(UroToday.com) The 2023 American Society of Clinical Oncology (ASCO) annual meeting held in Chicago, IL between June 2nd and June 6th was host to a prostate, testicular, and penile cancers oral abstract session. Dr. Andrew J. Armstrong presented the results of a pooled analysis of multiple phase III NRG/RTOG trials aimed at developing and subsequently validating an AI-derived digital pathology-based biomarker to predict the benefit of long-term ADT addition to radiotherapy in men with localized, high-risk prostate cancer.
The current standard of care for men with high-risk, localized prostate cancer remains radiotherapy plus long-term (2-3 years) concurrent/adjuvant ADT. However, not all high-risk men will require long-term ADT, subjecting a significant proportion to long-term side effects. As such, can we identify men with high-risk, localized prostate cancer with a low risk of metastasis after short-term ADT, and thus spare them the risks of long-term ADT?
Current genomic and clinical risk stratification tools remain mainly prognostic (i.e., predict disease course, irrespective of treatment received). There is a clear need for predictive biomarkers, which provide information regarding what specific treatment confers the most benefit for specific patient populations. Artificial intelligence tools leveraging digital pathology data may provide a solution to the issues of predictive biomarker paucity.
Dr. Spratt and colleagues have previously demonstrated that a digital pathology-derived AI biomarker may be used to predict short-term ADT benefits for intermediate risk patients (ASCO GU 2022). As demonstrated below, intermediate risk patients who were biomarker negative did not benefit from short-term ADT. Conversely, those with a positive biomarker status had significantly decreased distant metastases rates when short-term ADT was added to radiotherapy (HR: 0.33, p<0.001):
In this study, the ArteraAI® long-term ADT predictive biomarker was trained on NRG/RTOG 9408, 9413, 9902, 9910, 0126, and 0521 to predict the benefit of long-term ADT use on distant metastases. After the model was locked, the ArteraAI® long-term ADT predictive biomarker was validated in the NRG/RTOG 9202 trial that compared short-term and long-term ADT in combination with radiotherapy.1 The predictive utility of this biomarker was evaluated for ADT duration with Fine-Gray or Cox proportional hazards interaction models. The events rates were estimated using cumulative incidence function curves.
The ArteraAI Multimodal Artificial Intelligence (MMAI) Architecture relies on clinical data (age, PSA, Gleason Score, T-stage) and H&E pathology data to derive an AI score that predicts the magnitude of benefit from long-term ADT with respect to the outcomes of distant metastasis or death with distant metastasis.
But what is this artificial intelligence model praying attention to (i.e., how is this AI predictive score generated)? As demonstrated below, the AI predictive score appears to be mostly driven by imaging/pathology features (43%), with reliance on the other clinical factors, including T-stage (18%), age (16%), and baseline PSA (8.4%).
After the ArteraAI predictive model was trained using data from the 9408, 9413, 9902, 9910, 0126, and 0521 NRG/RTOG trials, it was subsequently validated in the phase III NRG/RTOG 9202 trial, which included 1,192 patients, of whom 80% had high-risk disease. Patients were randomized in a 1:1 fashion to either prostatic/pelvic radiotherapy + short-term ADT (4 months) or radiotherapy + long-term ADT (28 months). This was a positive trial that demonstrated that long-term ADT was associated with improved disease-free survival, distant metastatic, prostate cancer-specific, and overall survivals, compared to short-term ADT. For the purposes of this analysis, the outcomes of interest were development of distant metastases (primary) and death with distant metastases (secondary). Other survival outcomes were not primarily evaluated due to concerns about the AI model not being able to predict overall survival outcomes and that the outcome of prostate cancer-specific mortality may be confounded by the competing risk of other-cause mortality (>50% of cohort).
With regards to baseline characteristics, stratified by biomarker status (positive/negative), we note that biomarker positive patients had a higher baseline PSA (25.1 versus 14 ng/ml) and had high or very high-risk disease more frequently (87% versus 67%).
As mentioned above, patients receiving long-term ADT had decreased distant metastasis rates compared to patients receiving short-term ADT (HR: 0.64, 95% CI: 0.50 – 0.82, p<0.001).
The validation analysis in the NRG/RTOG 9202 cohort demonstrated that the ArteraAI biomarker was a significant predictor of the utility of long-term ADT use in this cohort (interaction p=0.04). As demonstrated below, patients in the biomarker negative group (n-407) derived no benefit from long-term ADT use, compared to short-term ADT use (HR: 1.06, 95% CI: 0.61 – 1.84, p=0.84). Conversely, patients in the biomarker positive group had significant improvements in the distant metastasis rates with long-term ADT (HR: 0.55, 95% CI: 0.41 – 0.73, p<0.001). Accordingly, approximately 1/3 of men with high-risk prostate cancer could have safely avoided long-term ADT based on the results of this predictive biomarker analysis.
This was further demonstrated on Cox multivariable modeling where the ArteraAI MMAI biomarker was shown to significantly predict benefit of long-term ADT addition to radiotherapy with respect to the outcome of distant metastasis (HR: 2.35, 95% CI: 1.72 – 3.19, p=0.04).
A similar biomarker effect was observed for the outcome of death with distant metastases, as demonstrated below:
As expected, the MMAI model was unable to reliable predict the benefit of long-term ADT addition to radiotherapy for the outcomes of metastasis-free and overall survivals. 
How can we translate these results into our practice? Do results of this predictive biomarker supersede those from NCCN risk grouping? In other words, should we rely on one or both for clinical decision making? Based on these results, it appears that 29% of patients with at least one NCCN high/very high-risk feature would benefit from treatment shortening, given the lack of benefits with long-term ADT for the outcomes of distant metastasis and death with distant metastasis. Conversely, it appears that 43% of NCCN intermediate risk patients may benefit from treatment intensification with long-term ADT to reduce the rates of distant metastasis and death with distant metastasis.
Although long-term ADT addition to radiotherapy improves the rates of distant metastasis, ~20% of AI biomarker positive men still experience distant metastasis or death with distant metastasis within 15 years. As such, we need to do a better job to identify these patients and implement early treatment intensification to further improve survival outcomes.
Dr. Armstrong concluded that:
- They have trained and validated the first predictive biomarker to guide ADT duration in locally advanced prostate cancer with validation in a completed phase III randomized trial of radiotherapy plus short- versus long-term ADT.
- The ArteraAI long-term ADT predictive model was successfully developed using prostate biopsy histopathology image data and standard clinical variables
- Approximately 1/3 of men with high-risk prostate cancer could safely avoid long-term ADT based on similar rates of distant metastasis and death with distant metastasis outcomes with short-term ADT
Presented by: Andrew J. Armstrong, MD, Professor, Department of Medicine, Duke Cancer Institute Center for Prostate and Urologic Cancer, Duke University, Durham, NC
Written by: Rashid Sayyid, MD, MSc – Society of Urologic Oncology (SUO) Clinical Fellow at The University of Toronto, @rksayyid on Twitter during the 2023 American Society of Clinical Oncology (ASCO) Annual Meeting, Chicago, IL, Fri, June 2 – Tues, June 6, 2023.
References:
- Lawton CAF, Lin X, Hanks GE, et al. Duration of Androgen Deprivation in Locally Advanced Prostate Cancer: Long-Term Update of NRG Oncology RTOG 9202. Int J Radiat Oncol Biol Phys, 2017;98(2):296-303.