Influence Of Dose/Volume Parameters, Equivalent Uniform Dose And Linear Quadratic Model On Prediction Power Of Ntcp Model For Imrt: Based On The Analysis Of Brain Injury

To investigate the influence of dose/volume parameters, equivalent uniform dose (EUD) and linear quadratic (LQ) model on prediction power of normal tissue complication probability (NTCP) model for intensity-modulated radiation therapy (IMRT) and to explore preferable predictors. Seventy-four nasopharyngeal carcinoma (NPC) patients who had developed brain injury (BI) and 146 patients 5-year-free of BI after IMRT were retrospectively assessed. Dosimetric parameters of left and right half-brains were exported from treatment planning system. Four hundred and forty sets of dosimetric features were collected, including doses covering certain volume (DV), doses per fraction(dpf) covering certain volume (dV), absolute volumes receiving specific dose (VD), absolute volumes receiving specific dpf (Vd), EUDs with the a value ranged from 1 to 20, EUDs of certain volume receiving specific dose (EUDVD), EUDs of certain volume receiving specific dpf (EUdVd), and biological equivalent dose (BED) with the α/β ratio ranged from 0.01 to 10 Gy. NTCP were analyzed by receiver operating characteristic curve (ROC) and Logistic regression. Area under the curve (AUC) of DV increased with volume and reached maximum (D1cc, 0.782) between 0.9 cm3 and 2 cm3 then decreased. AUC of D1cc increased as the α/β ratio decreased and reached maximum (0.797) when α/β<1Gy. AUC of d1cc (0.805) was larger than any AUC of D1cc. AUC of EUD increased with a value and reached maximum plateau (0.798) when a = 12-15 then decreased (α/β = 3Gy). AUC of EUD increased as the α/β ratio decreased and reached maximum (0.803) when α/β≤1Gy. AUC of EUd (0.808) was larger than any AUC of EUD. The optimal value of a gradually decreased as the reference volume received specific dose decreased. The largest AUCs of EUDVD (α/β = 1Gy) was found in EUDV45Gy, EUDV50Gy and EUDV55Gy (0.812). The largest AUCs of EUdVd was found in EUdV1.6Gy, EUdV1.7Gy and EUdV1.8Gy (0.818). AUC of VD(α/β = 1Gy) /Vd increased with dose and reached maximum at VD75Gy (0.805)/Vd2.23Gy (0.815) then decreased. Cut off points of EUdV1.7Gy (a = 8), Vd2.23Gy and d1cc were 1.988Gy, 0.897cm3 and 2.212Gy respectively, and tolerance dose/volume of 50% probability were 2.091Gy, 6.1cm3 and 2.413Gy respectively. DV (such as D1cc) has more poor prediction power than VD and EUD. AUC increase with the decrease of α/β ratio and reach maximum when α/β≤1Gy. EUD could further improve the accuracy of prediction models, and when certain low-dose volume is removed, the appropriate value of a decreases and the AUC increases. In long course radiotherapy, dpf could replace the uncertain BED when the difference in fraction numbers is less. Under such conditions EUdV1.7Gy is a preferable predictor of BI, and Vd2.23Gy or d1cc is a simply and substitutable predictor.