Development of normal tissue complication probability models for radiotherapy induced xerostomia factoring in sub-volumes of the parotid glands

Radiotherapy is a well established treatment for head and neck cancer. Unfortunately, due to the incidental irradiation of nearby organs at risk, patients may experience a number of side effects. One of the more prominent is radiation induced xerostomia (the feeling of a dry mouth). Xerostomia has been shown to severely impact on patient quality of life, and given its subjective nature, it is understood to be a complex end-point, influenced by many parameters. Variation across the salivary glands, both in terms of quantity and consistency of produced saliva, means there are a number of factors to consider when attempting to improve rates of xerostomia. Parotid-sparing intensity modulated radiotherapy is established as the standard of care for treating these patients. Even with advances in treatment, a significant proportion of patients still suffer with long term xerostomia, necessitating further improvements in both the prediction and prevention of this condition. Focussing on the prediction of xerostomia, there are a number of published prediction models, however these studies suffer from a lack of external validation. In addition there is scope for further improvement of these prediction models, given advances in the understanding of functionally important sub-volumes of the salivary glands. This thesis begins by presenting the independent external validation of an existing normal tissue complication probability (NTCP) model (published by Beetz et al., 2012). Applying the model to a cohort of local patients, the ability to predict moderate-to-severe patient reported xerostomia was assessed. The model was found to offer some discrimination between those with and without moderateto- severe xerostomia, however the measured area under the curve (AUC) of 0.59 was lower than the published AUC of 0.68. Using the validated model as a basis, this thesis then goes on to present the development of two new NTCP models. Developed using a cohort of local patients, these models were created both with and without the inclusion of the individual parotid lobes, which were hypothesised to represent functionally important areas of the parotid glands. The models were created using the same methodology as the previously validated model, which has been recently published. A comparison of NTCP models produced with and without the inclusion of the parotid sub-volumes found no significant difference in performance. Following internal validation, the performance of each model was also comparable to that of the published model (when applied to the same cohort of patients). It was concluded that no additional benefit was observed when including the lobes of the parotids into NTCP model development, however the work did identify a strong correlation between smoking status and moderate-to-severe xerostomia. In addition, the challenges of NTCP model development when a cohort of patients all have little variation in dosimetric parameters (due to treatment plan standardisation) were highlighted.