AI-Based Pathology Image Classifier May Elucidate AR Benefit in mHSPC

An artificial intelligence (AI)-based pathology image classifier (APIC) may help elucidate androgen receptor (AR) therapy benefit with enzalutamide (Xtandi) among patients with metastatic hormone-sensitive prostate cancer (HSPC), according to an AI biomarker study analyzing hematoxylin and eosin (H&E) specimens from the phase 3 ENZAMET/ANZUP 1304 trial (NCT02446405) presented at the 2026 American Society of Clinical Oncology (ASCO) Annual Meeting.

Specifically, a negative APIC scan was found to be predictive of enzalutamide benefit in the overall population. Among 64% of patients for whom the scan was negative, an 18.2% difference in 5-year overall survival (OS) was observed (HR, 0.50; 95% CI, 0.35-0.73). By contrast, APIC positivity was associated with slightly worse outcomes with enzalutamide, with the remaining 36% experiencing a –3.2% difference in 5-year OS (HR< 1.04; 95% CI, 0.67-1.62).

When excluding docetaxel, the role of APIC negativity was slightly more profound. Among 63% of patients identified as APIC-negative who did not receive docetaxel, the difference in 5-year OS was 21% with enzalutamide vs conventional nonsteroidal antiandrogen (NSAA; HR, 0.40; 95% CI, 0.23-0.69). APIC positivity was associated with a 5.2% difference in 5-year OS favoring enzalutamide vs NSAA (HR, 0.79; 0.43-1.46), ruling out docetaxel as an explanation for the APIC-negative benefit.

Regarding treatment-biomarker interaction, the most pronounced benefit with enzalutamide was for patients who had APIC negativity and low volume disease (HR, 0.31; 0.16-0.62; P <.001). Additionally, a multivariable analysis found that the biomarker/treatment interaction held significance after controlling for non-prognostic variables (P = .016). Finally, APIC score was found to be inversely associated with immune infiltration ( = –0.69; P < .001).

“APIC is an interpretable AI biomarker computable from routine H&E,” Sebastian R. Medina, graduate student in the Department of Radiology and Imaging Sciences at Emory University, stated in the presentation. “APIC [negativity] is predictive of [OS] benefit from androgen deprivation therapy [ADT] plus enzalutamide, and it identifies [patients] who may not require treatment intensification. APIC [positivity] defines a poor prognostic group who may benefit from the addition of non-AR directed therapy.”

In the phase 3 protocol, 1125 patients with metastatic HSPC were randomly assigned 1:1 to receive ADT plus enzalutamide (n = 563) or conventional NSAA (n = 562). Patients received docetaxel at physician discretion, and 492 patients had digitized H&E data and were included in the follow-up.

In the biomarker cohort, the proportion of patients younger than 70 in the standard care (n = 248) and enzalutamide arms (n = 244) was 58.9% vs 57.0%. In the respective arms, most patients were from Australia or New Zealand (59.7% vs 65.2%), did not have early planned docetaxel (56.9% vs 54.1%), and had low volume disease (49.2% vs 52.5%). Additionally, 85.9% vs 89.3% of each group had visceral metastases, 60.1% vs 56.1% had synchronous metastases, and 68.5% vs 68.4% had a Gleason score between 8 and 10.

Of note, no significant differences were observed between the trial overall and biomarker cohorts, and those receiving triplet therapy were more likely to have high-volume, synchronous disease. The main study end point was OS.

The APIC was composed of 6 spatial and morphology features designed to quantify immune organization and tumor heterogeneity. In the CHAARTED protocol, APIC was predictive of docetaxel benefit independent of volume and timing of metastases.

One limitation of the study included the biomarker cohort being representative of 43.7% of the initial ENZAMET population based on tissue availability. Additionally, the investigators noted that APIC was developed in the CHAARTED protocol where ARPIs were not the standard of care. Moreover, a lack of ethnic diversity across both trial protocols and a lack of docetaxel randomization were identified as additional limitations.

Reference

Medina SR, Tokuyama N, Putcha V, et al. An AI-based pathology classifier to predict benefit from enzalutamide in metastatic hormone-sensitive prostate cancer (mHSPC) from ENZAMET (ANZUP 1304). J Clin Oncol. 2026;44(suppl 16):108. doi:10.1200/JCO.2026.44.16_suppl.108