Headline: Artificial intelligence using digital histopathology provides better prognostication than NCCN risk categories for men with localized prostate cancer.
The Study: What’s the least popular part of most prostate radiation regimens? ADT. For years we’ve sought to better individualize recommendations for the addition of ADT, an imminent need highlighted by several headlining studies of ASTROs past such as critical looks at 9601 and 0815. This year offers a solution with multimodal artificial intelligence (MMAI) based on deep learning platforms that incorporate clinical data and digitized prostate biopsy slides. The MMAI 5-year metastasis model was used to categorize 5569 enrollees in five phase 3 RTOG trials assessing the benefit of adding various ADT regimens to definitive radiation (9202, 9408, 9413, 9910, and 0126). According to the NCCN classification, 10% were low risk, 55% were intermediate risk, and 35% were high risk. The MMAI model classified 60% as favorable, 30% as moderate, and 10% as unfavorable risk. So a lot of downstaging going on here. The 3 risk groups identified by the MMAI model showed a broader range of 10-year distant metastasis risk (3%, 12%, and 37%) compared to the 3 NCCN groups (3%, 6%, 17%). Put another way, 83% of NCCN intermediate risk patients and 13% of high risk patients were identified as MMAI-favorable with a <5% risk of DM at 10 years. At the same time, MMAI identifies a subset of patients with a substantially higher risk of DM than the spectrum of disease we currently call high risk.
TBL: The MMAI deep learning model provides a more accurate prognostic assessment of DM risk than the current NCCN risk categories.
Citation(s)
- Tward, ASTRO 2022