Statistical Analysis

The exploratory primary endpoint was OS of all patients stratified by exposure to REG and/or TFTD as follows: cohort A (both REG and TFTD) and cohort B (either REG or TFTD). The following pretreatment clinical data and baseline laboratory values were used in the analysis as covariates: age, sex, body mass index, ECOG PS, primary tumor site, surgery on primary tumor, RAS status, metastatic tumor site (liver metastasis, lung metastasis, lymph node metastasis, peritoneal dissemination, and bone metastasis), number of metastatic sites, pathologic type, time from initiation of first-line chemotherapy to initiation of later-line treatment, serum albumin, serum aspartate transaminase (AST), serum C-reactive protein (CRP), and serum carcinoembryonic antigen (CEA). Each cutoff value of quantitative data was set with reference to that of albumin, AST, CRP, AST, and CEA in the REGOTAS study (13).

OS was defined as the time from the start of initial REG or TFTD to death or last follow-up. Quantitative data are expressed as median and interquartile range (IQR). The Mann–Whitney U-test was used to compare the continuous variables, and Fisher's exact test was performed to compare the categorical variables. Survival curves were estimated using the Kaplan–Meier method, and differences between the groups were tested by the log-rank test. Hazard ratios (HRs) were estimated using the Cox proportional hazard model. OS was analyzed using univariate and multivariate Cox regression analyses. The backward selection method was conducted for the selection of factors retained (P < 0.2) in the multivariate analysis. The predictive factor for OS between each group was explored using subgroup analyses with the multivariate Cox model including interaction terms.

A 1:1 matching using the propensity score (propensity score-matched dataset) was performed as a sensitivity analysis. Patients in the two groups were matched by a difference of propensity score within 0.05. The propensity score was calculated with a multivariate logistic regression model including 19 prognostic variables (Supplementary Table 1). All P-values < 0.05 were considered statistically significant. All statistical analyses were performed using EZR (Saitama Medical Center, Jichi Medical University, Saitama, Japan), which is a graphical user interface for R (The R Foundation for Statistical Computing, Vienna, Austria).