3d-Crt Based Predictive Models For Patient-Rated Xerostomia And Sticky Saliva Are Not Valid Among Patients Treated With Imrt

The main objective of the current study was to investigate the external validity of 3D-conformal radiotherapy (3D-CRT) based predictive models for patient-rated xerostomia (XER6M) and sticky saliva (STIC6M) at 6 months after completion of primary chemoradiation (CHRTprim) among patients treated with intensity modulated radiotherapy (IMRT) for head and neck cancer (HNC). A study population of 167 consecutive HNC patients treated with 3D-CRT and (CH)RTprim was used as training dataset to develop multivariate logistic regression predictive models using a bootstrap technique. A population consisting of 162 patients treated with IMRT and (CH)RTprim was used as external validation dataset. External validity was described by the explained variation (R2 Nagelkerke) and the Brier score. The discriminative abilities of the models were calculated using the area under the Receiver Operating Curve (AUC) and calibration (the agreement between predicted and observed outcome) was quantified using the Hosmer-Lemeshow “goodness-of-fit” test. The multivariate analysis showed the mean parotid dose, age, and baseline xerostomia (none vs. a bit) to be the most important predictors for XERm6 in the 3D-CRT training dataset. For STICm6, the mean submandibular dose, age, the mean sublingual dose, and baseline sticky saliva (none vs. a bit) were most predictive for STICm6 in the 3D-CRT dataset. Model performance of these models was good with an AUC of respectively 0.82 and 0.84. As compared to the 3D-CRT cohort overall model performance of the 3D-CRT based predictive models for XER6M and STIC6M was significantly worse in terms of the Brier score and Nagelkerke R2 in the IMRT cohort. The AUC for both 3D-CRT based models in the IMRT treated patients were markedly lower (0.66 and 0.64, respectively). The Hosmer-Lemeshow test showed a significant disagreement between predicted risk and observed outcome (Table). The 3D-CRT based models for patient-rated xerostomia and sticky saliva turned out to be not sufficiently valid among patients treated with IMRT. The main message from these findings is that models developed in a population treated with a specific technique cannot be generalized to a population treated with another technique without validation.TablePerformance of the predictive models for XER6m and STIC6m in training and validation datasetPerformance measureXER6M 3D-CRTXER6M IMRTSTIC6M 3D-CRTSTIC6M IMRTBrier (scaled)0.3130.0460.339-0.003R2 Nagelkerke0.4000.0460.4420.082Area under the curve0.824 (0.761 – 0.866)0.657 (0.573 – 0.740)0.840 (0.782 – 0.898)0.644 (0.555 – 0.733)Hosmer-Lemeshow testChi-square = 7.6 (p = 0.47)Chi-square = 17.41 (p = 0.04)Chi-square = 5.16 (p = 0.74)Chi-square = 30.3 (p = 0.04) Open table in a new tab