(UroToday.com) The 2025 American Society of Clinical Oncology (ASCO) Annual Meeting held in Chicago, IL, was host to the Poster Session: Genitourinary Cancer - Prostate, Testicular, and Penile Cancer. Dr. Chien-Kuang Cornelia Ding presented Poster 5106: External validation of a pathology-based multimodal artificial intelligence biomarker for predicting prostate cancer outcomes after prostatectomy.
Despite radical prostatectomy (RP) and salvage therapies, up to one-third of men with recurrent prostate cancer progress to metastatic disease, and approximately 5–10% ultimately die from the disease. This highlights the critical need for more accurate post-surgical risk stratification to guide management. A post-RP multimodal artificial intelligence (MMAI) model was previously developed using histopathology images and clinical data from phase III trials and validated in patients with biochemical recurrence following RP.1
The current validation study assesses the performance of the post-RP MMAI model in a real-world institutional cohort of prostatectomy patients, both before and after PSA recurrence. The objective is to determine whether this model can provide independent prognostic information that may support more personalized post-surgical treatment decisions.
Dr. Cornelia Ding and colleagues conducted a retrospective analysis of patients with localized prostate cancer who underwent radical prostatectomy at UCSF between 2000 and 2012. Eligible patients had available Hematoxylin & eosin-stained RP slides, corresponding clinical data, and long-term follow-up (n=1,230). A subset of this cohort (n=738) was selected for model validation using a random split stratified by CAPRA-S risk group. Of these, 640 patients had complete whole slide image and clinical datasets available for analysis.
Multimodal artificial intelligence (MMAI) scores were generated using the Artera RP MMAI model, which integrates histopathologic features from whole slide RP H&E images and clinical variables (Figure 1). CAPRA-S scores were calculated using standard methods as previously described by Dr Cooperberg.2
The association between the post-RP MMAI score and clinical outcomes was evaluated using Fine and Gray competing risk models along with cumulative incidence analysis. The primary endpoint was the development of any metastasis (MET). Secondary endpoints included bone metastasis (BMET), disease progression (DP)—defined as two consecutive PSA values ≥0.2 ng/mL or receipt of salvage therapy—and prostate cancer–specific mortality (PCSM). For MET, BMET, and DP, all-cause mortality was treated as a competing risk, while non–prostate cancer deaths were considered the competing risk for PCSM.
Subgroup analyses were also conducted to evaluate the model's prognostic performance in specific clinical contexts:
- Undetectable PSA subgroup: patients whose first post-RP PSA was <0.05 ng/mL at ≥6 months post-surgery.
- Salvage-eligible subgroup: patients with a detectable PSA (≥0.05 ng/mL) and/or those who received salvage therapy. For this group, MMAI scores were computed using recurrence or pre-salvage PSA data.
Baseline characteristics of the validation cohort, including both the undetectable PSA and salvage-eligible subgroups, are summarized below. The median age was similar across the full cohort and both subgroups. However, a notable difference was observed in the rate of positive surgical margins, which was higher in the salvage-eligible group (35%) compared to the undetectable PSA subgroup (18%). The median follow-up for the full validation cohort was 11.3 years, allowing for robust assessment of long-term clinical outcomes.
Notably, after adjusting for CAPRA-S, the post-RP MMAI score remained independently associated with adverse outcomes following surgery across clinical subgroups and all endpoints in univariate analysis (Figure A). These associations persisted in multivariate analysis, where MMAI continued to demonstrate prognostic significance for all endpoints when compared to CAPRA-S as a continuous variable. (Figure B).
Moreover, the post-RP MMAI model effectively identifies distinct risk groups (high vs. low) and enhances prognostic discrimination in salvage-eligible patients. Specifically, MMAI provides superior risk stratification for bone metastasis and prostate cancer–specific mortality compared to CAPRA-S, as shown in the cumulative incidence curves. Notably, while CAPRA-S failed to significantly differentiate outcomes in the salvage-eligible subgroup, MMAI maintained clear separation between risk groups with statistically significant differences in event rates at both 5 and 10 years.
Dr. Cornelia Ding concluded the presentation with the following key messages:
- This study represents the first external validation of a digital pathology-based post-RP MMAI model in patients following radical prostatectomy.
- The RP MMAI model effectively stratified risk for adverse outcomes, including metastasis, independent of CAPRA-S.
- These findings support the model’s potential utility in guiding personalized post-surgical management, particularly in informing early or intensified salvage treatment strategies.
- Compared to other advanced risk stratification tools, RP MMAI offers notable advantages in accessibility, efficiency, and cost.
Future directions for research include:
- Investigating the role of RP MMAI in post-surgical surveillance and treatment decision-making.
- Evaluating its risk stratification performance in prospective clinical trials.
- Assessing its applicability in contemporary cohorts imaged with modern diagnostic modalities.
Presented by: Chien-Kuang Cornelia Ding, MD, PhD, Assistant Professor, Pathology. School, School of Medicine University of California, San Francisco. San Francisco, CA.
Written by: Julian Chavarriaga, MD – Urologic Oncologist at Cancer Treatment and Research Center (CTIC) via Society of Urologic Oncology (SUO) Fellow at The University of Toronto. @chavarriagaj on Twitter during the American Society of Clinical Oncology (ASCO) 2025 Annual Meeting, Chicago, IL, Fri, May 30 – Tues, Jun 3, 2025.
Reference:
- Morgan TM, Ren Y, Tang S, Zwerink W, Chen E, Mitani A, et al. PD42-11 DEVELOPMENT AND VALIDATION OF A MULTIMODAL ARTIFICIAL INTELLIGENCE (MMAI)-DERIVED DIGITAL PATHOLOGY-BASED BIOMARKER PREDICTING METASTASIS FOR RADICAL PROSTATECTOMY PATIENTS WITH BIOCHEMICAL RECURRENCE IN NRG/RTOG TRIALS. Journal of Urology [Internet]. 2024 May 1 [cited 2025 May 26];211(5S):e897. Available from: https://doi.org/10.1097/01.JU.0001008560.54103.65.11
- Cooperberg MR, Hilton JF, Carroll PR. The CAPRA-S score: A straightforward tool for improved prediction of outcomes after radical prostatectomy. Cancer. 2011 Nov 15;117(22):5039-46. doi: 10.1002/cncr.26169. Epub 2011 Jun 3. PMID: 21647869; PMCID: PMC3170662.