Knowledge Engineering–Based Quality Evaluation of RTOG 1308 Photon Treatment Plans

Abstract

To create a knowledge-based mathematic model (RapidPlanTM in Varian Treatment Planning System) using individual anatomic information and dosimetric information from high-quality photon treatment plans submitted to RTOG1308 clinical trial (Phase III randomized trial comparing overall survival after photon versus proton chemoradiotherapy for inoperable stage II-IIIB NSCLC). And to use the model as quality assurance (QA) tool for other submitted photon treatment plans from the same trial. Thirty-four photon treatment plans submitted to RTOG 1308 were initially reviewed by Imaging and Radiation Oncology Core (IROC) RTQA center and scored (score 1 for per protocol, 2 for variation acceptable and 3 for deviation unacceptable). Score ‘1’ plans were used to generate a knowledge base model in Varian Eclipse TPS. All thirty-four plan were re-optimized using the model. Dose volume histograms (DVHs) of the model based re-optimized plans were plotted against the original ones (as submitted by the treating institutions). Tumor control probability (TCP) and normal tissue complication probability (NTCP) were calculated using Equivalent Uniform Dose based Lyman-Kutcher-Burman (LKB) model for clinical efficacy evaluation of the re-optimization. Among six plans that were not compliant to protocol criteria, only one plan was found to be of poor quality by model analysis. Average DVHs comparison and TCP/NTCP calculations show that model aided plan optimization can improve target coverage while achieving similar or slightly better organ at risk (OAR) sparing. The average target equivalent uniform dose was increased from 74.05Gy to 75.8Gy after re-optimization. When 1991 Burman Emami data is used for LKB NTCP calculations, average lung pneumonitis probability decreased from 6.97% to 5.22%. Average Pericarditis probability decreased from 0.004% to 0.001%. Average severe esophugitis probability decreased from 35.05% to 22.47%, and average spinal cord necrosis probability decreased from 0.61% to 0.28% after re-optimization. RapidPlan model was shown to be an efficient tool for RT QA and plan optimization. IROC RTQA has started the process of QA plans submitted to RTOG 1308 using this model. Plan optimization using the model has the potential to improve overall quality and cross-center uniformity.