathologic factors, the recurrence nomogram is better able to account for tumor and patient heterogeneity, thereby providing a more individualized outcome prognostication than that afforded by the AJCC categoric system. By identifying both the high- and low-risk patients within any particular stage, the nomogram is expected to aid in treatment planning and future trial design.
Adenocarcinoma of the colon afflicts more than 1 million individuals worldwide. It is the most common GI malignancy in the United States, affecting more than 100,000 individuals annually.1 Although surgical resection is the mainstay of treatment for patients with localized, nonmetastatic disease, adjuvant chemotherapy can reduce the risk of recurrence by 40% to 50% in high-risk patients.2,3 However, the current chemotherapy regimens have significant adverse effects, with the vast majority of patients experiencing some peripheral neuropathy and as many as 40% of patients experiencing grade 3 or 4 neutropenia.2,3 Therefore, tools that predict recurrence are critical in helping physicians counsel and advise patients about whether or not to pursue adjuvant chemotherapy. The present gold standard for risk assessment is the TNM anatomic tumor staging system, adopted from the original scheme devised more than 75 years ago by Dukes4 and currently supported by the American Joint Commission on Cancer (AJCC) and the International Union Against Cancer.5 The TNM staging system has been widely adopted because it is easy to implement and parsimonious. It is based on the following two elements for patients without metastatic disease: extent of tumor penetration into the intestinal wall and number of locoregional lymph nodes containing metastases. Patients are broadly grouped into stage cohorts (stages I to III) with distinct outcome; however, in this simple scheme, the predictive accuracy is limited. Patients within each group are assumed to have homogenous outcomes when, in fact, outcomes are quite heterogeneous as a result of variability in clinicopathologic features and tumor biology.6 Additionally, the categoric nature of the AJCC staging scheme forces the transformation of continuous variables into categoric ones, further limiting predictive accuracy. Thus, although the AJCC staging scheme provides an estimate of outcome, clinicians and patients are often left wondering where a particular outcome lies on the clinical spectrum. Attempts to refine the prognostic estimates by subcategorization have proven to be less than satisfactory7-9 and highlight the limitations of any staging system that creates distinct groupings. Nomograms have been developed for a variety of malignancies in an attempt to improve outcome prediction and provide patients and physicians with a more understandable outcome measure when making treatment-related decisions. These statistically based tools provide the overall probability of a specific outcome. They do not produce risk groups but, rather, attempt to combine all proven prognostic factors and quantify risk as precisely as possible.10 Previous studies in prostate cancer, soft tissue sarcoma, and gastric cancer suggest improved predictive accuracy using this approach relative to the formation of risk groups.11-13 Nomogram-based staging platforms are also flexible and readily allow the addition of other prognostic markers to the prognostic model, including adjuvant therapy2 as well as histologic and molecular determinants of tumor biology.14 We describe the construction and performance characteristics of a nomogram predicting time to recurrence for patients with resectable colon cancer. We propose that this nomogram provides more individualized outcome predictions and could aid clinicians and patients in the treatment decision-making process.
Patient Cohort Query of institutional and service databases identified 1,320 patients surgically treated at Memorial Sloan-Kettering Cancer Center for AJCC stages I to III colon cancer between January 1, 1990 and December 27, 2000. Charts were reviewed and data confirmed in all patients. All patients underwent complete (R0) resection, with no evidence of metastatic disease on initial work-up or on surgical exploration. All lesions were located from the cecum to the rectosigmoid (> 18 cm from the anal verge). Clinicopathologic variables that potentially predict tumor recurrence were collected, including age, sex, tumor location, preoperative carcinoembryonic antigen level, tumor differentiation, number of positive lymph nodes, number of negative lymph nodes, lymphovascular invasion, perineural invasion, and depth of tumor penetration into the colon wall (T stage). In cases where pathologic variables were missing from the original reports, specimens were rereviewed by a pathologist. The study was presented to and approved by the institutional review board before initiation of the study.
