Inter-institutional analysis demonstrates the importance of lower than previously anticipated dose regions to prevent late rectal bleeding following prostate radiotherapy

Abstract

PurposeTo investigate whether inter-institutional cohort analysis uncovers more reliable dose–response relationships exemplified for late rectal bleeding (LRB) following prostate radiotherapy. Material and methodsData from five institutions were used. Rectal dose–volume histograms (DVHs) for 989 patients treated with 3DCRT or IMRT to 70–86.4 Gy@1.8–2.0 Gy/fraction were obtained, and corrected for fractionation effects (α/β = 3 Gy). Cohorts with best-fit Lyman–Kutcher–Burman volume-effect parameter a were pooled after calibration adjustments of the available LRB definitions. In the pooled cohort, dose–response modeling (incorporating rectal dose and geometry, and patient characteristics) was conducted on a training cohort (70%) followed by final testing on the remaining 30%. Multivariate logistic regression was performed to build models with bootstrap stability. ResultsTwo cohorts with low bleeding rates (2%) were judged to be inconsistent with the remaining data, and were excluded. In the remaining pooled cohorts (n = 690; LRB rate = 12%), an optimal model was generated for 3DCRT using the minimum rectal dose and the absolute rectal volume receiving less than 55 Gy (AUC = 0.67; p = 0.0002; Hosmer–Lemeshow p-value, pHL = 0.59). The model performed nearly as well in the hold-out testing data (AUC = 0.71; p < 0.0001; pHL = 0.63), indicating a logistically shaped dose–response. ConclusionWe have demonstrated the importance of integrating datasets from multiple institutions, thereby reducing the impact of intra-institutional dose–volume parameters explicitly correlated with prescription dose levels. This uncovered an unexpected emphasis on sparing of the low to intermediate rectal dose range in the etiology of late rectal bleeding following prostate radiotherapy.