Prism 6.0d for macintosh

The statistical software used was SAS 9.3 (SAS Institute Inc., Cary, NC, USA) and Prism 6.0d for Macintosh (GraphPad Software, San Diego CA, USA; www.graphpad.com). Hazard ratios (HR), 95% confidence intervals (CI) and p values were calculated from univariate and multivariable Cox proportional hazards models. The proportional hazards assumption was found to be upheld for each univariate and multivariable regression. Initially each confounder that had a significant HR in univariate analysis (p < 0.1) was included in the multivariable model with the predictor being AR nuclear localization or FKBP5 expression or AR nuclear localization and FKBP5. However, there were too few observations to account for the ten covariates. Therefore, backwards stepwise elimination was performed. The confounder with the highest p value was eliminated, one at a time, until the final most parsimonious model had all confounders with p < 0.05 or p < 0.2 depending on the model. Normalized fold expression data were compared using unpaired t-test. Correlations between androgenresponsive genes in esophageal tissues were determined using linear regression. All statistics were considered significant when the two-tailed p < 0.05.