AI-Based Pathology classifier Predicts Sensitivity to Enzalutamide in Metastatic Hormone-Sensitive Prostate Cancer: A Biomarker Analysis of the ENZAMET Trial
Sebastian Medina, Naoto Tokuyama, Vaishnavi Putcha, Tilak Pathak, Pingfu Fu, Hayley Thomas, Vinod V. Subhash, Sonia Yip, Hui-Ming Lin, Lisa G. Horvath, James G. Kench, Alison Zhang, Martin R. Stockler, Anthony M. Joshua, Arun Azad, Samantha R. Oakes, Ian D. Davis, Christopher J. Sweeney, Anant MadabhushiAbstract
Purpose: The ENZAMET trial established that enzalutamide added to androgen deprivation therapy (ADT) improves overall survival (OS) in metastatic hormone-sensitive prostate cancer (mHSPC), but treatment response is heterogeneous. We tested whether the Artificial Intelligence Pathology Image Classifier (APIC), previously developed and locked in CHAARTED, predicts overall survival benefit from enzalutamide in ENZAMET. Experimental Design: This biomarker study applied APIC to digitized hematoxylin and eosin specimens from ENZAMET (ANZUP 1304, NCT02446405). ENZAMET randomized participants with mHSPC to ADT plus enzalutamide or non-steroidal antiandrogen (NSAA), with early concurrent docetaxel permitted. APIC was applied without modification or retraining. The prespecified primary analysis evaluated the treatment-by-APIC interaction for OS using Cox models adjusted for standard clinical factors. Prespecified sensitivity analysis excluded participants receiving early concurrent docetaxel. Results: Among 1125 participants, 492 with digitized tissue (median follow-up 68.8 months) were eligible. APIC classified 316 (64.0%) as APIC-negative and 176 (36.0%) as APIC-positive. APIC significantly modified enzalutamide benefit (interaction p=0.014). APIC-negative disease showed improved survival with enzalutamide versus NSAA (HR=0.50, 95%CI:0.35–0.73; 60-month OS 76.5% vs 58.3%), while APIC-positive showed limited evidence of benefit (HR=1.04, 95%CI:0.67–1.62; 60-month OS 55.6% vs 58.8%). In low-volume disease, enzalutamide improved survival in APIC-negative (HR=0.31, 95%CI:0.16–0.62) but not APIC-positive (HR=1.04, 95%CI:0.53–2.06; interaction p=0.015). For non-docetaxel-treated participants, enzalutamide benefit was greater in APIC-negative (HR=0.40; 60-month OS 82.0% vs 61.0%) than APIC-positive (HR=0.79; 60-month OS 62.8% vs 57.6%). Conclusions: The presence of APIC negative disease identified participants with a major improvement in OS with addition of enzalutamide to ADT compared with addition of NSAA.