A multimodal AI (MMAI) digital pathology model was evaluated as a predictive biomarker for abiraterone benefit in 1,137 non-metastatic, clinically very high-risk prostate cancer patients from two STAMPEDE phase 3 trials, randomized to long-term ADT (LT-ADT) alone vs. LT-ADT plus abiraterone; primary endpoint was metastasis-free survival (MFS).
MMAI very high-risk patients (N=268, ~24%) had a large abiraterone benefit (HR 0.47; 95% CI 0.31–0.70; 5-yr MFS 62% → 81%), while MMAI standard high-risk patients (N=869) showed no significant benefit (HR 0.83; 95% CI 0.63–1.09; 5-yr MFS 82% vs. 84%); interaction p=0.02.
- Post-hoc analysis of RCT data limits causal inference and requires prospective validation. - MMAI threshold (75th percentile) was pre-specified but derived from prior work, not this dataset. - Two sequential trials with no shared controls were pooled, introducing potential heterogeneity.
MMAI scoring of diagnostic biopsy pathology could help identify the ~24% of very high-risk localized prostate cancer patients who derive clear MFS benefit from adding abiraterone to LT-ADT, potentially sparing the majority from unnecessary toxicity and cost. Prospective validation is needed before routine clinical use.
Explore related topics
