Margriet Kwint

Real world evidence to audit NTCP-models for acute esophagus toxicity 141 7 Discussion This study illustrates that real world data (RWD), gathered from an electronic toxicity registration (ETR), is a useful method to audit NTCP-models. The showcase, used in this study, exemplifies that with the use of RWD, the accuracy of the AET prediction models were influenced by the dose de-escalation of the involved mediastinal lymph nodes for CCRT in NSCLC patients. The use of digital toxicity registration in electronic medical records is becoming more common practice in many radiotherapy departments. With this approach, a more systematic and accurate recording of radiotherapy toxicity can be achieved (6, 7). With regular adaptations in radiotherapy treatment schedules, techniques or prescription doses it is necessary to assure and maintain treatment quality. The use of RWD simplifies and accelerates quality assurance and provides tools to validate NTCP models in current clinical practice. Of course, (randomized) clinical trials remain a powerful method for developing scientific evidence. Nonetheless, RWD can be used to complement or audit the knowledge gained from traditional clinical trials, and provides additional insights on generalizability of RCT outcomes to the real world population. But to interpret this RWD right, it is important to use adequate analytic approaches to analyze the data and to use the right data-sources to collect the data (1). AET is a common and severe toxicity of CCRT which has a negative influence on quality of life. Therefore, it is important to predict AET to mitigate the risk and to improve the therapeutic ratio of CCRT. The literature available on prediction models for AET and IMRT concluded that the high dose volumes are more important than the low dose volumes (14-19). In our previous work, the V50 was identified as an accurate predictor of AET for IMRT (16). In our hospital, the V50 is currently used as predictor for AET and dose constraint for the treatment planning. The meta-analysis of Palma et al. (17) analyzed the data of 1082 patients and concluded that the V60 has the best predictive ability (3D-CRT and IMRT). In the guidelines for radical radiotherapy from the Advisory Committee on Radiation Oncology Practice (ACROP) of the European Society for Radiotherapy and Oncology (ESTRO), the V60 is advised as dosimetric parameter to predict AET (26). We therefore decided to audit both parameters. AUCs for both the V50-model and the V60-model showed a moderate accuracy. Palma et al. found a lower AUC of 0.583 for the V60-model for grade ≥2 AET compared to ours