Using AI Risk Estimates in Prostate Cancer Treatment Decisions

Daniel Spratt, MD, emphasizes that the MMAI test is not a stand-alone score: It incorporates prostate-specific antigen level, T stage, and the clinical data behind National Comprehensive Cancer Network (NCCN) risk groups and combines them with digital pathology. A low MMAI score therefore reflects the full clinical picture plus additional information. In his own practice, Spratt starts with the NCCN risk group to frame treatment options, then uses the MMAI result to judge whether a patient's cancer is more favorable or more aggressive than anticipated. He notes that the NCCN guidelines state that advanced risk stratification tools can support choosing treatment options from adjacent risk categories.

He walks through 2 discordant scenarios. For a patient with clinically high-risk features, such as Gleason 8 disease, but a low MMAI result, Spratt explains that the patient remains a candidate for radical prostatectomy, and the favorable score may give the urologist and patient more confidence that surgery alone could be sufficient, with a lower likelihood of needing postoperative therapy than with a high score.

In the reverse case, an NCCN low-risk patient with a high MMAI result, which he describes as uncommon but possible, Spratt says active surveillance remains an option but should not be one-size-fits-all. Rather than scheduling a repeat biopsy in 3 years, he would bring that patient back within 1 year because of a higher chance of progressing off surveillance.

Spratt adds that the absolute risk estimate is what he finds most useful in practice, and that this analysis supports confidence that those estimates are accurate within about a percent in a large cohort that included many surgical patients, supporting shared decision-making about de-escalation, surveillance, intensification, or referral for radiation.

In the next segment, Spratt outlines the questions urologists should ask before adopting any AI-based tool and shares his key takeaway from the benchmarking data.