AI biomarker predictive of long-term ADT benefit in patients with prostate cancer

Investigators have validated the first AI-derived predictive biomarker of long-term androgen deprivation therapy (ADT) benefit among men with localized high-risk prostate cancer using multiple phase 3 NRG/RTOG trials.¹

Patients who were AI-biomarker positive showed reduced DM with long-term ADT, whereas men who were AI-biomarker negative showed no benefit in DM.

The data were presented at the 2023 American Society of Clinical Oncology Annual Meeting in Chicago, Illinois.

“The ArteraAI long-term ADT predictive model was successfully developed using prostate biopsy, histopathology imaging data, and standard clinical variables, with an implication that about a third of high-risk patients could be spared the long-term side effects of long-term ADT and have excellent outcomes,” said lead author Andrew J. Armstrong, MD, MSc, during the presentation. Armstrong is a medical oncologist at Duke University Department of Medicine in Durham, North Carolina.

The biomarker training set included data on 2641 men from RTOG 9408, 9413, 9902, 9910, and 0521. The model was validated using data on 1192 men from RTOG 9202.

"The ArteraAI long-term ADT predictive biomarker was trained on NRG/RTOG phase 3 studies across 1000s of patients using digital pathology tools to machine learn the AI algorithms important in predicting the benefits of distant metastases prediction with long-term ADT," Armstrong added.

All men included in the study were receiving radiotherapy. The primary end point was distant metastasis (DM).

In the validation cohort, long-term ADT was shown to improve DM (P < .001), which is consistent with the literature. The AI model was also found to be prognostic for DM (P < .001). Patients who were AI-biomarker positive (n = 785, 66%) showed reduced DM with long-term ADT (sHR .55, 95% CI .41-.73, P < .001). Conversely, men who were AI-biomarker negative (n = 407, 34%) showed no benefit in DM (sHR 1.06, 95% CI .61-1.84, P = .84).

The investigators also compared DM outcomes among patients who received long-term ADT vs those who received short-term ADT. A difference of 13% was seen among patients who were AI-biomarker positive between the 2 groups. There was a 2% difference in patients who were AI-biomarker negative.

Further, National Comprehensive Cancer Network (NCCN) clinical risk was found to be prognostic, but not predictive of benefits to long-term ADT.

“There was no interaction between NCCN risk and the benefits of long-term ADT, illustrating that the pathology digital tool is able to see things that our current clinical data is not able to predict,” said Armstrong during the presentation.

A similar biomarker effect was seen in regard to death with DM among patients who were AI-biomarker negative. In AI-biomarker positive men, risk of death with DM was reduced with long-term ADT (HR = .57, P = .11).

Armstrong added, “The clinical impact of the digital pathology AI biomarker illustrates that about 29% of our high-risk patients tested AI biomarker negative, and they could thus be spared the long-term [adverse] effects of ADT for 2 to 3 years… Conversely, [in] another group of what we call NCCN clinical intermediate-risk patients, about 43% were AI-biomarker positive and may benefit from longer extensions of their ADT to reduce distant metastases.”

Reference

1. Armstrong AJ, Liu VY, Selvaraju RR, et al. Development and validation of an AI-derived digital pathology-based biomarker to predict benefit of long-term androgen deprivation therapy with radiotherapy in men with localized high-risk prostate cancer across multiple phase III NRG/RTOG trials. J Clin Oncol. 2023;16(suppl)5001: doi:10.1200/JCO.2023.41.16_suppl.5001