AI Helps Tailor Hormone Therapy for Prostate Cancer

TOPLINE:

A third of patients with high-risk prostate cancer identified as biomarker-negative by an artificial intelligence (AI)–derived tool were safe to forgo long-term androgen deprivation therapy (ADT). Patients identified as biomarker-positive by the tool had a 14% absolute reduction in the estimated 15-year risk for distant metastasis with long-term vs short-term therapy.

METHODOLOGY:

• Long-term ADT is beneficial in patients with high-risk prostate cancer but is associated with treatment-related toxicities. Predictive tools can guide the duration of ADT.
• Researchers developed a multimodal AI (MMAI)–derived predictive biomarker (the MMAI Prostate LT-ADT Predictive Model) using digital images of prostate biopsies before treatment and data from six phase 3 randomized radiotherapy trials to identify patients with high-risk disease who may benefit from extending short-term ADT to long-term ADT.
• The model was validated using a seventh trial (RTOG 9202) that included 1192 patients with localized, nonmetastatic prostate cancer (median age, 70 years) who were randomly assigned to receive radiotherapy with either short-term (4 months; n = 590) or long-term (28 months; n = 602) ADT. Researchers compared outcomes between treatments.
• The primary endpoint was the time to distant metastasis, whereas the secondary endpoint was death with distant metastasis. The median follow-up duration in the validation cohort was 17.2 years.
• In the validation cohort, 66% of patients were classified as long-term ADT MMAI biomarker–positive if they were predicted to benefit from long-term ADT and 34% as long-term ADT MMAI biomarker–negative if predicted to not benefit from long-term ADT.

TAKEAWAY:

• In the overall validation cohort, long-term ADT significantly reduced the risk for distant metastasis. The estimated 15-year risk for distant metastasis was 26% with short-term ADT and 17% with long-term ADT (subdistribution hazard ratio [sHR], 0.64; P < .001).
• The MMAI biomarker predicted distant metastasis (P = .04), with biomarker-positive patients showing reduced distant metastasis with long-term ADT (sHR, 0.55; P < .001) but biomarker-negative patients showing no significant benefit from extended therapy (sHR, 1.06; P = .84).
• The absolute difference in the estimated 15-year risks for distant metastasis between long-term and short-term ADT was 14% in biomarker-positive patients and 0% in biomarker-negative patients, suggesting that the former could avoid prolonged ADT.
• Consistent results were obtained for death with distant metastasis. The estimated 15-year risk was 15% with long-term ADT and 23% with short-term ADT (sHR, 0.64; P < .001). The biomarker also strongly predicted distant metastasis regardless of treatment duration (sHR, 2.35; P < .001).

IN PRACTICE:

“This digital pathology AI biomarker was successfully able to differentiate benefit from [short-term ADT] vs [long-term ADT] and identify approximately one third of men with high-risk [prostate cancer] who may be spared the cost and morbidity of [long-term] ADT without jeopardizing [distant metastasis] outcomes,” the authors wrote. “By sparing these men from unnecessary prolonged ADT, our findings not only contribute to personalized treatment strategies but also alleviate the burden of adverse effects and impaired quality of life associated with extended therapy.”

SOURCE:

The study, led by Andrew J. Armstrong, MD, ScM, Duke University Medical Center in Durham, North Carolina, was published online in Journal of Clinical Oncology.

LIMITATIONS:

The inclusion of patients with slides of both biopsy and transurethral resection of the prostate specimens may have affected model accuracy. Complete clinical data and quality digital pathology images were required for biomarker score generation. Additionally, supervised and self-supervised modeling approaches had limitations as they might propagate biases present in labeled and unlabeled data.

DISCLOSURES:

The study was supported by Artera Inc and through several grants from the National Cancer Institute. Six authors reported being employed with Artera, and seven authors declared having stock and other ownership interests with Artera. Several other authors disclosed receiving research funding and having other ties with various sources.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.