A multi-modal AI prostate test changed decisions about short-term androgen deprivation therapy (ST-ADT) for men receiving radiotherapy for intermediate-risk prostate cancer, according to interim findings from the Australian ASTuTE trial.
The prospective, multicenter implementation study included 786 patients. Before testing, clinicians and patients recorded whether they planned to add ST-ADT to curative radiotherapy. They reconsidered that decision after receiving prognostic and predictive results from the biomarker.
Among 230 patients initially recommended ST-ADT, 176, or approximately three-quarters, decided against treatment after testing. The results, presented at ASTRO 2026, also supported adding ST-ADT for 21 patients who had not initially planned to receive it. Across the full cohort, planned ST-ADT use fell by a net 20.7 percent.
“What stood out was how significantly the [multi-modal AI prostate test’s] result could reshape treatment conversations, giving doctors and patients concrete information to weigh together versus a dictated outcome,” said Eric Wegener, a radiation oncologist and the study’s principal investigator. “That’s what let 176 men reconsider hormone therapy and helped us identify 21 more who stood to benefit from it.”
ST-ADT lowers testosterone and can improve the effectiveness of radiotherapy in prostate cancer. However, its benefits vary among patients, and treatment can cause sexual dysfunction, hot flashes, persistent fatigue, osteoporosis, metabolic changes, cardiovascular complications, and mood disorders.
Treatment decisions have traditionally relied on clinical and pathologic factors such as prostate-specific antigen level, tumor stage, Gleason score, and the proportion of biopsy cores containing cancer. These measures remain central to risk assessment but do not always establish whether an individual patient will benefit from adding hormone therapy.
The multi-modal AI prostate test combines these clinical factors with information extracted from a digitized hematoxylin-and-eosin-stained biopsy slide. The selected slide contains the tumor core with the highest Gleason grade used by the local pathologist in making the diagnosis. A certified pathologist reviews the digital image for suitability before analysis.
The system produces two types of information. Its prognostic component estimates the patient’s 10-year risk of distant metastasis, while its predictive component assesses whether adding four to six months of ST-ADT to radiotherapy is likely to provide benefit.
This distinction is important for laboratories and pathology teams. A prognostic result estimates the likelihood of a future outcome, whereas a predictive result estimates the likelihood of benefit from a particular treatment. Combining both may provide more clinically useful information than risk classification alone.
The ASTuTE results indicate that the test influenced decisions in both directions. Most changes involved patients avoiding planned ST-ADT, but testing also identified a smaller group for whom treatment was newly recommended.
Investigators also reported that clinicians became less likely to recommend ST-ADT before receiving test results as the study progressed. Although the reason was not established, this pattern suggests that growing experience with the biomarker may have influenced how clinicians assessed risk beyond individual test reports.
“The ASTuTE trial demonstrated how a validated AI biomarker changes practice in the real world, across centers nationally, not just in a controlled setting,” said Jarad Martin, a study investigator. “We even saw clinicians’ baseline recommendations evolve across the duration of the trial, which tells us that the [multi-modal AI prostate test] is shaping how oncologists think about risk more broadly.”
The trial’s primary endpoint was the proportion of shared treatment recommendations that changed after testing. Longer-term outcomes, including quality of life and disease control, will be evaluated through registry linkage after a median follow-up of 5 years.
The current findings therefore demonstrate an effect on decision-making rather than proof of improved patient outcomes. They do not yet establish whether avoiding ST-ADT in biomarker-negative patients maintains cancer control or whether adding treatment in biomarker-positive patients reduces recurrence or mortality.
Further assessment will also be needed to examine analytical reproducibility, slide-quality requirements, performance across patient groups, and integration with routine digital pathology workflows. The results show how pathology-derived AI biomarkers could support treatment selection in addition to conventional risk classification.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
