Artificial intelligence to predict outcome after [177Lu]Lu-PSMA for metastatic castration-resistant prostate cancer: Preliminary results from a multicentric prospective study.

Abstract

e17077 Background: [ 177 Lu]Lu-PSMA-617 was approved for metastatic castration-resistant prostate cancer, but only 46% of patients achieved a PSA response in the phase 3 VISION trial. Identifying patients unlikely to benefit from PSMA-targeted therapeutic radiopharmaceuticals is an urgent unmet need. Methods: Patients with mCRPC who progressed on taxane-based chemotherapy and ARSI, received baseline [ 68 Ga]Ga-PSMA-11 PET/CT, were VISION-eligible, and underwent [ 177 Lu]Lu-PSMA-617 or [ 177 Lu]Lu-PSMA-I&T were prospectively enrolled at 2 academic institutions. We report here the interim results of the University of Grenoble-Alpes cohort only (IRB: CEMEN 202406). Primary outcome was the prognostic value of an artificial intelligence-based technology (SelectPSMA, NucsAI) for PSA50 (≥50% decline from baseline), PSA progression-free survival (PSA-PFS; time to PSA progression by PCWG3), and overall survival (OS) after [ 177 Lu]Lu-PSMA radiopharmaceuticals. The Fisher's exact test and Kaplan-Meier analysis with log-rank test were used to test associations between SelectPSMA output and outcome data. Results: Of 72 mCRPC patients screened, 60 (83%) were enrolled between August 2023 and September 2024 and received [ 177 Lu]Lu-PSMA. 57/60 (95%) patients were treated previously with taxanes, while all 60 patients had received ARSIs. The median follow-up in survivors was 9.5 (IQR 7.1-15.1) mo and 47/60 (78%) patients achieved PSA progression at last follow-up. 38/60 (63%) patients achieved PSA50, the median (95%CI) PSA-PFS was 5.1 (3.3-7.1). 9/60 (15%) patients were classified by SelectPSMA as non-responders (PSMA-NR) and 51/60 (85%) as likely responders (PSMA-R). PSMA-NR was associated with significantly lower likelihood to achieve PSA50 (22% vs 71%; p=0.009) and shorter PSA-PFS (median (95%CI): 1.2 (0.7-NR) vs 6.6 (4.6-9.5) mo; p<0.001) compared to PSMA-R. At the time of this analysis, OS data was immature with low number of events in PSMA-NR group (n=4). Conclusions: Preliminary results show that SelectPSMA, an AI-based technology, identified patients with lower likelihood of PSA response and shorter progression-free survival after [ 177 Lu]Lu-PSMA radiopharmaceuticals. Results of multi-centric analysis including mature OS data will be presented at the conference. Clinical trial information: CEMEN 202406 .