3121 - Evaluating the Correlation between Genomic Classifier and Digital Pathology-Based Multi-Modal AI Biomarkers in Localized Prostate Cancer
Presenter(s)
A. A. Olabumuyi1, O. Jordan2, J. H. Chang2, C. Datnow-Martinez1, S. E. Delacroix Jr3, D. E. Spratt4, X. Shi5, Y. Liu6, E. Davicioni7, H. C. Huang8, M. Tierney8, M. Kim9, S. P. Kanani9, Z. H. Rana10,11, J. K. Molitoris1, M. V. Mishra10, Y. Kwok1, P. Sutera12, M. P. Deek13, and P. T. Tran5; 1Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, MD, 2School of Medicine, University of Maryland, Baltimore, MD, 3Mary Bird Perkins Cancer Center and Gulf South NCORP, Metairie, LA, 4Case Western Reserve University, Cleveland, OH, 5University of Maryland School of Medicine, Baltimore, MD, 6Veracyte, San Diego, CA, 7Veracyte Inc., San Diego, CA, 8ArteraAI, Los Altos, CA, 9Radiation Oncology Dept, University of Maryland, Baltimore, MD, 10Maryland Proton Treatment Center, Baltimore, MD, 11Department of Radiation Oncology, University of Maryland Medical Center, Baltimore, MD, 12University of Rochester, Rochester, NY, 13Department of Radiation Oncology, Rutgers Cancer Institute of New Jersey, New Brunswick, NJ
Purpose/Objective(s): Prostate cancer is the 2nd most common cancer in men, with 1 in 8 diagnosed in their lifetime. Risk stratification relies on T-stage, PSA, and Gleason score, while biomarker tests like Decipher Prostate Genomic Classifier (GC) and Artera Multi-Modal AI (MMAI) enhance prognostic accuracy. Both are validated for localized disease and provide risk information on clinical outcomes (e.g., distant metastasis, mortality) influencing decisions. MMAI can also predict short-term ADT (ST-ADT) benefit. The NRG GU010 trial uses GC to stratify unfavorable intermediate-risk (UIR) patients for treatment intensification. Whether GC and MMAI assess overlapping biology differently or offer complementary insights remains unclear. This study evaluates their correlation and association.
Materials/Methods: We conducted a retrospective single-institution study of localized prostate cancer patients who underwent GC and MMAI testing between May–December 2024. Demographics and clinical characteristics were summarized. GC scores were derived from RNA profiling per Decipher test (Veracyte, San Diego, CA). MMAI scores were generated from digitized H&E slides integrating image-based and clinical features per ArteraAI test (Artera, Los Altos, CA). Pearson’s correlation assessed GC/MMAI score relationships, and cross-tabulation with percent agreement evaluated prognostic risk group association. GC-based ST-ADT classification (<0.4, =0.4 per GU010) was compared with MMAI ST-ADT biomarker.
Results: A total of 76 patients [age: 54–83y, mean 68.9y; 52 (68%) White, 19 (25%) Black; 73 (96%) Non-Hispanic] were analyzed, with most classified as NCCN UIR [very low/low: 1 (1.3%), favorable intermediate: 20 (26.3%), UIR: 52 (68.4%), high: 2 (2.6%)]. GC and MMAI scores showed moderate correlation (r = 0.487, p < 0.001), increasing in UIR patients (r = 0.533, p < 0.001). Categorical risk groups were significantly associated (p = 0.008), with 43.6% of MMAI low-risk patients classified as GC low-risk and 66.7% of MMAI intermediate-risk patients as GC high-risk. No patients were MMAI high-risk. Overall ST-ADT agreement was 40% (30/75), and 36% (19/52) in UIR. Among GC = 0.4 (GU010 darolutamide arm), 4.5% were MMAI ST-ADT positive. Among GC < 0.4 (GU010 ADT deintensification), 17.0% were MMAI ST-ADT positive (Table 1).
Conclusion: Decipher GC and Artera MMAI scores are modestly correlated, and categorical prognostic scores are not interchangeable. Given Artera ST-ADT MMAI positive tumors were nearly all GC low risk, and almost all higher GC score patients were ST-ADT MMAI negative, further work is needed to understand the clinical implications of this observation in contemporary cohorts.
Table 1: Cross-tabulations between GC and MMAI| Artera MMAI | ||||
| Low | Int | High | ||
| Decipher GC | Low (<0.45) | 24 (43.6%) | 3 (13.4%) | 0 (0.0%) |
| Int (0.45-0.6) | 15 (27.3%) | 4 (19.0%) | 0 (0.0%) | |
| High (>0.6) | 16 (29.1%) | 14 (66.7%) | 0 (0.0%) | |
| Artera ST-ADT MMAI | ||||
| Positive | Negative | |||
| Decipher GC (GU 010 cutoffs) | < 0.4 | 9 (17.0%) | 44 (83.0%) | |
| =0.4 | 1 (4.5%) | 21 (95.5%) | ||