Statistical Analysis
Descriptive statistics were calculated, and bivariate testing methods included t and chi-squared tests. Overall survival (OS) was calculated as time from diagnosis to either date of death due to any cause or censoring at 5 years post-diagnosis. Subjects who were still alive on December 31st, 2013 were censored on that date. Hazard ratios (HRs) and 95% confidence intervals (CI) for the independent effects of race on overall survival were estimated by Cox proportional hazards regression models. Sequential adjustment sets were used to examine the relative contribution of demographic, clinical and SES variables to the racial survival disparity. Variables that were missing >10% of observations were left out of the survival analysis, which was determined as an a priori threshold. These included rural/urban household designation (n=44 missing) and driving distance to clinic where biopsy was performed (n=49 missing). All variables were treated as categorical variables for the survival models. The proportional hazards assumption was met for all covariates and there was no evidence of multicollinearity on variance inflation factor testing. STATA 16 (StataCorp, College Station, TX) was used for all analyses.