A Nomogram for Predicting Survival for Metastatic Colorectal Cancer
A Nomogram for Predicting Survival for Metastatic Colorectal Cancer
Purpose: To develop a tool for predicting survival after liver resection for patients with stage IV colorectal cancer. By using a nomogram we are trying to improve on the current practice of using prognostic scores for evaluating risks of therapeutic failure.
Patients and Methods: All patients admitted to Memorial Sloan-Kettering Cancer Center (MSKCC) for curative intent for treatment of metastatic disease from colorectal cancer between January 1986 and December 1999 were included. A nomogram was developed as a graphical representation of a Cox proportional hazards regression model. The nomogram was verified for discrimination and calibration, both employing bootstrapping to obtain relatively unbiased estimates.
Results: Using nodal status of the primary tumor, disease-free interval, size of the largest metastatic tumor, preoperative carcinoembryonic antigen, bilateral resection, extensive resection (lobectomy or more), gender, number of hepatic tumors, primary cancer site (colon vs. rectum), and age, the nomogram achieved a concordance index of 0.61, statistically significantly greater than chance. The nomogram also had very good calibration.
Conclusion: This nomogram is a predictive tool, upon external validation, that can routinely be used to counsel patients in making treatment decisions. The discriminatory ability of the nomogram indicates that this population should not be considered homogeneous with respect to risk of death.
Prior to the routine use of hepatic resection for metastatic colorectal cancer, all patients died of disease. This is the reason that traditional TNM staging classifies all patients with metastatic disease to the liver as stage IV disease. However, it has become clear that patients with hepatic metastases are not homogeneous with respect to risk of recurrence and death, particularly following hepatectomy. Liver resection results in a 5-year survival for one-third of patients. Individual patient decision making requires better tools for assessing prognosis to provide patient selection for surgery, for potentially morbid and toxic neoadjuvant and adjuvant therapy, and for potentially costly radiologic preoperative workup.
To assess risk and to aid in patient care, investigators have attempted to use prognostic scoring systems. Cady et al, Iwatsuki et al, Nordlinger et al, and Fong et al have each proposed prognostic scoring systems, using between 2 and 7 clinical criteria to aid in the assessment of risk of recurrence. Of these, the clinical risk score (CRS) as proposed by Fong et al has been used most widely. This scoring system has been validated by independent databases, and has been demonstrated useful not only in predicting recurrence, but also in predicting yield for diagnostic tests such as laparoscopy and PET scanning.
The purpose of the current study is to improve upon previous prognostic scoring systems. Rather than count risk factors, a nomogram takes the specific value for each factor into consideration. For example, instead of cons idering a preoperative carcinoembryonic antigen (CEA) value above 200 to be a risk factor, the nomogram considers the patient's actual CEA value, and calculates risk accordingly. A nomogram should be more specific to the individual patient and thus predicts more accurately. To test this hypothesis we compared predictions from the nomogram with those of the CRS.
Abstract and Introduction
Abstract
Purpose: To develop a tool for predicting survival after liver resection for patients with stage IV colorectal cancer. By using a nomogram we are trying to improve on the current practice of using prognostic scores for evaluating risks of therapeutic failure.
Patients and Methods: All patients admitted to Memorial Sloan-Kettering Cancer Center (MSKCC) for curative intent for treatment of metastatic disease from colorectal cancer between January 1986 and December 1999 were included. A nomogram was developed as a graphical representation of a Cox proportional hazards regression model. The nomogram was verified for discrimination and calibration, both employing bootstrapping to obtain relatively unbiased estimates.
Results: Using nodal status of the primary tumor, disease-free interval, size of the largest metastatic tumor, preoperative carcinoembryonic antigen, bilateral resection, extensive resection (lobectomy or more), gender, number of hepatic tumors, primary cancer site (colon vs. rectum), and age, the nomogram achieved a concordance index of 0.61, statistically significantly greater than chance. The nomogram also had very good calibration.
Conclusion: This nomogram is a predictive tool, upon external validation, that can routinely be used to counsel patients in making treatment decisions. The discriminatory ability of the nomogram indicates that this population should not be considered homogeneous with respect to risk of death.
Introduction
Prior to the routine use of hepatic resection for metastatic colorectal cancer, all patients died of disease. This is the reason that traditional TNM staging classifies all patients with metastatic disease to the liver as stage IV disease. However, it has become clear that patients with hepatic metastases are not homogeneous with respect to risk of recurrence and death, particularly following hepatectomy. Liver resection results in a 5-year survival for one-third of patients. Individual patient decision making requires better tools for assessing prognosis to provide patient selection for surgery, for potentially morbid and toxic neoadjuvant and adjuvant therapy, and for potentially costly radiologic preoperative workup.
To assess risk and to aid in patient care, investigators have attempted to use prognostic scoring systems. Cady et al, Iwatsuki et al, Nordlinger et al, and Fong et al have each proposed prognostic scoring systems, using between 2 and 7 clinical criteria to aid in the assessment of risk of recurrence. Of these, the clinical risk score (CRS) as proposed by Fong et al has been used most widely. This scoring system has been validated by independent databases, and has been demonstrated useful not only in predicting recurrence, but also in predicting yield for diagnostic tests such as laparoscopy and PET scanning.
The purpose of the current study is to improve upon previous prognostic scoring systems. Rather than count risk factors, a nomogram takes the specific value for each factor into consideration. For example, instead of cons idering a preoperative carcinoembryonic antigen (CEA) value above 200 to be a risk factor, the nomogram considers the patient's actual CEA value, and calculates risk accordingly. A nomogram should be more specific to the individual patient and thus predicts more accurately. To test this hypothesis we compared predictions from the nomogram with those of the CRS.
