Validation study on magnitude of tumorshrinkage as a prognostic marker

Statistical plan: The primary endpoint of the study will be OS, and the secondary endpoint will be PFS as defined in the studies. The primary prognostic factor of interest will be depth of remission (tumour shrinkage), defined using the sum of longest diameter (SLD) of target lesions on imaging as per the RECIST v1.0 guidelines, as per study protocols. As with the prior analysis, patients will be categorized by maximal tumor shrinkage from baseline (-100% to <-60%, =60% to <-30%, =-30% to <0%, =0 to <+20%, =+20%) and calculated as (nadir on study SLD - baseline SLD) / baseline SLD * 100%. Patients without post-baseline scans will be grouped and analyzed separately.Estimates of OS and PFS will be calculated using the Kaplan-Meier method and 95% confidence intervals (CI) reported. Log-rank and univariate Cox regression analyses will be used to investigate prognostic ability of depth of remission, and other factors, with OS or PFS. A multivariate Cox proportional hazards model will be used to evaluate the effect of depth of remission, adjusted for sex, race, treatment, and Memorial Sloan-Kettering Cancer Center (MSKCC) risk factors. An interaction term will be evaluated to determine whether any effect due to depth of remission is constant across treatments. To correct for the bias inherent in the analysis of time-to-event outcome a 6-month landmark analysis will be performed, hence, any patient who died or had disease progression prior to 6-months will be excluded from the analyses with OS and PFS outcomes respectively. All tests will be two-sided and statistical significance will be defined at the a=0.05 level.Statistical Power: Response rates, OS, and PFS are similar between the COMPARZ study, the Seidel et al study and in the Grünwald analysis (Grünwald et al. ESMO 2013 #2702). Hence, it will be assumed for statistical power calculations that the COMPARZ study will have similar results to what was observed by Grünwald et al. In the Grünwald et al study, 10%, 20%, 42%, 14%, 6% and 8% patients had tumour shrinkage in each depth of remission category, with observed hazard ratios of 0.27, 0.70, REFERENCE, 1.62, 1.92 and 4.37 in the multivariable regression model respectively. Applying these groups to the COMPARZ study, and excluding the ~200 patients who have no survival data beyond 6 months, it is expected that there will be approximately 90, 180, 378, 126, 54 and 72 patients per group. Assuming similar results as in the Grünwald study and with follow-up and accrual as described in the COMPARZ study, it is expected that there would be >96% power to detect a difference between the =60% to <-30% subgroup and the reference group (=-30% to <0%), or alternatively between the =0 to <+20% and the reference group (calculated using NCSS-PASS, v2005). Given that the multivariable Cox model will evaluate depth of remission across all subgroups simultaneously, there should be sufficient statistical power to detect differences between categories even after adjusting for the other covariates which will be included in the model.