Does Artificial Intelligence Change the Treatment Decision in Prostate Cancer? The DIRECT-AI Registry Enrols Its First Patient

KEY FINDINGS

• First patient enrolled: The first patient in the DIRECT-AI registry was included on 13 August 2026 at The Urology Place in San Antonio, Texas.
• Non-interventional design: The study is observational and non-interventional; no additional procedure, additional biopsy or additional test beyond routine care is applied.
• A two-phase structure: The first phase measures the change in treatment choice, physician confidence and patient-physician communication; the second phase measures hard endpoints such as distant metastasis and survival at 2 and 5 years.
• What distinguishes the test: The ArteraAI Prostate Test is positioned as a test included in the NCCN guidelines for localized prostate cancer that provides both prognostic and predictive information.
• No contribution in the biomarker-negative group: n=1,046; adding short-course ADT to radiotherapy did not significantly change the rate of distant metastasis (HR 0.92; 95% CI 0.59-1.43; P=0.71).
• Marked benefit in the biomarker-positive group: n=673; adding short-course ADT reduced the risk of distant metastasis by approximately two thirds (HR 0.34; 95% CI 0.19-0.63; P<0.001). The test identifies the 34% of patients who will benefit from ADT.

Background: An Unresolved Question in Localized Prostate Cancer

The fundamental difficulty in managing localized prostate cancer is the wide heterogeneity in the biological behaviour of the disease. The clinical course of two patients with the same Gleason score, the same PSA value and the same clinical stage can diverge radically. This heterogeneity leads to a two-way error, particularly in intermediate-risk disease: one group of patients is treated more aggressively than necessary and exposed to urinary, sexual and hormonal side effects, while another group carries the risk of metastatic progression because of undertreatment.

The most concrete example of this dilemma is the decision on short-course androgen deprivation therapy added to definitive radiotherapy. Randomized trials have shown that short-course ADT reduces metastasis and mortality on average in the intermediate-risk population. However, the costs of ADT - hot flushes, loss of libido, sarcopenia, metabolic syndrome, bone loss and cardiovascular burden - are entirely unnecessary for patients who will derive no benefit.

Traditional risk classifications - NCCN, D'Amico, CAPRA - are prognostic tools; they predict the patient's course. But they are not predictive; they do not tell us which patient will benefit from a particular treatment. This is precisely why multimodal artificial intelligence (MMAI) biomarkers have entered the clinical agenda.

The distinction between prognostic and predictive: A prognostic marker answers the question "how aggressive is this disease?"; a predictive marker answers "is there a return on adding this treatment for this patient?" In personalised oncology, the real clinical value lies in the answer to the second question.

How Does the ArteraAI Prostate Test Work?

The test processes digitised images of standard haematoxylin and eosin (H&E) stained prostate biopsy slides together with the patient's clinical variables (PSA, Gleason grade group, clinical T stage, age) in a single deep learning architecture. The algorithm integrates tissue-level morphological patterns that the pathologist cannot distinguish by eye with clinical parameters to produce both a prognostic estimate and a prediction of treatment benefit.

The methodological strength of the method comes from the source of its development and validation: the model was trained and validated on patient-level data from phase 3 randomized controlled trials conducted within NRG Oncology/RTOG. That it rests not on retrospective institutional series but on cohorts in which selection bias was controlled through randomization is the most critical element strengthening the biomarker's predictive claim.

Regulatory and clinical position

• August 2025: The FDA granted the test de novo authorisation in the non-metastatic prostate cancer indication.
• June 2026: The test was made available for clinical use to support treatment planning in metastatic hormone-sensitive prostate cancer (mHSPC).
• NCCN Guidelines: It appears in the localized prostate cancer section as a test that both predicts treatment benefit and provides long-term prognosis.
• Laboratory: It is run through a CLIA-certified and CAP-accredited laboratory in Jacksonville, Florida.

DIRECT-AI Study Design

Key Results: What Does the Test Rest On?

DIRECT-AI itself has not yet produced results; the registry has only just begun. However, the validation data forming the clinical rationale for the test at the centre of the study demonstrate the discriminating power of the biomarker in the decision to add short-course ADT to radiotherapy.

The difference between these two rows constitutes the essence of personalised oncology: of two patients in the same clinical risk group, hormonal therapy reduces the risk of metastasis by two thirds for one, while for the other it means only a burden of side effects.

Implications for Clinical Practice

1. Personalising treatment intensity

In a patient with intermediate-risk localized prostate cancer planned for radiotherapy, the decision on short-course ADT has until now been made largely through a generalisation based on the risk group. Adding a predictive biomarker to this decision carries the potential for a two-way gain: reducing unnecessary hormonal exposure and not leaving treatment incomplete in a patient who will genuinely benefit.

2. The quality of shared decision-making

Prostate cancer is a disease in which treatment options are close to oncological equivalence but their side effect profiles diverge radically. The decision is therefore not a technical choice for the physician to make alone but a process integrated with the patient's values. That the first phase of DIRECT-AI aims to measure physician confidence and the quality of the consultation directly is a noteworthy methodological choice.

3. The move from analytical performance to clinical benefit

That a biomarker is discriminating in randomized trial data is a necessary condition for clinical benefit; but it is not a sufficient one. The real value of DIRECT-AI is that it attempts to measure prospectively the step of clinical utility that is frequently skipped in the field of molecular and digital diagnostics.

4. Points to note in interpretation

• The registry is not randomized; attributing the change in treatment decisions causally to the test alone is methodologically limited.
• The study is being conducted by the manufacturer; external validation data from independent centres will be of critical importance.
• The test's performance in non-US populations and its integration into pathology workflows in Türkiye must be assessed separately.
• Year 2 and year 5 data must be awaited for long-term oncological endpoints; what we have today is the discriminating power of the biomarker, not a change in decisions.

Conclusion

The launch of the DIRECT-AI registry marks a meaningful threshold in the maturation of artificial intelligence-based biomarkers in prostate cancer. The field is moving from the stage of demonstrating an algorithm's statistical performance to the stage of questioning whether that algorithm genuinely improves the decision in the clinic and the patient's long-term course.

The defining question for the period ahead in prostate cancer management will not be "can artificial intelligence predict accurately?" but "does artificial intelligence support the right decision?" For our patients, the meaning of this distinction is concrete: being protected from the burden of unnecessary hormonal therapy, or receiving the treatment genuinely needed in good time.

References

• Clarke H. Registry launches to assess real-world clinical utility of ArteraAI Prostate Test. Urology Times, 13 August 2026.
• Artera. Artera Enrolls First Patient in DIRECT-AI Registry. Press release, 13 August 2026.
• Artera. DIRECT-AI Registry study information and ArteraAI Prostate Cancer Test technical/clinical data.
• Spratt DE, et al. Meta-analysis of Individual Patient-level Data for a Multimodal Artificial Intelligence Biomarker in High-risk Prostate Cancer: Results from Six NRG/RTOG Phase 3 Randomized Trials.
• External Validation of a Digital Pathology-based Multimodal Artificial Intelligence Architecture in the NRG/RTOG 9902 Phase 3 Trial. European Urology Oncology.