ESMO Congress 2026

KEYNOTE-365 is a phase 1b/2 multi-cohort platform study of pembrolizumab in metastatic castration-resistant prostate cancer (mCRPC), with prior cohorts published successively: cohort A combined pembrolizumab with olaparib in molecularly unselected, docetaxel-pretreated patients ^[2]; cohort B combined it with docetaxel and prednisone in chemotherapy-naïve patients after failure of or intolerance to abiraterone or enzalutamide ^[1]; cohort C combined it with enzalutamide in abiraterone-pretreated patients ^[3]; cohort D combined it with abiraterone in chemotherapy-naïve patients ^[4]; cohort E combined it with lenvatinib ^[5] and cohort G used vibostolimab coformulated with pembrolizumab ^[6], both in docetaxel-pretreated patients. These cohorts all had safety, PSA response rate, and ORR as primary endpoints, systematically exploring multiple combination directions against the backdrop of pembrolizumab monotherapy showing only manageable safety and modest antitumor activity in mCRPC ^[5]; for example, cohort G reported a confirmed PSA response rate of 18% and an ORR of only 6.9% in the RECIST-measurable population among 40 treated patients ^[6].

The ESMO26 update covers Cohort I, previously absent from the published literature (abstract 2217RO; the live posts mention Gunhild von Amsberg): pembrolizumab plus carboplatin/etoposide versus chemotherapy alone in neuroendocrine prostate cancer (NEPC), with a total of 70 patients (40 in the combination arm, 30 in the chemotherapy arm). Key numbers: median OS 10.2 vs 7.1 months, HR 0.33 (95% CI 0.17-0.63); rPFS 5.1 vs 4.0 months, HR 0.66 (95% CI 0.36-1.21); ORR 28% vs 15%, DCR 50% vs 48%, median DOR 6.2 vs 4.1 months; in the PD-L1 CPS ≥1 subgroup, median OS was 11.0 vs 4.8 months, HR 0.16. No new safety signals were observed.

Unlike the previously published cohorts, which had safety and response rates as primary endpoints and no survival readout against a chemotherapy comparator ^{[1][2][3][4][5][6]}, this dataset extends the immune-combination exploration from mCRPC at large to the specific histologic population of NEPC and provides OS and rPFS results versus chemotherapy alone: ORR rose from 15% to 28% with longer median DOR (6.2 vs 4.1 months) while DCR was similar between arms (50% vs 48%), and the OS difference was particularly pronounced in PD-L1 CPS ≥1 patients (HR 0.16), pointing to a potential direction for biomarker enrichment.

As summarized in the on-site commentary, the limitations are the small sample size (N=70), the post hoc nature of the PD-L1 subgroup analysis, and the absence of formal hypothesis testing; the signal is thus promising but not yet practice-changing, and information on confirmatory trials was not disclosed.