Sensitivity analysis of dose-response model parameters for the bladder in prostate cancer radiotherapy

Abstract

Introduction

This study evaluates how model parameter values affect dose–response maps (DRMs) in identifying high-risk bladder subregions associated with late urinary toxicities in prostate cancer patients post-radiotherapy.

Methods

Data from 1808 patients were analyzed for five late bladder toxicities. Baseline scores were subtracted from maximum toxicity at 12 and 24 months and dichotomized into grades ≥ 1 and ≥ 2. Bladders were segmented on computed tomography scans, and dose-surface maps (DSMs) were created on 91 × 90 voxel grids using spherical and cylindrical coordinates. Voxel doses were converted to equivalent dose in 2 Gy fractions (EQD2, α/β 1–3 Gy). Welch’s t and Mann–Whitney U equations were applied at each voxel location. Multiple comparisons were corrected via permutation testing (10–10000 iterations), and statistically significant voxels were identified using the 90th and 95th percentiles of Tmax/Umax. Sensitivity of parameters was assessed by varying one parameter at a time, with changes > 400 voxels (∼5% of 8190) classified as large and ≤ 400 as small.

Results

Urinary tract obstruction was the only toxicity significantly associated with bladder DSMs, focusing results on this outcome. After baseline adjustment and dichotomization, event/nonevent counts were 62/701 (grade≥1) and 21/742 (grade≥2; N = 763). DRM results showed large effects of toxicity grade threshold, coordinate system, statistical test equation, and Tmax/Umax thresholding. EQD2 α/β showed variable effects, large for cylindrical and small for spherical coordinates, while the number of permutations had only a small effect.

Conclusions

Parameter selection significantly influences high-risk subregion identification in DRMs, emphasizing the need for standardized parameter reporting for meaningful external comparisons.