Creation of Model
Model Validation
Descriptive statistics for the patient cohort are listed in Table 1. At last follow-up, 243 patients had experienced recurrence. Of those alive, median follow-up time was 55 months. Freedom from recurrence is displayed in Figure 1, according to the fifth and sixth AJCC staging systems.5 The nomogram included the following demographic and clinicopathologic variables: patient age, preoperative carcinoembryonic antigen level, tumor location, tumor differentiation, presence of lymphovascular and perineural invasion, number of positive and negative lymph nodes in the specimen, depth of tumor penetration (T stage), and whether the patient received adjuvant chemotherapy (Fig 2). Outcome was reported as 5- and 10-year freedom from recurrence. The associated calibration curves from the nomograms at 5 and 10 years are shown in Figure 3.
The concordance index of the nomogram was 0.77 for predicting freedom from recurrence at 10 years. This was superior to both predictions based on the AJCC fifth and sixth staging systems, with concordance indices of 0.73 and 0.74, respectively. A histogram of nomogram-predicted probabilities within each of the AJCC stages is shown in Figure 4 and depicts the variation in predicted outcome within each of the AJCC version 6 subgroups.
We have developed and internally validated a tool for predicting recurrence of colon cancer after surgery, assuming complete resection of the primary disease. In contrast to systems that assign prognosis based on risk groups, nomograms estimate risk based on a combination of variables, permitting a more individualized prediction of outcome. A wide variety of prognostic factors can be incorporated into the nomogram, including nonlinear and more complex relationships. Data do not have to be condensed into the categories necessary for traditional staging schemes. Therefore, it is not surprising that the recurrence colon cancer nomogram performs better than the AJCC system because it is better able to account for heterogeneity in tumor biology and patient characteristics. The price of a more individualized outcome measure is increased complexity. The AJCC staging system is easy to remember and use. One might argue that the current AJCC staging system could be refined to improve its accuracy by subcategorization; however, past attempts at creating more groups have been unsuccessful. In the sixth edition of the AJCC scheme released in 2002, for example, stages II and III patients were substratified into two and three groups, respectively. Model performance did not improve and, in fact, may have worsened. An analysis by O'Connell et al7 of more than 100,000 Surveillance, Epidemiology, and End Results colon cancer patients noted a loss of the orderly relationship between stage and outcome that is crucial to any categoric staging system. Specifically, survival curves crossed, with stage IIB patients faring worse than stage IIIA patients (5-year cancer-related survival rates of 72% for stage IIB and 83% for stage IIIA; P < .01).7 Although stage III patients generally receive adjuvant therapy, whereas stage II patients do not, Jeong et al9 have argued that this does not explain the finding reported by O'Connell et al.7 These irregularities created controversy and emphasize the limitations of any categoric staging system.8,9 In our cohort, we found similar results when plotting recurrence by AJCC version 6 grouping (Fig 1B). The heterogeneity in outcome within each stage is also demonstrated in Figure 4, which is a histogram of predicted outcome based on the nomogram for each AJCC stage classification. The variability in probability of recurrence is clearly evident. In the stage II cohort, for example, the nomogram predicted probability of 10-year freedom from recurrence spanning the entire gamut of 0.1 to 0.99, and significant heterogeneity remained when the cohort was subdivided into stages IIA and IIB. Similarly, great heterogeneity exists within the stage III cohort, with some stage III patients having estimated outcomes better than some stage II patients. Clearly, not all stage II or III patients should be treated identically. In this way, the nomogram could assist patients and physicians to decide on adjuvant treatment options. We would argue that the nomogram does not have to replace the estimates provided by the AJCC system but, rather, can be used to fine-tune and refine them. The emphasis on a more accurate outcome estimate may not be appealing to all physicians and patients. Certainly, many physicians currently use the prognostic factors included in a nomogram when discussing outcome and the potential benefit of adjuvant chemotherapy with patients.19,20 However, the nomogram provides a sounder mechanism for conveying the impact of multiple prognostic factors. The nomogram may be most valuable for those individuals in whom the potential benefit of additional therapy seems marginal. Patients and physicians can determine what risk of recurrence warrants additional therapy. This may be especially useful when discussing the role of adjuvant chemotherapy in patients with node-negative disease, for whom most trials have not found a clear significant benefit.2,3,7,20
