Metastatic hormone-sensitive prostate cancer is no longer approached as a single clinical entity. With androgen-deprivation therapy, androgen receptor pathway inhibitors, docetaxel, and triplet regimens now available, one of the central questions is not only whether treatment should be intensified, but which patients are most likely to benefit.
A new study published in JCO Precision Oncology evaluated whether routine diagnostic pathology slides could help identify patients with metastatic hormone-sensitive prostate cancer who have a higher-risk histologic profile and may benefit more from upfront treatment intensification.
The article, titled “Development and Validation of a Computational Histology Artificial Intelligence–Powered Biomarker in Metastatic Hormone-Sensitive Prostate Cancer on Randomized Phase III Trials,” was published on July 8, 2026.
Authors: Christopher J. Sweeney, Vrishab Krishna, Viswesh Krishna, Akshay Neema, Asit Tarsode, Vinod V. Subhash, Lisa G. Horvath, James G. Kench, Martin R. Stockler, Sonia Yip, Hayley Thomas, Umang Swami, Haochen Zhang, Snehal Sonawane, Waleed M. Abuzeid, Vivek Nimgaonkar, Ekin Tiu, Drew Watson, Lesli Kiedrowski, Trevor J. Royce, David D. Yang, Phuoc T. Tran, Charles J. Ryan, Paul L. Nguyen, Alicia K. Morgans, Anirudh Joshi, Neeraj Agarwal, and Ian D. Davis.
Background
Treatment selection in metastatic hormone-sensitive prostate cancer still relies largely on clinical and pathologic features, including metastatic volume, timing of metastatic disease, PSA level, Gleason score, ECOG performance status, age, and comorbidities.
These factors are useful, but they do not precisely define which patients need triplet therapy and which patients may be adequately treated with a doublet approach. This is important because systemic treatment intensification can improve outcomes, but may also increase toxicity.
The study tested whether hematoxylin and eosin-stained diagnostic slides contain enough information to support risk stratification using histology alone, without clinical variables as model inputs.
Read more about AI in Oncology on OncoDaily.
Study Design
The investigators used the computational histology artificial intelligence platform CHAI to analyze digitized whole-slide images from diagnostic H&E-stained specimens.
The biomarker was developed using data from the phase 3 CHAARTED/E3805 trial and then independently validated in the phase 3 ENZAMET/ANZUP 1304 trial.
CHAARTED included patients with metastatic hormone-sensitive prostate cancer who were randomly assigned to ADT alone or ADT plus docetaxel. ENZAMET included patients with metastatic hormone-sensitive prostate cancer who were randomly assigned to testosterone suppression plus enzalutamide or a standard nonsteroidal antiandrogen. In ENZAMET, docetaxel was allowed based on physician and patient choice.
Overall, the analysis included 1,191 patients: 507 patients from CHAARTED in the development cohort and 684 patients from ENZAMET in the independent validation cohort.
The CHAI platform extracted quantitative histomorphologic features from the digitized slides, including features related to nuclear morphology, cell spatial organization, immune infiltration, and stromal density. The model was developed in CHAARTED, locked, and then tested in ENZAMET. No ENZAMET data were used to develop or retune the biomarker.
Results
In the ENZAMET validation cohort, the CHAI biomarker classified 559 patients as favorable risk and 125 patients as unfavorable risk. Patients classified as unfavorable risk had significantly worse outcomes than those classified as favorable risk.
For overall survival, the unfavorable-risk group had a hazard ratio of 2.6 compared with the favorable-risk group. The 5-year overall survival rate was 39% in the unfavorable-risk group and 69% in the favorable-risk group. For progression-free survival, the unfavorable-risk group had a hazard ratio of 2.2. The 5-year progression-free survival rate was 24% in the unfavorable-risk group and 49% in the favorable-risk group.
The biomarker remained independently prognostic after adjustment for established clinical factors, including treatment, age, ECOG performance status, Gleason score, PSA, metastatic volume, and timing of metastatic disease. In the adjusted analysis, the biomarker remained associated with both overall survival and progression-free survival.
Adjusted results showed an overall survival hazard ratio of 2.34 and a progression-free survival hazard ratio of 2.17 for the unfavorable-risk group.
Treatment Intensification
The study also explored whether the biomarker could identify patients more likely to benefit from treatment intensification. In ENZAMET, an exploratory analysis compared patients treated with ADT plus enzalutamide with those treated with ADT plus enzalutamide plus docetaxel.
Among patients with favorable-risk disease, triplet therapy was not associated with a clear improvement over doublet therapy. In contrast, among patients with unfavorable-risk disease, triplet therapy was associated with improved outcomes.
Overall survival favored triplet therapy in the unfavorable-risk group, with a hazard ratio of 0.45. Progression-free survival also favored triplet therapy, with a hazard ratio of 0.42.
The interaction between biomarker status and treatment intensification was significant for both overall survival and progression-free survival, suggesting that the unfavorable-risk group may be more likely to benefit from the addition of docetaxel.
Read more about STAMPEDE Updates on OncoDaily.
Limitations
The treatment-intensification analysis should be interpreted carefully. Docetaxel use in ENZAMET was not randomized, but selected by physicians and patients. Therefore, the triplet therapy finding remains exploratory and needs confirmation in randomized cohorts.
The analysis also included only patients with available digitized H&E slides and clinical outcome data, not the entire randomized populations of CHAARTED and ENZAMET.
Another limitation is generalizability. Black patients were underrepresented in the ENZAMET validation cohort, which may limit how broadly the findings can be applied across diverse patient populations.
The study was supported by Valar Labs, Inc.
Conclusion
This study developed and validated a histology-only AI biomarker for metastatic hormone-sensitive prostate cancer using data from two phase 3 randomized trials.
The CHAI biomarker identified a subgroup of patients with substantially worse overall survival and progression-free survival, independent of conventional clinical risk factors. In exploratory analysis, patients with unfavorable-risk disease appeared to derive greater benefit from triplet therapy with ADT, enzalutamide, and docetaxel.
These findings support further validation of AI-based pathology biomarkers as tools for risk stratification and treatment selection in metastatic prostate cancer.
The full article is available in JCO Precision Oncology.