44. Metamodelling of late rectal bleeding in patients undergoing radiotherapy for prostate cancer
Published on Dec 1, 2018in Physica Medica2.485
· DOI :10.1016/J.EJMP.2018.04.054
Purpose To develop a modelling procedure for late rectal bleeding (LRB) that goes beyond maximum likelihood model fitting of data. Thus, to create a metamodel for grade 2–3 (G23) and grade 3 (G3) LRB starting from literature evidence and validate it on a large population. Methods Metamodelling allows to highlight specific features of the model itself. Characteristics that predictive models for LRB have in common are: the global harmony of radiobiological parameters (volume effect parameter, n, steepness of the curve, k, and dose parameter associated to 50% of complication probability, D50) and specific clinical factors that frequently occur. Models including rectal Equivalent Uniform Dose (EUD) with/without patient-related dose-modifying factors (DMF) were retrieved by literature search. Dosimetric coefficients were resolved by weighted mean of published values, using their standard deviation as weight. Identified clinical features, expressed by DMF, were differently weighted taking into account the prevalence of the features and the size of the study. Finally, both factors were inserted in a modified logit-EUD model. Metamodel was validated on a pooled population (3DCRT/IMRT) of three international cohorts. Performance was assessed through calibration. Results Literature search identified rectal EUD, previous abdominal surgery, hormone therapy and use of cardiovascular drugs as relevant features: associated coefficients are presented in Table 1 . Validation cohort included 1591 pts with 240 (15%) LRBG23 and 98 (6.2%) LRBG3 pts. The calibration (Fig. 1) showed the following results: concerning LRBG23 the slope was equal to 0.22 (R2 = 0.38); the corresponding values for LRBG3 were 1.12 and 0.97. Conclusions A metamodel for prediction of LRB was derived from literature. LRBG3 model was successfully validated on a large population proving to be a valuable tool for predicting toxicity before RT. The model for LRBG23 predicted very well toxicity rate below 25% (which involved 87% of the population) while partially failing at higher probabilities.