A new prospective multicenter study (ASTuTE Trial) published in Clinical and Translational Radiation Oncology in 2026 has found that artificial intelligence-based biomarker testing significantly influences treatment decisions regarding short-term androgen deprivation therapy (ADT) in patients undergoing radiotherapy for intermediate-risk prostate cancer.
The study, titled “Artificial Intelligence Impact on Decision-Making Regarding Short-Term ADT With Prostate Radiotherapy: ASTuTE Clinical Trial Interim Results,” was conducted by Eric Wegener, Jarad Martin, and colleagues across Australian oncology centers.
The interim analysis of 200 patients demonstrated that incorporating a multimodal artificial intelligence (MMAI) biomarker into shared decision-making changed the final ADT decision in 27.5% of cases.
Most changes favored treatment de-escalation, with more than 70% of patients who initially chose ADT subsequently deciding against it after reviewing their AI-derived biomarker results.
The findings highlight the potential of artificial intelligence to support more individualized treatment decisions while raising important questions about whether reducing ADT exposure can preserve long-term cancer outcomes.
ASTuTE Trial Design and Patient Population
The ASTuTE trial is a prospective, multicenter Australian implementation study evaluating how AI-derived biomarkers influence decisions about short-term ADT in patients receiving curative-intent radiotherapy for intermediate-risk prostate cancer.
The study is being conducted across 28 centers, with a planned enrollment of 800 patients.
The published interim analysis included the first 200 patients enrolled between December 2023 and August 2024.
According to National Comprehensive Cancer Network (NCCN) risk classification:
- Unfavorable intermediate-risk prostate cancer: 150 patients (75%).
- Favorable intermediate-risk prostate cancer: 50 patients (25%).
The study incorporated the commercially available ArteraAI multimodal biomarker, which analyzes digitized hematoxylin and eosin-stained prostate biopsy slides together with conventional clinicopathological information.
The system generates two types of information: a prognostic estimate of the patient’s 10-year distant metastasis risk and a predictive assessment of the potential benefit of adding short-term ADT to radiotherapy.
The primary endpoint was the proportion of patients whose shared ADT decision changed after receiving the biomarker results.
Secondary endpoints include quality of life, cost-effectiveness, and long-term oncological outcomes, which will be assessed as the trial matures.
AI Biomarker Testing Changes Treatment Decisions in More Than One in Four Patients
Before biomarker testing, clinicians and patients independently recorded their preferences regarding short-term ADT.
They then discussed these preferences and agreed on an initial shared treatment decision.
The same process was repeated after reviewing the AI biomarker report.
Following AI testing, the shared ADT decision changed in 55 of 200 patients (27.5%; 95% CI, 21.4–34.2%).
The changes were strongly directed toward treatment de-escalation.
Among the 74 patients who initially chose to receive short-term ADT, 52 patients (70.3%) changed their decision to omit ADT after reviewing the biomarker results.
In contrast, among 126 patients who initially decided against ADT, only three (2.4%) changed their decision in favor of treatment.
The difference between these two directions of decision change was statistically significant (P<0.001).
Overall, the proportion of patients with a shared decision to receive ADT decreased from 37% before testing to 12.5% afterward.
These findings suggest that incorporating AI-derived risk estimates into consultations may substantially change treatment recommendations, particularly by identifying patients who may have limited predicted benefit from hormone therapy.
However, the study measured decisions rather than confirmed treatment adherence or long-term outcomes.
Treatment De-Escalation Is Particularly Pronounced in Unfavorable Intermediate-Risk Disease
One of the most notable findings was the difference in decision changes between conventional prostate cancer risk groups.
AI testing changed shared ADT decisions in:
- 35% of patients with unfavorable intermediate-risk disease.
- 4% of patients with favorable intermediate-risk disease.
The difference was statistically significant (P<0.001).
This finding is particularly relevant because short-term ADT is commonly recommended alongside definitive radiotherapy for unfavorable intermediate-risk prostate cancer.
However, conventional clinical risk classification does not necessarily identify which individual patients will derive the greatest benefit from ADT.
The AI-derived prognostic results demonstrated substantial variability within the NCCN risk categories.
The median estimated 10-year distant metastasis risk was:
- 2.6% for unfavorable intermediate-risk patients.
- 1.7% for favorable intermediate-risk patients.
Interestingly, 28% of patients classified as unfavorable intermediate risk had estimated distant metastasis risks comparable to the median for favorable intermediate-risk disease.
Conversely, 28% of favorable intermediate-risk patients had risk estimates overlapping the median for the unfavorable group.
These findings suggest that conventional clinical categories contain substantial biological heterogeneity that may not be fully captured by established staging and grading systems.
Most Patients Are Classified as Unlikely to Benefit From Short-Term ADT
The predictive biomarker suggested that 170 of 200 patients (85%) were unlikely to derive additional benefit from short-term ADT, while 30 patients (15%) were classified as potentially benefiting.
The proportion of biomarker-positive patients was similar across favorable and unfavorable intermediate-risk groups.
