Artificial intelligence-derived biomarker predicts ADT benefit in prostate cancer subset
Click Here to Manage Email Alerts
SAN FRANCISCO — An artificial intelligence-derived digital pathology-based biomarker can help guide treatment decisions for men with localized intermediate-risk prostate cancer, according to study results presented at ASCO Genitourinary Cancers Symposium.
The biomarker (ArteraAI-Predict ADT) demonstrated that a majority of men treated with radiation therapy as part of a large randomized phase 3 trial did not require androgen derivation therapy and could have avoided the adverse events and costs associated with that treatment.
This is the first validated predictive biomarker for the benefit of ADT with radiotherapy in localized intermediate-risk prostate cancer, according to researcher Daniel E. Spratt, MD, chair of the department of radiation oncology at University Hospitals Cleveland Medical Center.
“This is a 10 out of 10. This is one of the most exciting things I’ve been a part of,” Spratt told Healio.
Men with intermediate or high-risk localized prostate cancer treated with radiotherapy typically also receive ADT. However, no validated predictive biomarker has been available to guide ADT use or duration in this setting.
“Many would say predictive biomarkers are the holy grail of precision medicine,” Spratt said. “You can personalize your treatment decisions irrespective of prognosis. Unfortunately, we have zero validated predictive biomarkers in localized prostate cancer, despite it being the most common cancer in men worldwide.”
Spratt and colleagues hypothesized that there is a wealth of unused biological information unrecognized in prostate cancer histopathology, and that using artificial intelligence may identify nonhuman-interpretable features that could allow for the creation and validation of the first predictive biomarker to guide ADT use for men with localized disease.
Spratt and colleagues performed their analysis using information from five randomized phase 3 NRG Oncology trials that included men who received radiotherapy with or without ADT.
Researchers digitized pretreatment biopsy slides and compiled patient data and clinical data from four trials to create a training set to develop the AI-derived predictive biomarker, with the goal of predicting distant metastasis.
They developed a multimodal deep learning architecture to learn from the clinicopathologic and digital imaging histopathology data and identify variations in outcomes between treatment groups.
An independent biostatistician performed validation using data from the fifth trial — NRG/RTOG 9408 — which included men randomly assigned to radiotherapy with or without 4 months of ADT.
Median follow-up in the training cohort (n = 3,935) was 13.6 years (interquartile range, 10.2-17.7). Median follow-up in the validation cohort (n = 1,719) was 17.6 years (interquartile range, 15-19.7).
Model scores in the validation set showed the majority (63%) of men would be predicted to derive no benefit from the addition of ADT to radiotherapy.
“Many people will say, ‘What is in this model?” Spratt said. “Most of it comes from the digital histopathology imagery. Tumor grade or Gleason primary or secondary combined scores contribute very little to this model, likely because the relevant biological and predictive nature is captured in the digital histopathology imagery.”
Independent validation showed a substantial reduction of 15-year distant metastasis with the addition of ADT to radiotherapy in the biomarker-positive group (HR = 0.33; 95% CI, 0.19-0.57). However, researchers reported no benefit in the biomarker-negative group (HR = 1; 95% CI, 0.63-1.56). The biomarker-treatment interaction reached statistical significance (P = .0021).
Competing-risk analysis showed a nearly 10 percentage point absolute reduction in cumulative incidence of distant metastasis at 15 years with ADT vs. without in the biomarker-positive group (15.2% vs. 5.4%), but results showed no absolute reduction in the biomarker negative group with ADT vs. without (6.1% vs. 6.5%).
“As this gets clinically deployed, we will have a tool that should be relatively easy to gain access to,” Spratt told Healio. “Every patient should have pathology slides. They don’t need any special treatment and no special analysis is required. It’s basically digitizing the slides and running them through a computational model. It’s the difference between having a result in a few days vs. 3 to 4 weeks if you send tissue out.”
Spratt said he hopes his institution will be a pilot site for clinical use of this biomarker later this year.
Collapse
Spratt DE, et al. Abstract 223. Presented at: ASCO Genitourinary Cancers Symposium; Feb. 17-19, 2022; San Francisco.
Click Here to Manage Email Alerts