Use of Receiver Operating Characteristic (ROC) Analysis To Determine the Most Optimal Prognostic Threshold for Adjuvant! in Lymph Node Negative Breast Cancer Patients < 55 Years.

Abstract Background: Adjuvant!1 is a web based tool to predict the 10-year relapse free survival (RFS), breast cancer specific survival (BCSS) and overall survival (OS) probabilities using age, tumor size (pT), lymph node status (pN), histological grade (Grade 1-3), Estrogen receptor (ER) status in the estimation. A challenge is to find optimal cut-off values of the Adhuvant probabilities to compare the predictions with other prognostic factors.Patients and methods: In the Dutch nationwide Multicenter Morphometric Mamma Carcinoma Project 3472 patients have been enroled2, of which 516 <55 years with T1-3N0M0 underwent surgery and local radiation therapy when indicated. Adjuvant systemic treatment was not given according to the Dutch guidelines at that time. Median follow-up is 10 years. The 10-year RFS, BCSS and OS probabilities were evaluated with Adjuvant! Since these estimated predicted probabilities are continuous, optimal thresholds can be evaluated by the Receiver Operating Characteristic (ROC) analyses. Concordances between the Adjuvant predicted outcomes and, classical prognosticators were evaluated by the Spearman-correlation test.Results: The optimal cut-points for Adjuvant-RFS, Adjuvant-BCSS and Adjuvant-OS were 62% (Area Under the Curve=AUC==0.603, p=0.0004), 86% (AUC=0.635, p=0.0001) and 72% (AUC=0.598, p=0.0002) respectively. Using these thresholds, a moderate correlation was found between all the Adjuvant! endpoints and pT, Grade 1-3 and ER status (r=0.34 – 0.63). In the univariate survival analysis (Kaplan Meier) the above Adjuvant thresholds were highly prognostic. For Adjuvant-RFS >63/≤63 the 14-year survival was 77% and 64% respectively (p=0.001; log-rank). Moreover, for Adjuvant-BCSS (≥86/<86) the corresponding numbers were 82% and 67% ( P < 0.0001; log-rank). For Adjuvant-OS (≥72/<72) the survivals were 77% and 64% respectively (p<0.0001; log-rank). With multivariate analyses, Adjuvant! variables were part of the model only for BCSS and OS. However, Grade1-3 was prognostically stronger than Adjuvant! regarding both BCSS and OS. Furthermore, when the three features that constitute grade (tubular formation, nuclear atypia and mitosis count), only the latter was included in the final model.Discussion: ROC analysis can obtain optimal prognostic threshold values from web-based prognostic tools using time dependent endpoints3. A high grade of dependency between Adjuvant! and its’ underlying variables must be taken into account when the multivariate model is put up. Mitotic activity in the tumor seems to overrule Adjuvant!. This should be further explored.1 www.adjuvantonline.com2 Baak JPA, et al.: Pathol Res Pract 185:664-70, 19893 Zlobec I, et al.:J Clin Pathol 60:1112-1116, 2007. Citation Information: Cancer Res 2009;69(24 Suppl):Abstract nr 3178.