Propensity Matching Survival Analysis

A propensity score matching (PSM) method (nearest-neighbor matching without replacement, using a logistic propensity model and a caliper of 0.2) was used to control for age (in terciles), sex, pathologic stage group (0, 1, 2, 3, or 4), and tumor size (discretized into 4 levels according to Eighth Edition AJCC). Clinical stage group was also viewed as a possible confounder, but was not selected due to moderate rank correlation with pathologic stage group. More complex propensity models, which would have allowed controlling for more variables, were not supported. The log-rank test was used to compare the entire survival experiences of the samples resulting from the PSM procedure. With the same PSM samples, a bootstrap method (the ordinary method, with 10,000 replicates) was used with Kaplan–Meier estimators at 5 years to estimate ratios of 5-year survival probabilities and calculate 95% confidence intervals (CIs).