However, the analysis revealed an important distinction between predicted relative treatment benefit and absolute risk of distant metastasis.
Among the 30 patients with a positive predictive biomarker, only 13 ultimately chose to receive ADT.
The remaining 17 patients (57%) decided against ADT despite a biomarker result suggesting potential benefit.
Further analysis suggested that absolute distant metastasis risk strongly influenced these decisions.
Among biomarker-positive patients with a low estimated 10-year metastasis risk of less than 3%, 12 of 13 chose to omit ADT.
In contrast, among biomarker-positive patients with an intermediate estimated risk of 3%–10%, 12 of 17 elected to receive ADT.
These findings demonstrate that patients and clinicians do not rely exclusively on whether a biomarker predicts treatment benefit.
Instead, they also consider the magnitude of the underlying cancer risk when deciding whether the potential benefits of ADT justify its adverse effects.
Overall, 85.5% of final shared decisions aligned with the predictive biomarker classification.
AI Testing Narrows Differences Between Clinician and Patient Preferences
Another important finding concerned the relationship between physician recommendations and patient preferences.
Before receiving biomarker results, clinicians were considerably more likely than patients to favor short-term ADT.
Initially:
- 46% of clinicians recommended ADT.
- 29.5% of patients preferred ADT.
This difference was statistically significant (P<0.001).
After reviewing the AI biomarker report, the proportion favoring ADT decreased in both groups:
- 15% of clinicians recommended ADT.
- 12% of patients preferred ADT.
The difference between these post-test preferences was no longer statistically significant (P=0.38).
These results suggest that an individualized biomarker report may help align clinician recommendations with patient preferences by providing additional information about expected treatment benefit.
Importantly, the final decision remained a shared process rather than an automated recommendation determined solely by the AI model.
Could AI Reduce Unnecessary Hormone Therapy and Its Side Effects?
Short-term ADT has demonstrated benefits when combined with radiotherapy in appropriately selected patients with intermediate-risk prostate cancer.
However, androgen deprivation is associated with adverse effects that may substantially influence quality of life.
These can include hot flashes, fatigue, sexual dysfunction, metabolic changes, and other physical and psychological symptoms.
The ASTuTE findings suggest that AI-based biomarker testing could potentially reduce ADT exposure in patients whose predicted absolute benefit is small.
Such treatment de-escalation may ultimately improve quality of life and reduce treatment-related healthcare costs.
However, the interim results do not establish that omitting ADT is oncologically safe in patients whose treatment decisions change after AI testing.
Longer-term follow-up is required to determine whether reductions in ADT use affect biochemical recurrence, distant metastasis, prostate cancer-specific mortality, or overall survival.
The study also has not yet reported the mature patient-reported quality-of-life outcomes or cost-effectiveness results needed to confirm the anticipated benefits of de-escalation.
Important Limitations and the Need for Long-Term Validation
Although the findings demonstrate a substantial influence on shared decision-making, several limitations should be considered.
First, ASTuTE is an open-label, single-arm implementation study without a randomized control group.
Consequently, the observed changes may partly reflect the influence of participating in a clinical trial, clinician expectations, or increased willingness to de-escalate treatment when supported by a novel biomarker.
Second, the AI model’s predictive performance was previously evaluated retrospectively using randomized clinical trial datasets in which radiotherapy techniques and dose schedules differed from many contemporary approaches.
The ongoing ASTuTE study includes modern radiotherapy schedules, including stereotactic body radiotherapy and focal boosting, making prospective outcome validation particularly important.
Third, most patients had relatively low AI-estimated absolute distant metastasis risks, potentially contributing to the pronounced tendency toward treatment de-escalation.
The study was also conducted in collaboration with ArteraAI, the biomarker developer, and several investigators reported employment or financial relationships with participating organizations.
These factors reinforce the importance of independent validation and mature clinical outcome data.
Finally, the present analysis included only 200 of the planned 800 patients and was designed to evaluate changes in treatment decisions rather than cancer control.
Moving Toward AI-Guided Personalized Prostate Cancer Treatment
The ASTuTE trial offers prospective real-world evidence that multimodal AI biomarkers can meaningfully influence decisions about short-term ADT in patients receiving radiotherapy for intermediate-risk prostate cancer.
AI testing changed 27.5% of shared treatment decisions and reduced the proportion of patients choosing ADT from 37% to 12.5%.
The effect was particularly pronounced in unfavorable intermediate-risk disease, where conventional risk classification often supports routine addition of hormone therapy.
The findings also demonstrate that integrating absolute prognostic risk with predicted treatment benefit may help clinicians and patients make more individualized decisions.
However, changing treatment decisions is not the same as demonstrating improved patient outcomes.
Continued enrollment and long-term follow-up will be essential to determine whether AI-guided ADT de-escalation can maintain cancer control while reducing unnecessary treatment and improving quality of life.
If validated, this approach could represent an important step toward integrating digital pathology and artificial intelligence into personalized prostate cancer radiotherapy.