ESTRO 2024 - Abstract Book

S2248

Clinical - Upper GI

ESTRO 2024

data of 25 patients were available at the time of the analysis (adenocarcinoma: 12, squamous cell carcinoma: 13): 3/25 with complete response at imaging refused surgery and 2/3 were cCR. Of the 22/25 patients submitted to immediate surgery, one was not radically operated due to evidence of peritoneal carcinomatosis and evidence of macroscopic residual; of the 21 patients with radical surgery, the responses according to the Mandard score were: TRG1: 8; TRG2: 3; TRG3: 5; TRG4: 4; TRG5: 1. Then, for the considered end-point (complete response: pCR+cCR), the events were 10/25 (40%). Significant tumor regression was seen in most patients with only 4/25 patients not showing regression at tpost. Median volumes at tmid and tpost were 63% and 24% of the volumes at tpre; regression at tmid was significantly larger for complete responding patients compared to the others (44% (IQR: 38-63) vs 79% (56-93) of Vpre, p=0.03). An example of volume regression for a patients experiencing pCR is shown in Figure 1. Both ERImid and ERIpost identified pCR/pCR+cCR patients, with ERImid showing better performance (AUC: 0.78, 95%CI: 0.54-0.94, p=0.014) compared to both ERIpost and % volume regression: the rate of pCR was 70% vs 12% for ERImid values ≤ or >26 (best cut -off value, p=0.017). A two-variable logistic model combining ERImid and Vpre improved performances (AUC: 0.93, 95%CI: 0.72-1.00, p<0.0001). In Figure 2, the ROC curves discriminating complete response for ERImid alone and in combination with Vpre are shown. Inter-observer variability in contouring GTV was found to be small for all three time points (median DICE: 0.91-0.94) and did not affect the results.

Figure 1: Example of tumor regression for a patient experiencing pCR