Sas versions 9 1 and 9

The SLDs at baseline and the percentage change at nadir from baseline were summarized by independent review for pancreatic and metastatic lesions in each treatment arm using descriptive statistics. A nonparametric method, the Wilcoxon rank sum test, was used to test the treatment effect between arms. A multivariate analysis for OS was performed with a proportional hazards model (Cox regression model) for patients without missing values for any variable. The significance level required for entry into the model was 0.20 and for stay was 0.10. The model tested treatment group, geographic region (North America vs others), age, KPS, presence of liver metastases, and baseline pancreatic SLD. Baseline pancreatic SLD was a continuous variable representing the baseline sum of longest diameters of pancreatic lesions. Pearson correlation coefficients were calculated to measure the strength of the linear relationship between the percentage change of pancreatic versus metastatic SLDs from baseline at weeks 8, 16, and 24 for each treatment arm. All statistical tests were 2-sided and performed with SAS versions 9.1 and 9.2 (SAS Institute, Cary, NC). All P values were derived from Wald χ² tests. The HRs were computed using the exponential function applied to the regression estimate.