Citations

Purpose Factors contributing to early mortality after initiation of treatment of metastatic colorectal cancer are poorly understood. Materials and Methods Data from 22,654 patients enrolled in 28 randomized phase III trials contained in the ARCAD (Aide et Recherche en Cancérologie Digestive) database were pooled. Multivariable logistic regression models for 30-, 60-, and 90-day mortality were constructed, including clinically and statistically significant patient and disease factors and interaction terms. A calculator (nomogram) for 90-day mortality was developed and validated internally using bootstrapping methods and externally using a 10% random holdout sample from each trial. The impact of early progression on the likelihood of survival to 90 days was examined with time-dependent Cox proportional hazards models. Results Mortality rates were 1.4% at 30 days, 3.4% at 60 days, and 5.5% at 90 days. Among baseline factors, advanced age, lower body mass index, poorer performance status, increased number of metastatic sites, BRAF mutant status, and several laboratory parameters were associated with increased likelihood of early mortality. A multivariable model for 90-day mortality showed strong internal discrimination (C-index, 0.77) and good calibration across risk groups as well as accurate predictions in the external validation set, both overall and within patient subgroups. Conclusion A validated clinical nomogram has been developed to quantify the risk of early death for individual patients during initial treatment of metastatic colorectal cancer. This tool may be used for patient eligibility assessment or risk stratification in future clinical trials and to identify patients requiring more or less aggressive therapy and additional supportive measures during and after treatment.

666 Background: Early tumor shrinkage (ETS; ≥20% decrease from baseline) and early objective tumor response (EOTR; CR/PR by RECIST) are associated with improved PFS/OS to a similar extent than best overall response (BOR), confirmed response (ConfR) and non-progression rate (NPR; CR/PR/SD by RECIST) at the individual level in first-line mCRC (ASCO 2014, abstr 3538/3578). We assessed these RBEs at the trial level. Methods: Data were available on 12,185 pts enrolled in a total of 17 randomized trials of chemotherapy (CT) alone or with targeted agents (TAs). ETS, EOTR and NPR were assessed at 6/8/9 and 12 weeks (wks), whereas BOR and ConfR were ascertained within 26 wks from randomization. Association between treatment effects on RBEs and on PFS/OS was assessed by coefficients of determination (R2, adjusted for estimation errors) using a copula model). Results: None of the RBEs had consistently strong correlations with PFS or OS (Table). The number of trials was insufficient to obtain reliable estimates for antiangiogenics and anti-EGFR separately. Conclusions: Despite their individual-level associations with PFS and OS, RBEs do not appear to be surrogate endpoints in first-line mCRC, given their poor predictive ability. [Table: see text]

659 Background: Prospective survival prediction of patients with metastatic colorectal cancer is difficult. Prognosis estimation based on readily available clinicopathologic factors has the potential to inform clinical practice and improve risk stratification for clinical trials. We constructed prognostic nomograms for OS and PFS in mCRC using the multi-trial ARCAD database. Methods: Data from 19,678 mCRC pts accrued to 24 first line randomized phase III clinical trials since 1997 were used to construct and validate Cox models for PFS and OS, stratified by treatment arm within each study. Candidate variables included age, gender, BMI, performance status, colon vs. rectal cancer, prior chemotherapy, number of metastatic sites, sites of metastases (liver, lung, lymph nodes), and baseline bilirubin, albumin, white blood cell count, hemoglobin, platelets, absolute neutrophil count, and derived neutrophil:lymphocyte ratio (dNLR). Missing data (<11%) were imputed, continuous variables modeled with splines, and clinically relevant pairwise interactions considered if p<0.001. Final models were internally validated via bootstrapping to obtain optimism-corrected calibration and discrimination C-indices, and externally validated using a 10% holdout sample from each trial. Results: Nomograms for OS and PFS including remaining variables were well calibrated with C-indices of 0.66 and 0.60, respectively. Evaluation of external validity revealed good concordance; 71% and 67% respectively between predicted (> vs. <50% probability) and actual (yes/no) 1-year OS and 6-month PFS, and median 1-year OS and 6-month PFS predictions fell within the actual 95% Kaplan-Meier intervals. Gender, liver and lung metastases, and dNLR were not prognostic for OS; prior chemo, colon vs. rectum, dNLR, liver and lymph node metastases, and gender did not predict for PFS. No clinically relevant pairwise interactions were identified. Conclusions: The proposed nomograms are well calibrated and internally and externally valid. These tools have the potential to aid prognostication and patient/physician communication, and balance risk in randomized trials in mCRC.