Predictive Factors Of Radiation Induced Xerostomia In Lymphoma Patients Receiving Intensity Modulated Radiation Therapy

Purpose:To investigate the incidence and related predictors for radiation xerostomia in patients receiving Intensity Modulated Radiation Therapy(IMRT).To build Normal Tissue Complication Probability(NTCP)models and compare the predictive value of different models.Methods:We retrospectively reviewed clinical charts and radiation records of 271 collected from 271 patients with Hodgkin and Non-Hodgkin lymphoma who received IMRT in the neck and head area.Observer-graded xerostomia scores classified by Common terminology criteria for adverse events version 4.0(CTCAE4.0)and patient-reported xerostomia questionnaire(XQ)scores developed by University of Michigan were recorded by telephone follow-up.Clinical and dosimetric predictors associated with xerostomia were identified in univariate analysis using the Pearson x2 test.Mann-Whitney U test and logistic multivariate regression.Lyman-Kutcher-Burman(LKB)models and decision trees C5.0 were built to predict xerostomia.Model performance was described using Area Under the receiver operating Curve.Results:The mean follow-up time in our cohort was 29 months.185 cases(68.3%)had any xerostomia of any grade.57 cases(21.0%)had>2 Grade acute xerostomia.19 cased(7.0%)had>2 Grade late xerostomia.None had Grade 3 late xerostomia.A cohort of 130 patients rated the xerostomia questionnaire.There was significant difference between acute and late XQ scores(P<0.001).All the dosimetric factors of parotid glands,submandibular glands and major salivary glands were significantly cross-correlated with each other(P<0.001).Mean parotid dose was strongly correlated with mean submandibular dose,r=0.830(P<0.001).Multivariate analysis showed that age,mean parotid gland dose and Submandibular gland V5/V10 were the most important predictors of xerostomia of any grade(P<0.001);xerostomia before treatment and mean parotid gland dose were the most important predictors of ?2 Grade acute xerostomia;mean parotid gland dose were the most important predictors of ?2 Grade late xerostomia.LKB model parameters based on mean parotid gland dose for ?2 Grade acute xerostomia:TD50=33.5 Gy,m=0.48;for>2 Grade late xerostomia:TD50 = 43.2 Gy,m = 0.34.LKB model parameters based on mean major salivary gland dose for>2 Grade acute xerostomia:TD50 =39.4 Gy,m = 0.50;for>2 Grade late xerostomia:TD50 = 55.2 Gy,m = 0.36.For xerostomia of any grade,the multi-dosimetric logistic regression model consisting of age,mean parotid dose and submandibular glands V5/V10 showed higher prediction value than the single-dosimetric predictive models(P<0.05).The best predictive model for>2 Grade acute xerostomia only included mean parotid gland dose.Multivariate logistic regression model show no better AUC than LKB model based on parotid glands,submandibular glands and major salivary glands respectively(P>0.05).The accuracy rate of simplified decision tree to categorize xerostomia of any grade,?2 Grade acute xerostomia,>2 Grade late xerostomia was 76.75,82.29%,94.8%respectivelyConclusion:The incidence of xerostomia in lymphoma patients treated with IMRT was acceptable.Life quality of late stage was significantly better than the acute.As for xerostomia of any grade,mean parotid dose together with V5/V10 for submandibular glands could enhance the predictive value of the multivariate model.The optimal xerostomia model for>2 Grade xerostomia consisted of only the dosimetric factor;there was no significant performance gain using nondosimetric predictors.