An ARCAD database nomogram for prognostic prediction of treatment refractory metastatic colorectal cancer.

Abstract

107 Background: The "No Placebo Initiative" demonstrated that the survivals of patients (pts) receiving placebo in a salvage-line setting, exhibited no significant difference across trials (CORRECT, RECOURSE, CONCUR and TERRA). Therefore, we believe that a nomogram can be generated using the ARCAD database to predict survival of pts who completed standard therapy. Methods: Of 40,889 Individual patient data (IPD) from 59 studies, 723 pts received placebo from 4 randomized studies were pooled. The potential prognostic factors considered include BMI and Age. Imputed datasets were constructed by a multiple imputation method since the original dataset contains some missing values. Overall survival (OS) was defined as the time from randomization to death from any cause. Multivariate Cox proportional hazards models for OS were estimated based on the variables showing clinically and statistically significant by univariate analysis. Prediction performance of the final model was evaluated using the concordance index for survival data and calibration plots. Results: From univariate and multivariate analysis, ECOG performance status, Royal Marsden Hospital score (calculated by albumin level, LDH level, and number of metastatic site), primary site location, liver meta, peritoneal meta were selected as variables. We successfully developed a highly accurate nomogram to predict post-standard treatment prognosis for pts with mCRC. Concordance index for OS was 0.66. Conclusions: Although further validation in other cohorts is needed, this nomogram could be very useful to estimate the survival for clinical practice and enrolling in clinical trials. [Table: see text]