The use of real-world evidence to audit normal tissue complication probability models for acute esophageal toxicity in non-small cell lung cancer patients

Abstract

IntroductionThe aim of this work is to assess the validity of real world data (RWD) derived from an electronic toxicity registration (ETR). As a showcase, the NTCP-models of acute esophageal toxicity (AET) for concurrent chemoradiation (CCRT) for NSCLC patients were used to validate the ETR of AET before/after dose de-escalation to the mediastinal lymph nodes. Material and methodsOne hundred and one patients received 24 × 2.75 Gy and 116 patients received de-escalated dose of 24 × 2.42 Gy to the mediastinal lymph nodes. The validity and completeness of the ETR was analyzed. The grade ≥2 AET probability was defined according the V50 Gy and V60 Gy NTCP-models from literature. Validity of the models was assessed by calibration and discrimination. Furthermore, sensitivity and specificity for different cut-off points were determined. ResultsThe compliance of ETR was 73–80%, with sensitivity and specificity rates of 83% and 86% for grade ≥2 AET, respectively. Discrimination of both NTCP-models demonstrated a moderate accuracy (V50 model, AUC 0.71; V60-model, AUC 0.69). Dose de-escalation did not influence the accuracy of the V50-model; AUC before: 0.69, and AUC after: 0.71. For the V60-model the model-accuracy decreased after dose de-escalation; AUC before: 0.72 and AUC after: 0.62, respectively. ConclusionRWD is a useful method to audit NTCP models in clinical practice. The NTCP models to predict AET in NSCLC patients showed moderate predictive accuracy. For clinical practice, the V50Gy seems to be most stable for dose de-escalation without compromising safety and efficacy.