Session Item

Radiation induced xerostomia following radiotherapy in the head and neck region is a troublesome late effect affecting patients’ quality of life. Normal Tissue Complication Probability (NTCP) models for predicting xerostomia have been developed, but external validation is needed in order to determine validity and generalizability in patient cohorts not used for model development. Here we present results from the external validation of an NTCP model for predicting patient reported moderate to severe xerostomia in a multicenter cohort. 

A total of 167 patients had completed the EORTC HN35 questionnaire at baseline and 6 months post-radiotherapy where mouth dryness was scored as “not at all” (Xer1), “a little” (Xer2), “quite a bit” (Xer3) and “very much” (Xer4). The predictors in the original NTCP model were the mean dose (D_mean) to the contralateral parotid gland, baseline Xer2 and baseline Xer3-4, respectively (Table I). The model endpoint was moderate to severe xerostomia (i.e. Xer3-4) six months after completion of treatment. For model validation a closed testing procedure was applied. Successively, recalibration in the large (i.e. model intercept update), recalibration and model revision were performed, whereupon predictive performance was tested for significant improvements compared to original model in each step. Bootstrapping (n = 1000) was perform to investigate the robustness of the procedure.

At baseline, Xer1 was reported in 79 patients (47%), Xer2 in 53 patients (32%) and Xer3-4 in 35 patients (21%). Six months post-treatment the prevalence rate of Xer3-4 was 64%. The median (IQR) D_mean to the contralateral gland were 11.5 (5.5-17.1) Gy and 15.0 (6.3-22.4) Gy for those with Xer1-2 vs. Xer3-4 (p = 0.04). In the bootstrap closed testing procedure, calibration in the large was selected 889 times, model revision 82 times, recalibration of the model 28 times and the original model one time. The closed testing showed that by refitting the model intercept, the model matched the current patient cohort well with improved calibration compared to the original model (Table I/ Figure 1).  For patients with EORTC score 1 vs. 2-4 at baseline, median (IQR) estimated risk of Xer3-4 were 0.50 (0.44-0.80) and 0.76 (0.61-0.79). The corresponding observed rates of Xer3-4  in the two groups were 52% and 75%.

Six months after treatment the majority of patients reported moderate to severe xerostomia. The closed testing procedure resulted in an intercept update of the original model. The original model underestimated the risk of moderate to severe xerostomia, whereas the validated model was well-calibrated and had good discriminative ability.