Calibrating and benchmarking absolute malignant induction probabilities

Conference Paper

Madkhali, A . 2016

• نوع عمل المنشور: Poster
• الوسوم: Secondary cancer risk, radiotherapy
• اسم المؤتمر: International Conference on the use of Computers in Radiation Therapy
• عنوان المؤتمر: London, UK
• تاريخ المؤتمر: الاثنين, حزيران (يونيو) 27, 2016
• المنظمة الراعية: The Institute of Cancer Research, The Royal Marsden, University of Oxford

مستخلص المنشور

Introduction
More than half of cancer patients receive radiotherapy for radical or palliative purposes. Increasing survival rates in cancer patients make it important to study late side-effects, including secondary radiation-induced cancers. Although a number of predictive models exist, the absolute accuracy of these models in the radiotherapy dose range is limited partly due to scarcity of data and partly by extrapolation beyond historical data bounds. The aim of this work is to investigate conversion of malignant induction probabilities (MIP), which provide useful relative risk estimates, into absolute life time attributable risk estimates (LAR) and excess absolute risk (EAR) by calibrating and benchmarking our models using published outcome data.
Materials & Methods
An in-house modelling tool, which calculates voxel-wise risk estimates from patient-specific 3D dose distributions, was modified to generate linear-no-threshold (LNT) model-based risk estimates for the whole body and per organ using organ-equivalent dose. Second cancer risk was calculated for uniform whole-body exposure of 0.1 Gy for comparison with tabulated BIER VII data[1]. Model parameters initially used were taken from existing published reports for the relevant models. The calculated LAR was then compared to the BIER VII results and the linear coefficient, λ, was adjusted to make the model prediction better match the BEIR VII results. The calibrated λ was then used to calculate LAR and EAR for a 3DCRT plan and an actively scanned proton therapy plan for a case of adult medulloblastoma (MB). A similar calibration of parameters was performed for the linear quadratic (LQ) and linear model (LIN) malignant induction models.
Results
Initial calculations of LAR for single uniform exposure of 0.1 Gy produced an LAR of 1683 cases per 100,000 for an exposure at age 40, in comparison to 824 cases per 100,000 according to BIER VII report (figure 1.a). After calibration, calculations of LAR for single uniform exposure of 0.1 Gy produced a value of 837 cases per 100,000 for an exposure at age 40. Averaging over ages at exposure of 20 to 80 produced a value within 5% of the BIER VII report (figure 1.b). Calibration was also done for LIN and LQ model and improved fits were produced (data not shown). EAR was calculated for 3DCRT plan and an actively scanned proton therapy plan. 3DCRT was predicted to have higher EAR than actively scanned protons (figure 1.c), which reinforces previous relative modelling results of MIP.
 
Discussion & Conclusions
Our results show that our models can produce absolute lifetime attributable risk estimates for secondary cancer which are consistent with the values reported in the BEIR VII report for uniform irradiation to 0.1 Gy. The models can then be used to predict LAR for voxelized model taking into consideration heterogeneity in both the radiosensitivity and the 3D dose distribution.