Association of a post-prostatectomy digital pathology–based multimodal artificial intelligence (MMAI) biomarker with nodal and extrapelvic metastasis on PET/CT at the time of biochemical recurrence and event-free survival following salvage radiotherapy.

Abstract

378 Background: Clinical prostate cancer variables have limited ability to predict patterns of recurrence or outcomes after radical prostatectomy (RP). In this analysis, we evaluated the performance of a digital pathology multimodal artificial intelligence (MMAI) algorithm in predicting patterns of recurrence at the time of post-RP biochemical recurrence (BCR) and oncologic outcomes following PET-guided salvage radiotherapy (RT). Methods: The EMPIRE-1 (NCT01666808) and -2 (NCT03762759) randomized trials evaluated the association of PET-guided salvage RT on event-free survival (EFS) in men experiencing BCR following RP. A subset of patients with available digitized histopathologic images from the prostatectomy specimen was identified. A locked MMAI model (post-RP MMAI v1.1), which uses digital histopathology images and routine clinical data, was applied to generate MMAI scores (continuous and categorized [high- vs. low-risk]). The primary aim was to evaluate the association of MMAI score with 18 F-fluciclovine or 68 Ga-PSMA-11 PET uptake (defined as prostate bed [PB], pelvic lymph node [PLN], or extrapelvic [EP] uptake) at the time of BCR. The secondary aim was to examine the association of MMAI score with EFS (defined as freedom from biochemical/clinical/radiographic recurrence or initiation of systemic therapy) following salvage RT in a contemporary cohort managed with PET/CT and early salvage therapy. Results: Of the 223 patients enrolled on EMPIRE-1 or 2 who underwent PET scan, 92 had digital prostatectomy histopathology images available for MMAI analysis. Median (IQR) PSA at the time of PET/CT was 0.3 (0.2-0.7) ng/mL. MMAI score (continuous or categorized) was not associated with PB uptake on PET. MMAI score was significantly associated with PLN uptake on PET (continuous score per 1 SD increase: OR 2.68, 95% CI 1.57-4.98, P<.001; high vs. low group: OR 4.05, 95% CI 1.37-15.03, P=0.02). MMAI score was significantly associated with EP uptake on PET (continuous score: OR 4.63, 95% CI 1.92-14.47, P=0.003), with EP PET positivity only observed in MMAI high patients. These associations remained significant after pN-stage adjustment. After a median follow-up of 2.0 years, n=15 patients experienced an EFS event following post-RP salvage RT. Among these patients, 12 (80%) had an MMAI-high score. Post-RP MMAI score was significantly associated with EFS (HR 1.80 per SD increase, 95% CI 1.02-3.19, P=0.04). Conclusions: MMAI score from a post-RP model is associated with pelvic lymph node and extrapelvic PET uptake at the time of BCR, even with a low PSA at the time of PET/CT. This MMAI model was also prognostic for EFS following PET-guided early salvage RT.