In a similar manner, nomograms have tremendous potential as estimators of risk in clinical trial design. Patients can be stratified into treatment groups based on their estimated risk of recurrence or survival. This approach is currently being used in the Cancer and Leukemia Group B 90203 trial, which is a randomized study comparing prostatectomy alone versus estramustine and docetaxel before prostatectomy in high-risk patients.21 A nomogram is used to identify patients with a predicted probability of Close scrutiny of the nomogram reveals interesting interactions between variables, which is not surprising considering the complex nature of any malignancy. First, the negative association between number of positive lymph nodes, T stage, and recurrence is blunted by the administration of postoperative chemotherapy. This may be explained by the significant risk reduction associated with adjuvant chemotherapy, noted in prospective randomized trials.2,3 The interaction between adjuvant therapy, number of nodal metastases, T stage, and outcome justifies the addition of chemotherapy in our prognostic model. Lack of accounting for this interaction has been implicated as a deficit in other models.7 Although we have constructed an improved prognostic instrument based on a large data set of colectomy patients, its predictive accuracy is not perfect, and there is room for improvement. As prognostic markers are characterized, including molecular profiles,14 the nomogram can be easily modified, which is one of the benefits of this type of prognostic model. There are limitations to our analysis, including the fact that all surgical procedures were performed at a specialty cancer center; therefore, the nomogram may not be transferable to other patient cohorts. For example, the vast majority of patients were treated in an elective fashion after complete work-up. Therefore, few if any patients were included with complete obstruction or free perforation on presentation, and these variables were not included in the model. However, the limited number of surgeons involved and the standardized pathology limit variability and are advantageous when modeling outcome. Clearly, validation of any prognostic model is important before widespread adoption. Although there have been significant advances in chemotherapy for colon cancer over the last decade,2 the nomogram includes patients treated between 1990 and 2000, during which fluorouracil was the sole active agent used in the adjuvant setting. Therefore, this nomogram reflects recurrence in the era before administration of oxaliplatin as a component of adjuvant therapy. Beginning in 2004, subsequent to the publication of the MOSAIC study, infusional fluorouracil, leucovorin, and oxaliplatin became the standard adjuvant colon cancer regimen.2 Updated versions of the nomogram will need to refine prognosis in an era of more highly effective chemotherapy. Caution must be exercised when assessing the response to chemotherapy using the nomogram because the data are not from randomized trials. The treatment of metastatic disease has varied during this time period; however, this does not confound the nomogram, which reports time to first recurrence. In summary, we have created a nomogram that predicts recurrence after complete resection of nonmetastatic colon cancer. An electronic version is located for use at www.nomograms.org. This model improves on the estimates provided by the current AJCC staging system and provides an individualized risk assessment that is easily understood by physicians and patients. Using the nomogram, patients and physicians may better understand the risk of recurrence, which could influence their decision to pursue adjuvant therapy and plan long-term follow-up.
The author(s) indicated no potential conflicts of interest.
