ASTRO 2022: Prostate Cancer Risk Stratification in NRG Oncology Phase III Randomized Trials Using Multi-Modal Deep Learning with Digital Histopathology

( 2022 American Society for Radiation Oncology (ASTRO) Annual Meeting held in San Antonio, TX between October 23rd and 26th, 2022 was host to a plenary session on Monday, with Dr. Jonathan Tward discussing prostate cancer risk stratification using multimodal deep learning with digital histopathology, within the context of NRG Oncology phase III trials.


Dr. Tward began his presentation by noting that many decades ago the RTOG group set up a biobank that acted as a repository of clinical, pathologic, and imaging data from numerous trials, including the following five: RTOG 9202, 9413, 9910, 0126, and 9408. Using clinical data from over 5,000 patients with over 16,000 H&E pathologic specimens available, multimodal artificial intelligence (MMAI) was then used to develop a prognostic risk score (MMAI score), that can predict the risk of various oncologic outcomes (5- and 10-year biochemical failure, distant metastasis, prostate cancer-specific and overall mortalities).


ASTRO 2022 ARTERA Jonathan Tward_0 


As has previously been demonstrated by Dr. Osama Mohamad’s group at GU ASCO 2022, this MMAI score has been shown to be more a more accurate prognosticator of oncologic endpoints than NCCN risk groups.


ASTRO 2022 ARTERA Jonathan Tward_1 


Similarly, this MMAI score has been shown by Dr. Daniel Spratt’s group to predict patients that would benefit from concurrent ADT addition to radiation treatments for clinically localized prostate cancer patients


ASTRO 2022 ARTERA Jonathan Tward_2 


Dr. Tward next highlighted the advantages of artificial intelligence tools utilizing digital histopathology:

  1. Performance: prognostic and predctive of treatment benefit
  2. Short turnaround time: seconds of cloud computing
  3. Robustness: Tool developed from thousands of patients
  4. Widespread adoption: No consumption of tumor tissue
  5. Artificial intelligence tools can improve through continiuous adaptive learning


The purpose of the current work to be presented by Dr. Tward was to demonstrate that multi-modal artificial intelligence (MMAI) model could stratify patients into risk groups that more precisely and accurately reflect their prognosis compared to D’Amico/NCC risk groups.


This analysis included 5,569 patients from the 5 RTOG studies (RTOG 9202, 9413, 9910, 0126, and 9408) with a median follow-up of 11.4 years. Patients were stratified into deciles (N=557/decile) based on their MMAI prognostic scores. The risk of metastasis by decile was determined using competing risk regression models. Deciles were grouped into 3 cohorts based on risk of metastasis by 10 years:

  1. MMAI low: <10%
  2. MMAI medium: 10-25%
  3. MMAI high: >25%


These risk groups were subsequently evaluated for 10-year distant metastasis rates.


With regards to baseline patient characteristics, Dr. Tward noted that 90% of included patients were intermediate or high-risk. There was a relatively balanced distribution of patients from the five trials.

ASTRO 2022 ARTERA Jonathan Tward_3 



MMAI declie scores demonstrated that the lowest six deciles were associated with a low risk of distant metastasis (<10%), the next three were associated with a medium risk (10-25%), whereas the highest decile group was the only one associated with a high risk of >25%.


ASTRO 2022 ARTERA Jonathan Tward_4 



How do MMAI and NCCN risk group stratifications correlate? Amongst the 3,342 patients in the MMAI low group, 76% of patients were intermediate risk, with only 16% of patients in this group falling in the NCCN low risk category. In the MMAI medium group of 1,671 patients, 67.7% were NCCN high-risk.


ASTRO 2022 ARTERA Jonathan Tward_5 


But do these results suggest that MMAI risk groups have significantly different risks of metastasis compared to the corresponding NCCN risk groups? It appears that the risk of metastasis in the MMAI and NCCN low groups is actually similar. As demonstrated in the table below, the 10-year risk of distant metastasis in the NCCN and MMAI low risk groups is 3% in both groups. The corresponding rates in the medium risk groups is 12% in the MMAI medium and 6% in the NCCN intermediate. Notably, within the MMAI high group, the risk of 10-year metastasis in radiation-treated patients is 60% in NCCN intermediate risk (n=15) and 36% in NCCN high/very high risk (n=540). This is likely due to differences in the number of patients in each of these subgroups, as well as the fact that 96% of patients in the NCCN high/very high group received concurrent ADT.


ASTRO 2022 ARTERA Jonathan Tward_6 


Given these findings, Dr. Tward next emphasized the difference between prognostic and predictive biomarkers. The MMAI score is a prognostic biomarker that can stratify patients into excellent and poor prognoses. However, these results can help identify patients for whom early treatment intensification is necessary based on prognosis, which then brings into play predictive biomarkers (i.e. those that predict response to treatment), such as the work done by Dr. Spratt looking at predictors of response to ADT.

ASTRO 2022 ARTERA Jonathan Tward_7 



Notably, the MMAI identified 6-fold more patients than NCCN with the lowest risk of metastasis as demonstrated below:


ASTRO 2022 ARTERA Jonathan Tward_8 


One important takeaway from this study is that MMAI identifies a subset of patients (~10%) that a high risk of metastasis >= 30% at 10 years. Dr. Tward emphasized that these are the patients that need early treatment intensification and should be the subject of future clinical trials.


Dr. Tward summarized his presentations with the following take home messages:

  1. AI-derived prognostic biomarkers provide personalized risk estimates, that when grouped allows more streamlined communication
  2. ArteraAI MMAI prognostic tool identifies 6-fold more low-risk patients than NCCN. These patients might consider omitting ADT with RT (NNT >25)
  3. Prognostic biomarkers help with shared decision making to avoid futile treatment intensification in patients who will derive minimal absolute benefit from treatment
  4. Use of AI tools leveraging digital pathology improves prognostication, enabling oncologists to determine an optimal treatment plan for their patients



Presented by: Jonathan Tward, MD, PhD, FASTRO, Professor of Radiation Oncology, Huntsman Cancer Institute Genitourinary Cancers Center, Salt Lake City, UT

Written by: Rashid Sayyid, MD, MSc – Society of Urologic Oncology (SUO) Clinical Fellow at The University of Toronto, @rksayyid on Twitter during the 2022 American Society of Radiation Oncology (ASTRO) Annual Hybrid Meeting, San Antonio, TX, Sat, Oct 22 – Wed, Oct 26, 2022.