Identifying and predicting non-cancer mortality risk in mHSPC following doublet therapy in the STAMPEDE trial.

Abstract

134 Background: Body composition features derived from baseline computed tomography (CT) imaging can reveal clinical phenotypes which inform non-cancer mortality (NCM) risk in M1 HSPC. We measured the prognostic and predictive value of body composition phenotypes for NCM within the STAMPEDE Docetaxel (DOC) and Abiraterone (AAP) comparisons. Methods: A fully automated image analysis programme quantified body composition features from baseline CT scans in M1 patients recruited to STAMPEDE’s DOC and AAP comparisons. Random Forest feature importance identified six key imaging biomarkers: abdominal aortic calcium, bone mineral density, visceral adipose tissue, visceral adipose tissue: subcutaneous adipose tissue ratio, skeletal muscle area, and intramuscular adipose tissue. Unsupervised Gaussian mixture modelling defined three distinct phenotypic clusters. Fine-Gray competing-risk models, adjusted for age, WHO performance status, treatment, NSAID use, PSA, and Gleason grade group, evaluated the prognostic utility of cluster phenotypes for NCM. Predictive models incorporated cluster × treatment interaction terms to assess effect modification by therapy. Flexible parametric models generated 10-year cumulative incidence. Results: Among 937 men (docetaxel comparison: n=534; abiraterone comparison: n=553; n=150 contemporaneously recruited), median follow-up of 7.6 years, three distinct body composition phenotypes were identified: cardiovascular (high calcification, sarcopenia; n=163), reference (preserved muscle and bone; n=543), and metabolic (myosteatosis, visceral adiposity; n=231) phenotypes. Both adverse phenotypes predicted higher NCM versus the reference phenotype after adjustment for treatment and clinical factors: CV phenotype SDHR=2.08 (95% CI 1.23-3.52; p=0.006) and metabolic phenotype SDHR=1.75 (1.07-2.87; p=0.026). Age predicted NCM independently (SDHR=1.07 per year; p<0.001). A significant phenotype×AAP interaction revealed excess NCM risk in the metabolic phenotype (SDHR=3.78, 1.25-11.4; p=0.019). Ten-year NCM exceeded 20% in high-risk phenotypes versus <10% in the reference phenotype. Conclusions: CT body composition phenotypes from routine staging scans independently predict NCM following combination therapies in mHSPC identifying patients with metabolic phenotypes who experience excess mortality risk with AAP intensification. These findings support the role of imaging biomarkers in guiding personalised risk stratification and treatment selection.