- Email: [email protected]
3–30 Nonsentinel Node Metastasis in Breast Cancer Patients: Assessment of an Existing and a New Predictive Nomogram
3–30
Nonsentinel Node Metastasis in Breast Cancer Patients: Assessment of an Existing and a New Predictive Nomogram Degnim AC, Reynolds C, Pantvaidya G, et al (Mayo Clinic, Rochester, MN; Univ of Michigan, Ann Arbor) Am J Surg 190:543-550, 2005
Background.—The accurate prediction of nonsentinel node (NSN) metastasis in breast cancer patients remains uncertain. Methods.—The medical records of 574 breast cancer patients from 2 different institutions (Mayo Clinic and University of Michigan) with sentinel lymph node biopsy examination and completion axillary lymph node dissection were reviewed for multiple clinicopathologic variables. The Memorial Sloan Kettering Cancer Center nomogram performance for prediction of NSN metastases was assessed. A new model was developed with clinically relevant variables and possible advantages. Results.—The Memorial Sloan Kettering Cancer Center nomogram predicted the likelihood of NSN metastasis with an area under the receiver operating characteristic curve of .72 and .86. For predicted probability cut-off points of 5% and 10%, the false-negative rates were 0% and 14% (Mayo), and 17% and 11% (Michigan). A new model was developed with similar area under the curve but lower false-negative rates for low-probability subgroups.
TABLE 4.—Associations Between Clinicopathologic Features and Positive NSNs: Multivariate Analysis Variable Intercept Age Tumor size Size of metastasis ER positive Extracapsular extension Number of positive SNs Number of negative SNs Age/estrogen receptor positive Number of positive SNs/extracapsular extension
Coefficient
Standard Error
P Value
–5.58 .06 .19 .09 4.11 1.72 .68 ⫺.2 ⫺.08 ⫺.69
.02 .06 .03 1.39 .51 .3 .08 .03 .35
.008 .001 .0009 .003 .0008 .02 .01 .002 .04
(Reprinted from Degnim AC, Reynolds C, Pantvaidya G, et al: Nonsentinel node metastasis in breast cancer patients: Assessment of an existing and a new predictive nomogram. Am J Surg 190:543-550. Copyright 2005, with permission from Excerpta Medica Inc.)
Conclusions.—Predictive models for NSN tumor burden are imperfect (Table 4). Nomograms are graphic calculation models used to estimate outcome variables when predictor variables are known. These tools function in a manner similar to slide rules and can be based on linear, logarithmic, or more complex scales. In this report, the Memorial Sloan Kettering Cancer Center nomogram for predicting NSN metastases in breast cancer was applied to a novel patient population. This nomogram performed well when applied to the University of Michigan and Mayo Clinic cohorts—it predicted
NSN status in most subjects—but it was not perfect. This tool is useful in helping guide discussions with patients when deciding whether to proceed with a complete axillary lymph node dissection or observation. As demonstrated in this paper, statistical models can be improved by adding new, clinically relevant predictors. However, the gains in accuracy achieved by altering the model in this scenario may not translate to a new set of patients because of differences in tumor factors, surgical factors, or histopathologic evaluation of the lymph nodes. T. M. Breslin, MD
®
Breast Diseases: A Year Book Quarterly Vol 17 No 3 2006
255 255
Nonsentinel Node Metastasis in Breast Cancer Patients: Assessment of an Existing and a New Predictive Nomogram Degnim AC, Reynolds C, Pantvaidya G, et al (Mayo Clinic, Rochester, MN; Univ of Michigan, Ann Arbor) Am J Surg 190:543-550, 2005
Background.—The accurate prediction of nonsentinel node (NSN) metastasis in breast cancer patients remains uncertain. Methods.—The medical records of 574 breast cancer patients from 2 different institutions (Mayo Clinic and University of Michigan) with sentinel lymph node biopsy examination and completion axillary lymph node dissection were reviewed for multiple clinicopathologic variables. The Memorial Sloan Kettering Cancer Center nomogram performance for prediction of NSN metastases was assessed. A new model was developed with clinically relevant variables and possible advantages. Results.—The Memorial Sloan Kettering Cancer Center nomogram predicted the likelihood of NSN metastasis with an area under the receiver operating characteristic curve of .72 and .86. For predicted probability cut-off points of 5% and 10%, the false-negative rates were 0% and 14% (Mayo), and 17% and 11% (Michigan). A new model was developed with similar area under the curve but lower false-negative rates for low-probability subgroups.
TABLE 4.—Associations Between Clinicopathologic Features and Positive NSNs: Multivariate Analysis Variable Intercept Age Tumor size Size of metastasis ER positive Extracapsular extension Number of positive SNs Number of negative SNs Age/estrogen receptor positive Number of positive SNs/extracapsular extension
Coefficient
Standard Error
P Value
–5.58 .06 .19 .09 4.11 1.72 .68 ⫺.2 ⫺.08 ⫺.69
.02 .06 .03 1.39 .51 .3 .08 .03 .35
.008 .001 .0009 .003 .0008 .02 .01 .002 .04
(Reprinted from Degnim AC, Reynolds C, Pantvaidya G, et al: Nonsentinel node metastasis in breast cancer patients: Assessment of an existing and a new predictive nomogram. Am J Surg 190:543-550. Copyright 2005, with permission from Excerpta Medica Inc.)
Conclusions.—Predictive models for NSN tumor burden are imperfect (Table 4). Nomograms are graphic calculation models used to estimate outcome variables when predictor variables are known. These tools function in a manner similar to slide rules and can be based on linear, logarithmic, or more complex scales. In this report, the Memorial Sloan Kettering Cancer Center nomogram for predicting NSN metastases in breast cancer was applied to a novel patient population. This nomogram performed well when applied to the University of Michigan and Mayo Clinic cohorts—it predicted
NSN status in most subjects—but it was not perfect. This tool is useful in helping guide discussions with patients when deciding whether to proceed with a complete axillary lymph node dissection or observation. As demonstrated in this paper, statistical models can be improved by adding new, clinically relevant predictors. However, the gains in accuracy achieved by altering the model in this scenario may not translate to a new set of patients because of differences in tumor factors, surgical factors, or histopathologic evaluation of the lymph nodes. T. M. Breslin, MD
®
Breast Diseases: A Year Book Quarterly Vol 17 No 3 2006
255 255