Closing Risk Stratification Gaps in Localized Prostate Cancer With AI

Welcome back to another Urology Times Special Report. In this opening segment, Daniel Spratt, MD, chair and professor of radiation oncology at University Hospitals Cleveland Medical Center, Case Western Reserve University School of Medicine, and a member of the Case Comprehensive Cancer Center, discusses gaps in current risk stratification for localized prostate cancer and the growing role of artificial intelligence (AI) in addressing them.

Spratt explains that the primary tools used to risk stratify patients—prostate-specific antigen (PSA), clinical T category, and Gleason grade—were largely developed between the 1960s and 1990s. Even combined, he notes, these tools are only modestly accurate, which he estimates at roughly 60% to 70%, leaving a significant unmet need for tools that approach 80% to 90% accuracy when deciding whether to intensify treatment, de-escalate treatment, or treat at all. Genomic classifiers have improved on National Comprehensive Cancer Network risk groups in some studies, he says, but they carry limitations, including lack of approval in much of the world, turnaround time, and consumption of tissue.

Spratt describes AI as already embedded in daily practice at his center, from auto-contouring and auto-planning in radiation oncology to diagnostic imaging and, increasingly, prognostication and prediction based on digital pathology.

He then walks through what an individualized risk estimate adds beyond a risk category. In addition to Gleason grade, the pathology slide contains features that are human-interpretable, such as cribriform or intraductal histology, and others that currently are not. Multimodal AI tools can capture this information in a quantitative, objective way, Spratt explains, producing a continuous risk estimate rather than sorting patients into a handful of discrete groups—a refinement that can matter most for patients who fall near the boundary between risk categories.

In the next segment, Spratt explains what the STAR-CAP cohort is and why calibration benchmarking is a distinct, critical step beyond clinical validation.