Putting Data from a Post-RP Multimodal AI Prognostic Model Into Context

Every retrospective prognostic study carries limitations that must be weighed when translating findings into practice, and the external validation of the post–radical prostatectomy (RP) multimodal AI (MMAI) model at Mayo Clinic is no exception. In the final segment of this 5-part series, R. Jeffrey Karnes, MD, the Dr. Anson Clark Professor of Urology at Mayo Clinic in Rochester, Minnesota, discusses the constraints of the study design candidly and frames what prospective evidence would be required to fully establish MMAI's role in routine post-RP management. Chief among the limitations he identifies are the retrospective cohort structure, limited racial diversity, and the geographic and socioeconomic characteristics associated with a tertiary referral center—factors that may affect generalizability to broader practice settings.

Karnes notes that a prospective randomized trial, in which the MMAI score is incorporated into the decision-making process rather than applied retrospectively to archival tissue, represents the evidentiary gold standard—and that such a trial is currently under way, examining whether MMAI can guide intensification or de-intensification of hormonal therapy in the postoperative setting with radiation. He also describes the model's practical accessibility as one of its more notable attributes: because it is based on digitized hematoxylin and eosin slides available from standard pathology workflows, it does not require additional tissue extraction, is rapidly returnable, and is in principle deployable globally. He identifies the development of predictive—rather than purely prognostic—models as a key priority, noting that tools capable of predicting response to specific therapies would represent a meaningful advance beyond current risk stratification.

Looking further ahead, Karnes anticipates that digital pathology and multimodal AI will reshape oncology practice across multiple disease sites, including bladder and kidney cancers, in addition to prostate cancer. He acknowledges the broader implications for the pathology workforce—suggesting that AI may support higher-quality subspecialty reads and facilitate expert consultation across institutions through digital slide sharing—while expressing measured optimism about how competitive innovation in the AI model space may accelerate progress. His closing message for community urologists is direct: engage patients in a conversation about what the MMAI score is designed to answer before ordering it, establish a shared plan for both possible results, and recognize the test's demonstrated value in a post-RP setting where minimizing both overtreatment and undertreatment remains an enduring clinical challenge.