NTCP Modelling for Dermatitis Among Photon- and Proton-Treated Nasopharyngeal Carcinoma Patients
Abstract
Purpose
To develop normal tissue complication probability (NTCP) models for Grade 3 (G3) radiation dermatitis in nasopharyngeal carcinoma (NPC) patients treated with photon and proton therapy, and to evaluate their generalizability across the two treatment modalities.
Methods
Our data included 106 photon patients and 110 proton patients, treated for NPC in our institution from 1 Jan 2024 to 31 Dec 2025. On top of clinical characteristics, the dose-volume histogram of skin contour for each patient was extracted from the treatment planning system. Dermatitis was graded weekly since the start of treatment, up to 6-7 weeks. Two NTCP models were constructed with G3 dermatitis as the endpoint: 1) logistic regression (LR) model, and 2) Lyman-Kutcher-Burman (LKB) model. Leave-one-out cross-validation (LOOCV) was done within the photon and proton cohort, and the model trained on photon cohort was also tested on the proton cohort. The calibration plot and Receiver Operating Characteristic (ROC) curve were plotted, and the area under the ROC curve (AUC) was calculated.
Results
The LR model identified V45Gy to be highly predictive of >G2 toxicity. From the LKB model, the dose causing a 50% toxicity risk was found to be 50.0 Gy and 49.9 Gy for photon and proton cohort, respectively. LOOCV within photon and proton cohort individually achieved AUC of 0.53 to 0.85 for both LR and LKB models. LKB model trained on photon was found to be generalizable to the proton cohort (AUC = 0.86 [0.80-0.92]), but not so for LR model (AUC = 0.20 [0.12-0.28]).
Conclusion
In this work, two NTCP models have been constructed for G3 dermatitis in NPC patients treated with photon and proton therapy. The dosimetric features identified from the modelling could serve as clinical goals to be considered during treatment plan optimisation and evaluation, to minimise the occurrences of radiation dermatitis.