Posted: Friday, June 9, 2023
Use of artificial intelligence (AI) may aid in the development of a biomarker to predict benefit from long-term androgen-deprivation therapy in patients with localized, high-risk prostate cancer, according to presentation at the 2023 American Society of Clinical Oncology (ASCO) Annual Meeting based on data from multiple phase III NRG/RTOG randomized trials (Abstract 5001). The work, presented by Andrew J. Armstrong, MD, of Duke Cancer Institute, Durham, North Carolina, and colleagues, demonstrates reportedly the first time an AI-derived, digital pathology–based platform has been used to develop such a predictive biomarker.
The retrospective study used 3,833 patients who had received radiotherapy with or without androgen-deprivation therapy from several phase III randomized trials (RTOG 9408, 9413, 9902, 9910, and 0521). Using patient-associated data, the authors trained the AI model to predict the differential benefit of long-term androgen-deprivation therapy for 2,641 patients. Validation of the model was conducted using data associated with 1,192 patients. The authors noted that event rates, estimated by cumulative incidence and unrelated deaths, were treated as competing risks.
According to the investigators, the AI-derived model was prognostic for distant metastasis (hazard ratio [HR] = 2.35, 95% confidence interval [CI] = 1.72–3.19, P < .001). Patients who were considered to be AI biomarker–positive had reduced distant metastasis when treated with long-term androgen-deprivation therapy (HR = 0.55, 95% CI = 0.41–0.73, P <.001); however, no benefit was identified for men considered to be AI biomarker–negative (HR = 1.06, 95% CI = 0.61–1.84, P = .84). Furthermore, the 10-year distant metastasis rate was better among men who possessed the AI-derived biomarker. Lastly, the authors noted that the biomarker successfully identified 34% of men in the study who might benefit from short-term androgen-deprivation therapy and might avoid the side effects of long-term therapy.
Disclosure: For full disclosures of the study authors, visit coi.asco.org.