Risk prediction-modeling approach to identify best predictive parameters for urethral strictures after prostate brachytherapy.

Abstract

87 Background: Urethral strictures (US) are a rare complication of prostate brachytherapy (BXT), with prior studies showing radiation dose to the bulbomembranous urethra is associated with stricture formation. This retrospective case-control study explored clinical and dosimetric parameters associated with the development of BXT-related urethral strictures. Methods: A cohort of 34 patients developed urethral strictures after BXT at our institution for the period of 2008-2014. Each case was matched with two controls (68 controls) that had not developed a US according to similar baseline International Prostate Symptom Score (IPSS), planned prostate volume, post-implant prostate V150, and post-implant prostate D90 dosimetry parameters. US development was compared with clinical (i.e. age, IPSS, etc) and dosimetric (i.e. prostate, urethra, urethra segments) variables. Statistical modeling for risk prediction was applied, which included adjusted R2, Mallows’ C Selection (Cp), Schwartz’s information criterion (BIC), forward selection (FS), and backward selection (BS) to identify the parameters with prediction ability of toxicity. CV analysis was performed to select the best subset selection on the full data set in order to obtain the most predictive parameter selection. Results: The results show that the R2statistic increases from 6% (only one) to 33 %, (all of the parameters included). The table demonstrates minimum best-fit parameters in the different models. (See table.) CV with minimum standard error (MSE) identified a model with 5 parameters that included age, baseline IPSS, UD30, UD5, and U5mm V200Apex shows best prediction ability for US. Conclusions: This modeling approach, which is novel in BXT, helped to identify a combination of parameters with some predictive ability of radiation toxicity. Further evaluation is required to validation. [Table: see text]