Independent external validation using the EORTC HNCG-ROG 1219 DAHANCA trial data of NTCP models for acute oral mucositis

• Published in: Radiotherapy and oncology Vol. 161; pp. 35 - 39
• Main Authors: Sharabiani, M., Clementel, E., Andratschke, N., Collette, L., Fortpied, C., Grégoire, V., Overgaard, J., Willmann, J., Hurkmans, C.
• Format: Journal Article
• Language: English
• Published: Ireland Elsevier B.V 01.08.2021
• Subjects: External validation; Logistic models; NTCP; Oral mucositis; Oral mucositis; External validation; Logistic models; NTCP
• ISSN: 0167-8140; 1879-0887
• DOI: 10.1016/j.radonc.2021.04.006

•Six NTCP models on oral mucositis were externally validated.•Measures of calibration and discrimination were assessed.•Chemotherapy schedule does not affect the incidence of acute oral mucositis.•Mean dose-based logistic models remain valid in diverse treatment techniques. To externally validate previously published Normal Tissue Complication Probability (NTCP) models developed by separate teams for grade 3 oral mucositis (g3OM). Two models were validated: a logistic model, based on 144 head and neck cancer (HNC) patients receiving induction chemotherapy followed by chemo-IMRT; a multivariable logistic model for prediction of g3OM for 253 patients receiving radical treatment for the head and neck squamous cell carcinoma (HNSCC). The EORTC HNCG-ROG 1219 DAHANCA trial dataset, consisting of 169 patients was used as the validation cohort. This cohort was treated with accelerated fractionated chemo-IMRT, with/without the hypoxic radiosensitizer Nimorazole for HNSCC. External validity was assessed using the scaled Brier score. Calibration was assessed in terms of calibration curves as well as measures of mean and weak calibration. Hosmer-Lemeshow was used for goodness-of-fit test. Discrimination was calculated using the area under the receiver operating curve (AUC-ROC). The prevalence of g3OM in the validation cohort (35.5%) was similar to that of two development cohorts, i.e. 38.7% and 31.9% for Bhide logistic and Otter multivariable logistic models respectively. The scaled Brier scores showed good overall model performance. Perfect calibration was observed in the prevalence range of 20% to 40%. AUC-ROC was acceptable in external validation (0.67). The Hosmer-Lemeshow test showed good agreement between predicted and observed outcomes for two models. The NTCP models were validated and lead to valid predictions in a wide range of diverse treatment techniques and patient characteristics, also when Nimorazole is added as hypoxic radiosensitizer.