Margriet Kwint

Chapter 7 142 (AUC = 0.685 respectively). Our data is only IMRT based, and the dose heterogeneity due to IMRT, makes the higher dose parameters a better predictor for AET (14-18) compared to IMRT and 3D-CRT data in the study of Palma et al.(17). A significantly lower median V60 dose on the esophagus was found after the dose de-escalation on the pathological lymph nodes. This resulted also in a significantly lower incidence of AET grade ≥2. With the de-escalation of the prescription dose from 66 Gy to 58.08 Gy (60 Gy EQD2) in 24 fractions on the mediastinum (i.e., the esophagus area), we found that the model accuracy of the V50-model was equivalent but the model accuracy of the V60-model was declined (AUC= 0.722 compared to an AUC = 0.624), p=0.09). Qualitatively this can be understood by the realization that the dose parameter that predicts AET is basically the volume of the esophagus receiving a high dose and less related to the specific value of V60. With a dose prescription of 58 Gy, it is expected that the V60 correlates less well with this volume, leading to a decline in model performance. Therefore, the V50-model is more robust in our clinical practice. Since radiation techniques and prescriptions doses differ in time the use of EUD models to predict AET, might be less amendable for these changes, but are not widely used in clinical practice (18). We performed a sample test (N=77) to validate the quality of the electronic toxicity registration. This test showed a sensitivity of 83% and a specificity of 86% for the electronic registration of grade ≥2 AET. The compliance of the toxicity registration was 73-80%. This means that there is missing data. A limitation of this study is this missing data and that the sensitivity and specificity of <100% could have influenced the model accuracy. Comparison with the incidence of AET in the study of Van diessen et. al (22), similar incidences are found. This substantiate that RWD for ETR is reliable. To conclude, the use of real world data provides a useful method for quality assurance and for validation of NTCP-models in clinical practice. Both V50 and V60 NTCP-models showed moderate accuracy to predict acute oesophageal toxicity in NSCLC patients. For clinical practice, the V50Gy seems to be the most stable to the dose de-escalation and sensitive without compromising safety and efficacy.