Modeling of Acute Rectal Toxicity to Compare Two Patient Positioning Methods for Prostate Cancer Radiotherapy: Can Toxicity Modeling be Used for Quality Assurance?

Abstract

Purpose: Intensity Modulated Radiation Therapy (IMRT) allows for significant dose reductions to organs at risk in prostate cancer patients. However, the accurate delivery of IMRT plans can be compromised by patient positioning errors. The purpose of this study was to determine if the modeling of grade ≥ 2 acute rectal toxicity could be used to monitor the quality of IMRT protocols.

Materials and methods: 79 patients treated with Image and Fiducial Markers Guided IMRT (FMIGRT) and 302 patients treated with trans-abdominal ultrasound guided IMRT (USGRT) was selected for this study. Treatment plans were available for the FMIGRT group, and hand recorded dosimetric indices were available for both groups. We modeled toxicity in the FMIGRT group using the Lyman Kutcher Burman (LKB) and Univariate Logistic Regression (ULR) models, and we modeled toxicity in USGRT group using the ULR model. We performed Receiver Operating Characteristics (ROC) analysis on all of the models and compared the Area under the ROC curve (AUC) for the FMIGRT and the USGRT groups.

Results: The observed Incidence of grade ≥ 2 rectal toxicity was 20% in FMIGRT patients and 54% in USGRT patients. LKB model parameters in the FMIGRT group were TD₅₀=56.8 Gy, slope m=0.093, and exponent n=0.131. The most predictive indices in the ULR model for the FMIGRT group were D_25% and V_{50 Gy}. AUC for both models in the FMIGRT group was similar (AUC=0.67). The FMIGRT URL model predicted less than a 37% incidence of grade ≥ 2 acute rectal toxicity in the USGRT group. A fit of the ULR model to USGRT data did not yield a predictive model (AUC=0.5).

Conclusion: Modeling of acute rectal toxicity provided a quantitative measure of the correlation between planning dosimetry and this clinical endpoint. Our study suggests that an unusually weak correlation may indicate a persistent patient positioning error.