| Modeling of volume effects for treatment toxicity is paramount for optimization of radiation therapy.; This thesis proposes a new model for calculating volume effects in gastro-intestinal and genito-urinary normal tissue complication probability (NTCP) following radiation therapy for prostate carcinoma.; The radiobiological and the pathological basis for this model and its relationship to other models are detailed. A review of the radiobiological experiments and published clinical data identified salient features and specific properties a biologically adequate model has to conform to.; The new model was fit to a set of actual clinical data. In order to verify the goodness of fit, two established NTCP models and a non-NTCP measure for complication risk were fitted to the same clinical data.; The method of fit for the model parameters was maximum likelihood estimation. Within the framework of the maximum likelihood approach I estimated the parameter uncertainties for each complication prediction model. The quality-of-fit was determined using the Aikaike Information Criterion. Based on the model that provided the best fit, I identified the volume effects for both types of toxicities. Computer-based bootstrap resampling of the original dataset was used to estimate the bias and variance for the fitted parameter values.; Computer simulation was also used to estimate the population size that generates a specific uncertainty level (3%) in the value of predicted complication probability. The same method was used to estimate the size of the patient population needed for accurate choice of the model underlying the NTCP.; The results indicate that, depending on the number of parameters of a specific NTCP model, 100 (for two parameter models) and 500 patients (for three parameter models) are needed for accurate parameter fit.; Correlation of complication occurrence in patients was also investigated. The results suggest that complication outcomes are correlated in a patient, although the correlation coefficient is rather small. |
