Cardiac substructure dose signatures as predictors of late cardiovascular events in childhood cancer survivors.

Abstract

12123 Background: Radiation-associated cardiovascular disease (CVD) remains a major late toxicity in childhood cancer survivors. Whole-heart dose may mask heterogeneous substructure exposure, limiting precision in survivorship surveillance and modern radiotherapy planning. Methods: Using the Childhood Cancer Survivor Study (CCSS) open-access cardiac substructure dosimetry dataset (12 mean-dose structures; Gy), we modeled time from cancer diagnosis to CAD, CHF, valvular disease, and arrhythmia with elastic-net Cox (α=0.5; 10-fold CV). Bootstrap stability selection (200 resamples) quantified reproducible substructure predictors; “core” structures were defined by selection frequency ≥0.80. Parsimonious Cox models were fit using core structures (dose scaled per 5 Gy) and adjusted for sex, race, age at diagnosis, anthropometric category, and smoking. Model discrimination was assessed using the concordance index (C-index). Model-based predicted 20-year absolute risks were used to estimate dose thresholds for clinically interpretable risk targets (reference profile: median covariates; modal factor levels). Results: Among 25,481 survivors, events occurred for CAD (n=475), CHF (n=544), valvular disease (n=131), and arrhythmia (n=181). LAD and aorta emerged as reproducible, endpoint-specific predictors: LAD was core across CAD/CHF/valvular outcomes; aorta and LMCA were core for CAD/valvular; pericardium was core for CAD/CHF. In adjusted core models, aortic dose was associated with CAD (HR 1.59 per 5 Gy; 95% CI 1.14–2.21; C-index 0.796) and valvular disease (HR 1.96; 95% CI 1.15–3.32; C-index 0.773). LAD dose was associated with CAD (HR 1.26; 95% CI 1.05–1.53; C-index 0.796) and CHF (HR 1.23; 95% CI 1.02–1.47; C-index 0.818). Absolute risk translation identified interpretable thresholds: predicted 20-year CAD risk crossed ~1% at 13 Gy mean aortic dose and ~2% at 20 Gy; CAD risk crossed ~1% at 24 Gy mean LAD dose. Predicted 20-year CHF risk crossed ~0.5% at 21 Gy mean LAD dose. Predicted 20-year valvular risk crossed ~1% at 23 Gy mean aortic dose. Conclusions: Substructure-specific dosimetry yields reproducible radiation “signatures,” with LAD and aorta dose consistently linked to late CVD. These dose–risk thresholds provide actionable benchmarks for risk-stratified survivorship surveillance and may inform substructure-sparing dose constraints in contemporary radiotherapy planning moving beyond whole-heart dose to endpoint-specific substructure constraints. Substructure-specific dose thresholds for 20-year cardiovascular risk. Outcome Core Substructure HR per 5 Gy (95% CI) 20-Year Risk Target Dose Threshold (Gy) CAD Aorta 1.59 (1.14–2.21) 1% 13 CAD Aorta 1.59 (1.14–2.21) 2% 20 CAD LAD 1.26 (1.05–1.53) 1% 24 CHF LAD 1.23 (1.02–1.47) 0.5% 21 Valvular Aorta 1.96 (1.15–3.32) 1% 23