Evaluation of dose-response models and parameters predicting radiation induced pneumonitis using clinical data from breast cancer radiotherapy

Abstract

The purpose of this work is to evaluate the predictive strength of the relative seriality, parallel and LKB normal tissue complication probability (NTCP) models regarding the incidence of radiation pneumonitis, in a large group of patients following breast cancer radiotherapy, and furthermore, to illustrate statistical methods for examining whether certain published radiobiological parameters are compatible with a clinical treatment methodology and patient group characteristics. The study is based on 150 consecutive patients who received radiation therapy for breast cancer. For each patient, the 3D dose distribution delivered to lung and the clinical treatment outcome were available. Clinical symptoms and radiological findings, along with a patient questionnaire, were used to assess the manifestation of radiation-induced complications. Using this material, different methods of estimating the likelihood of radiation effects were evaluated. This was attempted by analysing patient data based on their full dose distributions and associating the calculated complication rates with the clinical follow-up records. Additionally, the need for an update of the criteria that are being used in the current clinical practice was also examined. The patient material was selected without any conscious bias regarding the radiotherapy treatment technique used. The treatment data of each patient were applied to the relative seriality, LKB and parallel NTCP models, using published parameter sets. Of the 150 patients, 15 experienced radiation-induced pneumonitis (grade 2) according to the radiation pneumonitis scoring criteria used. Of the NTCP models examined, the relative seriality model was able to predict the incidence of radiation pneumonitis with acceptable accuracy, although radiation pneumonitis was developed by only a few patients. In the case of modem breast radiotherapy, radiobiological modelling appears to be very sensitive to model and parameter selection giving clinically acceptable results in certain cases selectively (relative seriality model with Seppenwoolde et al (2003 Int. J. Radiat. Oncol. Biol. Phys. 55 724-35) and Gagliardi et al (2000 Int. J. Radiat. Oncol. Biol. Phys. 46 373-81) parameter sets). The use of published parameters should be considered as safe only after their examination using local clinical data. The variation of inter-patient radio sensitivity seems to play a significant role in the prediction of such low incidence rate complications. Scoring grades were combined to give stronger evidence of radiation pneumonitis since their differences could not be strictly associated with dose. This obviously reveals a weakness of the scoring related to this endpoint, and implies that the probability of radiation pneumonitis induction may be too low to be statistically analysed with high accuracy, at least with the latest advances of dose delivery in breast radiotherapy.