ESTRO 2021 Abstract Book

S607

ESTRO 2021

Purpose or Objective Photon-derived NTCP models have been used in clinical practice in the Netherlands to select head and neck cancer (HNC) patients for intensity modulated proton therapy (IMPT). The aim of this study was to externally validate photon-derived NTCP models for acute toxicity in HNC patients treated with IMPT. Materials and Methods This prospective cohort study included 135 HNC patients who were likely to benefit from proton therapy and were selected for IMPT in accordance with the criteria of the Dutch protocol. Published photon-derived NTCP models for 4 endpoints at 6 acute toxicity time points were assessed. Endpoints were grade 2-4 and grade 3- 4 physician-rated dysphagia (CTCAE v4.0), and moderate-severe and severe patient-rated xerostomia (EORTC QLQ-H&N35). Time points were week 3 to 7 (W3-W7) during treatment and approximately 6 weeks (range 5-7) after treatment (6W post). All endpoints were scored prospectively. Average NTCP predictions were compared with the observed prevalence and differences were tested for statistical significance (p ≤ 0.05). Model performance was evaluated with the area under the receiver operating characteristic curve (AUC), calibration in the large and calibration slope. In addition, the expected AUC distribution, based solely on the NTCP predictions, was calculated for each endpoint using a Monte Carlo simulation (1000 simulations). This expected AUC is an estimate of the highest AUC that can reasonably be obtained for this cohort and can serve as a reference for the observed AUC. Results The 135 patients had a mean age of 63 years. Most patients had an oropharyngeal tumor (54.8%), with a tumour stage T2-4 (63%) and positive lymph nodes (80.7%). They were mainly treated with bilateral neck irradiation (97%) and chemoradiation (45.2%). The average NTCP predictions did not significantly differ from the observed prevalence for all endpoints at all time points (p-value >0.05), except for grade 2-4 dysphagia at W7 (predicted 81% versus observed 72%, p-value 0.001), and moderate-severe xerostomia at 6W post (predicted 38% versus observed 47%, p-value 0.025) (Figure 1). When correcting for multiple testing these deviations were not significant. The median AUC was 0.69 (interquartile range (IQR): 0.67-0.72). The median calibration in the large and calibration slope were -0.05 (IQR: -0.29-0.07) and 0.86 (IQR: 0.75-1.00), respectively (Table 1).

Conclusion Photon-derived NTCP models for acute dysphagia and xerostomia were validated in HNC patients treated with proton therapy. Overall, the NTCP models had a good performance and predicted well with only minor deviations from the observed rates.