O-0013 Improved Early Prediction of Individual Prognosis for Patients with Metastatic Colorectal Cancer: Joint Modeling of Tumor Shrinkage with Volume Data

Abstract

ABSTRACT Introduction Piessevaux and colleagues proposed a decrease in Response Evaluation Criteria In Solid Tumors (RECIST)-based tumor size of 20% at week 8 following first-line therapy for metastatic colorectal cancer (mCRC) as a predictor of clinical outcome (progression-free survival [PFS] and overall survival [OS]). We expanded upon this idea (ASCO GI 2012) by transforming it into a strategy for individual prognosis. Further, a tumor volume algorithm was developed providing a better approximation to true tumor volume by using both the longest diameter and the longest orthogonal diameter of a target lesion. In this study, we compare the quality of individual patient prediction based on early tumor change according to either RECIST or the tumor volume algorithm. Methods The prognostic method is combined with the volume algorithm and applied to the data from two studies (OPUS, n = 337; CRYSTAL, n = 1198) and four treatment regimens (FOLFOX4 +/- cetuximab and FOLFIRI +/- cetuximab) in patients with mCRC. The influence of the treatment regimen on early PFS and OS prognosis is studied by joint modeling. The quality of early prognosis depending on the tumor size assessment used is compared by the logarithmic (log) scoring rule with respect to calibration and precision. Results Individual volume shrinkage and baseline volume are considered prognostic factors. Individual predictions depend on the chemotherapy administered and on whether cetuximab is added. Equivalent tumor baseline volumes and early volume changes predict an up to 20% higher 1-year PFS and 2-year OS rate for FOLFIRI than for FOLFOX. The addition of cetuximab to standard chemotherapy translates into a median increase in early shrinkage from 15% to 23% in CRYSTAL and from 19% to 30% in OPUS. This translates into an improvement in the 1-year PFS rate from 10% to 40% and in the 2-year OS rate from 40% to 60% in CRYSTAL patients and an improvement in the 1-year PFS rate from 7% to 28% and in the 2-year OS rate from 30% to 57% for OPUS patients. The quality of individual prediction with respect to calibration and precision over the four regimens is combined in one log score. The log score for the RECIST-based prediction is 2.45, which is significantly higher than that of the volume-based prediction of 1.97 (p Conclusion The tumor volume algorithm enables a better calibrated and more precise prediction of individual patient PFS and OS than RECIST-based tumor assessments. Based on early tumor changes for chemotherapy, +/- cetuximab, it is possible to calculate quantitative information on a patient's prognosis. The model has high potential to guide individual clinical decision making for patients with mCRC.