The Practical Value of Prediction: A Semireliable Assistant

Abstract

The proposed utility of a nomogram is to assist decision making for complete axillary lymph node dissection (CLND) in breast cancer patients who have a positive sentinel lymph node (SLN). According to the American Joint Committee on Cancer Cancer Staging Manual (sixth edition, 2002), a positive SLN is defined as tumor deposits >.2 mm. Isolated tumor cells, clusters smaller than .2 mm, are classified as N0. Of all patients with a positive SLN, about 30% to 50% will have additional non-SLNs that contain tumor cells detectable by hematoxylin and eosin. The size of these metastases is not well documented since only in the most recent update of the American Joint Committee on Cancer staging system has size been relevant. The standard of care is to perform CLND on all patients who have positive SLNs. Accurately predicting which patients with a positive SLN are the 50% to 70% with no additional nodal disease might spare them the morbidity of a CLND. However, the therapeutic impact of such management has not been determined. Small studies suggest that further dissection may not be necessary. Indeed, extrapolation from National Surgical Adjuvant Breast and Bowel Project B04 data might even suggest that axillary surgery itself may not impact survival. Van Zee et al. published an excellent nomogram for predicting the likelihood of additional nodal metastases in 2003. The Memorial Sloan-Kettering Cancer Center (MSKCC) nomogram used eight clinical and tumor characteristics that are readily available to most clinicians to predict the likelihood that additional positive nodes will be found if the patient undergoes CLND. These variables include tumor size, histological type, nuclear grade, presence of lymphovascular invasion, multifocality, estrogen receptor status, number of positive and negative SLNs, and the method of detection of the SLN metastases. Lambert et al. applied the nomogram to 200 consecutive patients at the University of Texas M. D. Anderson Cancer Center to assess its performance. They found that the area under the receiver operating characteristic (ROC) curve was .71. The area under the ROC curve is an overall assessment of the test s ability to predict accurately. ROC curves were developed in the 1950s in the study of how to interpret radio signals contaminated by noise. An ROC curve plots a test s performance with varying thresholds of sensitivity and specificity. As sensitivity improves, specificity suffers. For fewer missed diagnoses, there are more false positives. A value of .5 represents no ability to predict, and a value of 1.0 represents perfect accuracy. ROC curves are being used more frequently to describe the accuracy of radiological tests. Surgeons are often less familiar with this kind of tool because they are trained to manage uncertainty through a less quantitative process known as surgical judgment. ROC curves are compared by measuring the area under the curve as a quantification of accuracy. A rough correlation of surgical judgment and the area under the ROC curve values in terms of reliability might coincide like this: .9 to 1.0, intuitive genius professor; .8 to .9, dedicated midcareer surgeon; .7 to .8, average chief resident; .6 to .7, disinterested intern; and .5 to .6, motivated passer-by. In their original description of the MSKCC nomogram, Van Zee et al. explained the concept of a value of .77 of the ROC as follows: Received November 10, 2005; accepted November 16, 2005; published online February 1, 2006. Address correspondence and reprint requests to: Armando E. Giuliano, MD; E-mail: giulianoa@jwci.org.