Results of PD-L1 Analysis of Women Treated with Durvalumab in Advanced Endometrial Carcinoma (PHAEDRA)

Simple Summary Until recently the outcome for women with advanced endometrial carcinoma has been poor. New immune therapies have resulted in much better outcomes for some, but not all, women with advanced disease. Determining which women are likely to respond is important, both to identify potential responders and prevent overtreatment of women who are not likely to respond. In this study we looked to see if there were any other markers that could be used to better predict those tumors that would be more likely to respond to the immune therapy known as durvalumab versus those tumors more likely to be resistant. Using statistical methods to evaluate each of the components examined we determined the cut-off point with the best performance. We found that the presence of tumor associated inflammatory cells had the strongest association with response. We also found that an algorithm derived from our best performing cut-off points identified women not likely to respond to treatment. While the presence of inflammatory cells was not significant when mismatch repair status was considered, our novel algorithm was. This is a small study and the findings do require validation in a larger group of women. Women with advanced endometrial carcinoma (EC) with mismatch repair (MMR) deficiency have improved outcomes when treated with immune checkpoint inhibitors; however, additional biomarkers are needed to identify women most likely to respond. Scores for programmed death ligand 1 (PD-L1), immunohistochemical staining of tumor (TC+), immune cells (IC+) and presence of tumor-associated immune cells (ICP) on MMR deficient (n = 34) and proficient (n = 33) EC from women treated with durvalumab in the PHAEDRA trial (ANZGOG1601/CTC0144) (trial registration number ACTRN12617000106336, prospectively registered 19 January 2017) are reported and correlated with outcome. Receiver operating characteristic (ROC) analyses and area under the ROC curve were used to determine optimal cutpoints. Performance was compared with median cutpoints and two algorithms; a novel algorithm derived from optimal cutpoints (TC+ >= 1 or ICP >= 10 or IC+ >= 35) and the Ventana urothelial carcinoma (UC) algorithm (either TC+ >= 25, ICP > 1 and IC+ >= 25 or ICP = 1 and IC+ = 100). The cutpoint ICP >= 10 had highest sensitivity (53%) and specificity (82%), being prognostic for progression-free survival (PFS) (p = 0.01), while the optimal cutpoints algorithm was associated with overall survival (p = 0.02); these results were not significant after adjusting for MMR status. The optimal cutpoints algorithm identified non-responders (p = 0.02) with high sensitivity (88%) and negative predictive value (92%), remaining significant after adjustment for MMR. Although MMR status had the strongest association with response, further work to determine the significance of ICP >= 10 and the novel optimal cutpoint algorithm is needed.