Conception and design: Martin R. Weiser, Michael W. Kattan, W. Douglas Wong Administrative support: Martin R. Weiser, W. Douglas Wong Collection and assembly of data: Martin R. Weiser, Ron G. Landmann, Mithat Gonen, Jinru Shia, Joanne Chou, Larissa K. Temple Data analysis and interpretation: Martin R. Weiser, Ron G. Landmann, Michael W. Kattan, Mithat Gonen, Joanne Chou, Philip B. Paty, José G. Guillem, Larissa K. Temple, Deborah Schrag Manuscript writing: Martin R. Weiser, Ron G. Landmann, Jinru Shia, Leonard B. Saltz Final approval of manuscript: Martin R. Weiser, Jinru Shia, Deborah Schrag, Leonard B. Saltz, W. Douglas Wong
Supported in part by grants from the American Joint Committee on Cancer (M.R.W.) and from the Society of Memorial Sloan-Kettering (M.R.W.). Authors’ disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. Jemal A, Murray T, Ward E, et al: Cancer statistics, 2005. CA Cancer J Clin 55:10-30, 2005 2. Andre T, Boni C, Mounedji-Boudiaf L, et al: Oxaliplatin, fluorouracil, and leucovorin as adjuvant treatment for colon cancer. N Engl J Med 350:2343-2351, 2004 3. Adjuvant therapy for patients with colon and rectum cancer. Consens Statement 8:1-25, 1990[Medline] 4. Dukes CE: The classification of cancer of the rectum. J Pathol Bacteriol 35:323, 1932[CrossRef] 5. Green FL, Page DL, Fleming ID, et al: AJCC Cancer Staging Manual (ed 6). Chicago, IL, American Joint Committee on Cancer, 2002 6. Skibber J, Minsky B, Hoff P: Cancer of the colon, in DeVita V, Hellman S, Rosenberg S (eds): Cancer Principles and Practice of Oncology (ed 6). New York, NY, Lippincott Williams & Wilkins, 2001, pp 1216-1262 7. O'Connell JB, Maggard MA, Ko CY: Colon cancer survival rates with the new American Joint Committee on Cancer sixth edition staging. J Natl Cancer Inst 96:1420-1425, 2004 8. Burke HB: Outcome prediction and future of the TNM staging system. J Natl Cancer Inst 96:1408-1409, 2005 9. Jeong SY, Chessin DB, Schrag D, et al: Re: Colon cancer survival rates with the new American Joint Committee on Cancer 6th edition staging. J Natl Cancer Inst 97:1705-1706, 2005 10. Eastham JA, Kattan MW, Scardino PT: Nomograms as predictive models. Semin Urol Oncol 20:108-115, 2002[CrossRef][Medline] 11. Kattan MW, Leung DH, Brennan MF: Postoperative nomogram for 12-year sarcoma-specific death. J Clin Oncol 20:791-796, 2002 12. Kattan MW, Zelefsky MJ, Kupelian PA, et al: Pretreatment nomogram that predicts 5-year probability of metastasis following three-dimensional conformal radiation therapy for localized prostate cancer. J Clin Oncol 21:4568-4571, 2003 13. Kattan MW, Karpeh MS, Mazumdar M, et al: Postoperative nomogram for disease-specific survival after an R0 resection for gastric carcinoma. J Clin Oncol 21:3647-3650, 2003 14. Stephenson AJ, Smith A, Kattan MW, et al: Integration of gene expression profiling and clinical variables to predict prostate carcinoma recurrence after radical prostatectomy. Cancer 104:290-298, 2005[CrossRef][Medline] 15. Harrell FE Jr, Lee KL, Mark DB: Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 15:361-387, 1996[CrossRef][Medline] 16. Kattan MW, Eastham JA, Stapleton AM, et al: A preoperative nomogram for disease recurrence following radical prostatectomy for prostate cancer. J Natl Cancer Inst 90:766-771, 1998 17. Hanley JA, McNeil BJ: The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143:29-36, 1982 18. Harrell FE Jr: Regression Modeling Strategies With Applications to Linear Models, Logistic Regression, and Survival Analysis, Volume 19. New York, NY, Springer-Verlag, 2001 19. Schrag D: Defining optimal treatment for stage II colon cancer: Does decision analysis help? Gastroenterology 117:1005-1008, 1999[CrossRef][Medline] 20. Benson AB III, Schrag D, Somerfield MR, et al: American Society of Clinical Oncology recommendations on adjuvant chemotherapy for stage II colon cancer. J Clin Oncol 22:3408-3419, 2004 21. Eastham JA, Kelly WK, Grossfeld GD, et al: Cancer and Leukemia Group B (CALGB) 90203: A randomized phase 3 study of radical prostatectomy alone versus estramustine and docetaxel before radical prostatectomy for patients with high-risk localized disease. Urology 62: 55-62, 2003 (suppl 1)[Medline] Submitted August 24, 2007; accepted October 19, 2007.
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2008 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
60% of remaining free of disease recurrence 5 years after prostatectomy. The lead investigators argue that, by computing a continuous probability of failure, the nomogram provides a more accurate prediction than can be accomplished by placing a patient into one of two or three risk groups. More importantly, a nomogram allows flexibility in trial design when choosing the most appropriate cut point for dichotomization, which should better balance the study cohorts. This type of approach would be valuable in colon cancer, where the benefit of adjuvant therapy for node-negative disease has been difficult to prove.
