Can Artificial Intelligence Tell Us Who Needs Hormone Therapy in Prostate Cancer?

KEY FINDINGS

• The problem: The decision whether to add hormone therapy to radiotherapy in prostate cancer has until now been made largely by looking at the average of patient groups; yet not everyone in the same group benefits from this treatment.
• The new approach: Artificial intelligence examines the high-resolution digital image of the biopsy slide together with the patient's other medical data to produce an answer to the question "will this person benefit from hormone therapy?"
• Current data: In patients the AI judged as "will benefit", adding hormone therapy reduced the risk of distant metastasis by approximately two thirds; in the group judged as "will not benefit", no measurable contribution was found (1,719 patients in total).
• The new study: The first patient in the DIRECT-AI registry was enrolled on 13 August 2026; whether the test genuinely changes the physician's decision and long-term outcomes will be followed.
• No additional burden on the patient: The research is observational; no additional biopsy, additional test or different treatment is applied.
• The situation today: The test is currently in use in the United States; 2-year and 5-year results must be awaited to establish whether it genuinely makes a difference in clinical practice.

The Rarely Discussed Difficulty of Prostate Cancer

The best-known thing about prostate cancer is that it usually progresses slowly. That is true - but it is only half true. The real difficulty is this: this cancer behaves extraordinarily differently from one patient to another. In one man it sits silently for years without causing any problem; in another it spreads to distant parts of the body within a few years.

Being able to tell these two patients apart is one of the oldest and most critical questions in urology. Because in every situation in which we cannot distinguish them, we risk making one of two mistakes:

• Overtreatment: Treating a cancer that would in fact have run a quiet course aggressively, unnecessarily exposing the patient to urinary incontinence, loss of sexual function or hormonal side effects.
• Undertreatment: Underestimating an apparently quiet but in fact aggressive cancer and giving it the chance to spread.

The risk classifications we use today - "low", "intermediate" and "high" risk groups - achieve this distinction up to a point. The fate of two patients placed in the same risk group can still be very different from one another.

A Concrete Example: Should Hormone Therapy Be Added?

The place where this problem appears most clearly is the decision whether to introduce hormone therapy in a patient planned for radiotherapy. Prostate cancer cells need testosterone to grow; hormone therapy lowers testosterone to very low levels, cutting off the fuel of the cancer cells. When applied together with radiotherapy, large trials have shown that it increases the success of treatment in many patients.

But this treatment has a price: hot flushes, a marked reduction in sexual desire, loss of muscle mass, bone thinning, weight gain, mood swings and an additional burden on the cardiovascular system. These side effects are manageable; but they are entirely unnecessary for a patient who will derive no real benefit.

The fundamental issue here is this: large trials tell us that hormone therapy benefits the average of the patient group. But the average is not the real situation of any individual patient. The goal of personalised medicine is to unpick that mixture.

How Does Artificial Intelligence Come In?

Prostate cancer is diagnosed by biopsy; the tissue fragments taken are stained, placed on glass slides and assessed under the microscope by a pathologist. What is new is this: these glass slides can now be scanned at high resolution and converted into a digital image, and an artificial intelligence model can learn patterns invisible to the human eye by examining together both these images from thousands of patients and what happened to those patients over the years.

The system in question - the ArteraAI Prostate Test - combines digital biopsy images with the patient's other information (PSA value, the microscopic grade of the cancer, age, the extent of the disease) to produce two separate outputs:

• Prognostic estimate: How much trouble is this cancer likely to cause in the coming years?
• Treatment benefit estimate: Will adding hormone therapy to this patient have a concrete return?

The second question is far harder than the first and far more valuable clinically. Because knowing how serious a disease is one thing, and knowing which treatment will work is quite another.

What Do the Numbers Say?

The difference between the two rows of this table on its own explains why personalised medicine matters so much: of two patients who look identical on paper, hormone therapy can be a vital gain for one and merely a burden for the other.

What Is DIRECT-AI Testing?

The data above are encouraging, but they are all retrospective analyses. Showing that a test can make accurate predictions is not the same as showing that the test makes a real difference to patient benefit: a test may predict perfectly, but if physicians do not take the result into account nothing changes; or decisions may change but that change may not improve the patient's long-term health.

DIRECT-AI is designed precisely to fill this gap. The first patient in the research, to be conducted across the United States, was enrolled on 13 August 2026 at The Urology Place in San Antonio, Texas.

Dr Naveen Kella, founder of The Urology Place, notes that the research makes it possible to understand how the information produced by artificial intelligence fits into daily practice and how it helps in having more informed discussions with patients.

What Distinguishes This Test

The great majority of advanced diagnostic tests used in prostate cancer provide information about prognosis. The distinguishing feature of the ArteraAI Prostate Test is that it is positioned as a test included in international guidelines (NCCN) that can make both a prognostic estimate and a treatment benefit estimate at the same time.

The methodological reason for taking this claim seriously is important: the model was developed and validated not with retrospective data collected from the archives of individual hospitals but with patient data from internationally conducted randomized controlled trials. Because patients in such trials are allocated randomly to treatment arms, the question "were the patients in better condition the ones who received this treatment?" disappears.

A Measure of Healthy Scepticism

• This is not a randomized trial: DIRECT-AI does not divide patients into different groups; it only records what happens in real life. It is methodologically impossible to attribute all the changes observed directly to the test.
• The study is being conducted by the company that produces the test: This does not mean the data are invalid; but confirmation from independent centres will carry separate weight.
• Results require time: Year 2 and year 5 data must be awaited for long-term cancer outcomes. What we have today is data on the discriminating power of the test - not evidence that patient outcomes have improved.
• Geographical validity is a separate question: The test was developed in a US population; its performance in patient profiles and pathology workflows in other countries must be examined separately.

Glossary

• Biopsy: Taking small tissue samples from the prostate with a fine needle for diagnostic purposes.
• PSA: A protein produced by prostate tissue and measured by a blood test; it does not make a diagnosis on its own.
• Radiotherapy: Treatment aiming to destroy cancer cells with high-energy radiation, which may be an alternative to surgery.
• Hormone therapy (ADT): Drug treatment that suppresses the testosterone prostate cancer needs in order to grow.
• Metastasis: The spread of cancer from the organ in which it started to distant sites such as bone or lung.
• Biomarker: A measurable indicator that provides information about the course of a disease or whether a treatment will work.

Conclusion: The Difference Between the Right Prediction and the Right Decision

The debate about the place of artificial intelligence in medicine often revolves around the wrong question. The real issue is not whether an algorithm can predict more accurately than a human; it is whether that prediction makes the physician's decision more accurate and the patient's journey less gruelling.

What is noteworthy about DIRECT-AI is that it asks this second question directly and does not seek an easy answer. The field is moving from the stage of "can the algorithm predict accurately?" to "does the algorithm support the right decision?"

For our patients, the meaning of this distinction is extremely concrete: being protected from the months-long burden of an unnecessary hormone therapy, or receiving the treatment genuinely needed exactly on time. Personalised medicine is the move from telling a patient "in your group this usually happens" to being able to say "in your situation this is true".

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.
• 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.