ST-DoxPCa (AI “virtual spatial transcriptomics”) identifies an OS-benefiting subgroup for adjuvant docetaxel in high-risk localized prostate cancer

ST-DoxPCa is an AI/computational pathology biomarker built from routine H&E biopsies using a Vision Transformer to infer spot-level expression for a prostate gene panel (distilled to a locked 26-gene signature; includes genes such as PTEN, NKX3-1, ACPP, FASN, TMPRSS2) and aggregate to patient-level risk features. In NRG/RTOG 0521 (Phase 3; high-risk localized prostate cancer treated with RT+long-term ADT ± adjuvant docetaxel; n=350, RT+ADT n=169 vs RT+ADT+DTX n=181), ST-DoxPCa-positive patients had a significant OS benefit from docetaxel (HR=0.38, 95% CI 0.18–0.83; p=0.012), while ST-DoxPCa-negative patients had no OS benefit (HR=1.05, 95% CI 0.67–1.63; p=0.84)—notably aligning with the parent trial’s “modest/limited” unselected benefit signal that attenuated with longer follow-up (i.e., adding docetaxel to RT+ADT has not been a universally durable OS win in the full population).