1117 - Validation of a Prognostic Multimodal Artificial Intelligence (MMAI) Model in Asian Prostate Cancer (PCa) Patients (pts) from Singapore

Purpose/Objective(s): The digital pathology-based MMAI prognostic model (ArteraAI Prostate Test) was originally developed using biopsy image and clinical data from North American (NA) phase III clinical trials. Although the model has been shown to accurately predict outcomes in mostly Caucasian and African-American men with PCa, data is lacking on its performance in Asian pts, particularly in non-NA settings. Thus, we set out to validate the prognostic MMAI model in an Asian PCa cohort from Singapore.

Materials/Methods: MMAI scores were generated using diagnostic biopsy H&E images and clinical data (age, PSA, T-stage) from a PCa cohort treated at an institute in Singapore. MMAI association with distant metastasis (DM, primary endpoint), biochemical failure (BF, Phoenix’s definition), DM-free survival (DMFS), and overall survival (OS) was assessed using Fine and Gray (DM, BCR) or Cox (DMFS, OS) regression. Associations were assessed for MMAI scores (0-1) per standard deviation increase and categorical risk groups. A pre-defined statistical plan showed an underpowered analysis for the primary endpoint. Sub/Hazard ratio (s/HR) with 95% confidence intervals in [ ] and p-value are reported.

Results: MMAI scores were generated for 146 pts with complete image and clinical data (97.3% Asian). The cohort consisted of 2 NCCN low (1.4%), 23 favorable intermediate (int) (15.8%), 33 unfavorable int (22.6%), 23 high (15.8%), and 57 very high (39.0%), as well as 6 regional (reg, 4%) and 2 metastatic (met) pts (1.4%, removed in DM/BF analysis). With a median follow up of 6.0 years, the 5-year DM rate was 4% [1%-8%]. MMAI categorized 12 (8.3%), 54 (37.5%), and 78 (54.2%) men as low, int, and high risk, respectively, with meaningful risk stratification for 5-yr DM comparing MMAI low/int (no DM) to high risk (7% [2-14%]) groups. Reclassification was seen across all clinical risk group categories (Table 1), including 9/23 (39%) and 12/57 (21%) pts down-classified to MMAI low/int risk in NCCN high and very-high risk groups, respectively. MMAI score was significantly associated with risk of DM (sHR: 2.59 [1.40-4.78], p=0.003), BF (sHR: 2.09 [1.28-3.42], p=0.003), and DMFS (HR: 2.00 [1.19-3.36], p=0.009), but not OS (HR: 1.37 [0.63-3.00], p=0.43), and significance findings remained when adjusted for clinical risk group.

Conclusion: This study validates a biopsy MMAI model as an independent prognostic tool in Asian PCa pts, supporting performance of this model in different racial subgroups, as well as geographic robustness. Future studies may seek to evaluate the model in larger Asian pt cohorts to ensure its performance across diverse populations and geographies.