An artificial intelligence–digital pathology algorithm to predict outcomes in a cohort of men diagnosed with prostate cancer within a low resource setting.

Abstract

399 Background: In the era of precision oncology, personalized prognostic biomarkers have demonstrated superior prostate cancer (PCa) risk stratification, but their deployment in low- and middle-income countries (LMICs) is limited by resource and infrastructure constraints. Although digital pathology-based multimodal artificial intelligence (MMAI) model offers a scalable, low-cost alternative using routine hematoxylin and eosin (H&E) slides, few studies have evaluated the performance of these models on samples from LMICs. Here, we report the first evidence of a prognostic MMAI in predicting biochemical recurrence-free survival (BCRFS) and distant metastasis-free survival (DMFS) in a cohort of native African men (NAM) with PCa, aiming to bridge this gap and advance personalized care in low-resource settings. Methods: We performed a retrospective analysis of NAM with localized PCa who had available digital histopathologic images from biopsy (Bx) specimens. MMAI scores were generated by applying deep learning to digitized H&E Bx slides, integrating histopathologic features with clinical variables. This approach preserves tissue samples and can be readily implemented in LMICs. MMAI scores were evaluated as both continuous and categorical variables using pre-specified cutoff. The primary objective was to determine whether MMAI can predict BCRFS and DMFS using Harrell’s concordance index (C-index) derived from Cox proportional hazard models. Next, HTG EdgeSeq and gene set enrichment analysis (GSEA) was performed to identify distinct biological pathways that are associated with low/intermediate vs. high MMAI scores. Results: The final analytical cohort included 88 NAM with both MMAI and longitudinal clinical outcomes data. The median age of the cohort was 68 years (IQR: 63-73). The majority of patients (N=86, 98%) received radiation therapy as their primary treatment. The median follow-up time was 27 months (IQR: 13-44 months). Each 1–standard deviation (SD) increase in MMAI score was associated with higher risk of BCR (HR=4.38, 95% CI 1.87-10.27, P =.0007) with a C-index of 0.84, and DM (HR=2.68, 95% CI 1.14-6.32, P =.02) with a C-index of 0.86, indicating excellent discriminatory performance of MMAI. Lastly, GSEA revealed strong positive correlation between MMAI high and immune related pathways (NK cells, B cell markers, T cell markers) and negative correlation between MMAI high and MTORC1 Signaling, Unfolded Protein Response, and Androgen Response pathways. Conclusions: Our study provides the first evidence that MMAI score predicts BCRFS and DMFS in a native African PCa cohort. The performance and ease of deployment of MMAI in low resource settings has the potential to close the gap in personalized care for men with PCa globally.

Ähnliche Arbeiten