Spatial transcriptomics–guided computational pathology model to stratify docetaxel benefit in metastatic hormone-sensitive prostate cancer: CHAARTED trial (ECOG-ACRIN E3805).

Abstract

5002 Background: The CHAARTED trial demonstrated improved overall survival (OS) with the addition of docetaxel (DTX) to androgen deprivation therapy (ADT) in metastatic hormone-sensitive prostate cancer (mHSPC), but many patients do not benefit and experience significant toxicity. Clinical factors (e.g., metastatic burden) are suboptimal for predicting DTX benefit, indicating a need for better biomarkers. We developed ST-DoxPCa (Spatial Transcriptomics–Guided Docetaxel Therapy Stratification in Prostate Cancer), an artificial intelligence (AI)-driven biomarker that infers intratumoral gene expression from routine H&E slides (via virtual spatial transcriptomics) to stratify patients by likelihood of benefiting from DTX. Methods: ST-DoxPCa was developed using paired H&E and spatial transcriptomics from the HEST1K prostate cohort to train a Vision Transformer–based regressor to predict a 208-gene panel at spot/patch level; predictions were distilled into a biologically informed 26-gene signature and summarized into patient-level features capturing expression level and heterogeneity. In CHAARTED, a Cox model for overall survival (OS) was trained and internally validated in the ADT-only arm (n = 154; 50/50 split: n = 77 train, n = 77 validation) to derive an ST-DoxPCa risk score and fixed median threshold. This locked model was then applied to the ADT validation cohort (n = 77) and the held-out ADT+DTX arm (n = 129), and OS was compared between treatment arms within biomarker-defined risk groups using Kaplan–Meier estimates and Cox hazard ratios. Results: ST-DoxPCa stratified patients into biomarker-defined risk groups using aggregated spatial-expression features from a biologically informed gene panel; key genes included AR, PTEN, FASN, FOXA1, and ACPP. In ST-DoxPCa-positive (high-risk) patients, adding DTX to ADT significantly improved OS versus ADT alone (HR = 0.53, 95% CI 0.31–0.90; p = 0.018), with no benefit in ST-DoxPCa-negative (low-risk) patients (HR = 1.32, 95% CI 0.74–2.34; p = 0.34). Conclusions: ST-DoxPCa is a novel spatial transcriptomics guided histology biomarker that predicts differential benefit from docetaxel in mHSPC. ST-DoxPCa identified a subset of patients who achieve substantial survival gains from upfront DTX and another subset with no benefit, enabling personalized treatment intensification. ST-DoxPCa has the potential to spare low-risk patients unnecessary chemotherapy while directing therapy to those most likely to benefit. Clinical trial information: NCT00309985 .