Voxel-wise association of planned dose with measures of treatment failure and toxicity in prostate external beam radiotherapy

This study aimed to identify anatomically-localised associations between radiotherapy dose and measures of treatment failure, gastrointestinal toxicity and genitourinary toxicity outcomes. Planned dose distributions for 686 patients from the RADAR prostate radiotherapy trial were deformably registered onto common exemplar CT templates. Voxel-wise permutation dose-difference testing, Cox regression modelling and LASSO variable selection were used to identify regions where dose-variation was associated with the above treatment outcomes. Results were validated on datasets from the CHHiP and RT01 trials. Voxel-wise tests identified several dose-outcome associations localised to pelvic anatomy and related to treatment technique, both confirming and confounding previously established relationships.