Radiation Oncology Data and Modelling Side Effects after Radiation Therapy

Doctoral thesis, 2012

Although modern radiation therapy techniques have the ability to conform the dose distribution of ionizing radiation tightly around the volume to be treated, unwanted irradiation to surrounding organs remains a problem. The probability of a side effect arising in the normal tissue of a patient after radiation therapy can be modelled by sigmoid mathematical functions known as normal tissue complication probability (NTCP) models. Using statistical methods, these are fitted to input data representing the absence or presence of a studied symptom associated with the side effect in question and the dose distribution for potentially injured organs. NTCP models are increasingly being used in the clinic both for treatment evaluation and to guide optimization algorithms for inverse treatment planning although their predictions are associated with uncertainties to varying degrees. The purpose of this thesis is to investigate how different means to represent dose and ways to grade side effects contribute to uncertainties in radiation therapy side effect modelling. Using concepts from the literature and results from two recent Swedish studies including data on parotid gland complications in head and neck cancer patients and pubic bone pain in gynaecological cancer patients, respectively, current generations and representations of dose and side effect data and how these are used in modelling side effects are surveyed. Using similarities and differences in the data by the literature and these two studies, it is concluded that dose data for modelling purposes today is described by dose-volume histograms but can be made more detailed using a three-dimensional format like the structure-specific dose matrix as introduced in this thesis with additional information on dose representation. Side effect data are described by different scales to grade the same or similar symptoms, but also need to include information on factors which may influence modelling results such as effect-modifying factors as well as baseline symptom frequencies in non-irradiated individuals. Altogether, 15 items that capture the essential information needed for radiation therapy side effect modelling are identified, and the outlook for data integration and interoperability in radiation oncology would be improved by using these to form a semantic basis for this domain.