Evaluation of Prognostic and Predictive Value of PETSCAn in ColoRectal Cancer (CRC)

statistical hypothesis : Sample size calculation for FDG-PET in metastatic patients.

In the paper of Cascini et al. [1], they have 18 patients with a TRG response (TRG = Tumor Regression Grade) and 15 patients with no response.

SUV mean was measured at baseline and at day 12 in all 33 patients. All 33 patients showed a reduction in SUV-mean from baseline to day 12. The median reduction in SUV-mean was 63% (mean 66%) in responding tumors and 22% (mean 22%) for non responding tumors. Using a cutoff level of 52% a perfect accuracy could be obtained, i.e. all responding tumors had a SUV-mean reduction higher than 52% and all non responding tumors had a SUV-mean reduction lower than 52%.

(Standard deviation of the SUV-mean reduction was 25.) All 33 patients showed also a reduction in SUV-max from baseline to day 12. The median reduction in SUV-max was 62% in responding tumors and 28% in non responding tumors. (No information about mean or standard deviation of SUV-max available in [1].) Based on the data of SUV-mean we can do a power calculation based on the Wilcoxon test: In order to demonstrate an absolute difference in the mean of ΔSUV-mean of 44% between responding and non responding patients, with an estimated standard deviation of 25, using 35 patients (50% response, 50% no response), at a significance level of 5%, we obtain a power of 0.98.

For SUV-max, if we assume that the difference in the means is similar to the difference in the medians, i.e. 34% and the standard deviation is similar to the one of SUV-mean, i.e. 25, we obtain a power of 0.93 If we, to be safe, are a little bit less optimistic and estimate that the difference in the mean of ΔSUV-max between responding and non responding tumors would be a little lower than in [1]: 35%, and the standard deviation a little bit higher: 30, then we need 40 patients at a power of 0.90