Session Item

Osteoradionecrosis (ORN) of the mandible is an infrequent, yet a severely debilitating toxicity following head and neck cancer (HNC) radiotherapy. The first ORN Normal Tissue Complication Probability (NTCP) models for any grade (ORN_I-IV) and grade IV (ORN_IV) have recently been developed on a cohort of 1259 patients. The aim of this study was to externally validate these recently published multivariable NTCP models in a large HNC patient cohort treated with (chemo-)radiotherapy. 

The validation cohort consisted of 1154 HNC patients treated with definitive (chemo)radiotherapy. Patients who underwent a mandibulectomy or had previous radiotherapy in the HN area were excluded. Both cohorts used the ORN grading system published by Tsai et al. who defined grade 4 ORN as ORN requiring major invasive surgery. The original (logistic regression) NTCP models identified D_30% of the mandible and pre-radiotherapy (pre-RT) dental extraction as independent factors for both ORN_I-IV and ORN_IV. The original NTCP models were externally validated and the performance was described with the AUC and R². Additionally, the closed testing procedure (CTP) was performed to test which model update was appropriate in our cohort: 1) by calibration-in-the-large (re-estimation of model intercept), 2) recalibration (re-estimation of intercept and slope) or 3) model revision (re-estimation of all coefficients).

The prevalence of ORN_I-IV and ORN_IV in the validation cohort was 4% (n=44) and 1% (n=9), respectively, whereas the prevalence of ORN_I-IV and ORN_IV in the development cohort was 13.7% and 5%. Similar as in the published paper, all DVH parameters of the mandible were significantly associated to ORN with univariate analyses. The external validation showed a similarly good and comparable performance for the prediction of ORN_I-IV and ORN_IV (Table 1). Additionally, the CTP did not detect major differences in the validation cohort but recommended a re-estimation of the model intercept. The intercept decreased from -6.85 to -8.90 for ORN_I-IV and from -9.16 to -11.67 for ORN_IV, while the coefficients of D_30% and dental extraction remained the same. Please see figure 1 for the NTCP models (validation and previously published NTCP model). 

The previously published ORN NTCP models with D_30% of the mandible and pre-RT dental extraction as independent factors were externally validated and lead to similarly good and comparable performance for the prediction of ORN_I-IV and ORN_IV. The CTP suggested a minor intercept update in order to match our cohort, most likely due to the lower ORN prevalence in this cohort. An explanation for this difference is subject of further investigation. Overall, these models can be used to optimize radiotherapy treatment planning, including patient selection for proton or photon therapy, aiming at prevention of mandibular osteoradionecrosis.