Normal tissue complication probability modeling for acute esophagitis in patients treated with conformal radiation therapy for non-small cell lung cancer

Purpose: To evaluate the ability of a well-known normal tissue complication probability (NTCP) model to predict radiation esophagitis by determining updated model parameters and then comparing these results with the predictive value of other dosimetric parameters. Material and methods: Clinical and dosimetric data regarding esophagitis were analyzed in 101 inoperable/unresectable non-small-cell lung cancer patients treated by external beam irradiation. Grade 2 or higher esophagitis counted as events. Parameters (TD50, n, and m) of the Lyman normal tissue complication probability (NTCP) model were determined using maximum likelihood analysis, and compared to other dose/volume threshold values including: percentage of esophagus receiving > 40 Gy (V40) to > 75 Gy (V75), and maximum esophageal doses. Results: Sixteen patients developed grade 2-3 acute esophagitis (no G4 or 5). The maximum likelihood analysis produced new Lyman model parameters of: TD50 = 51 Gy, n = 0.44 and m = 0.32. The mean NTCP value is significantly lower (P<0.001) in the group of patients without esophagitis (13.5%) than with esophagitis (27.2%). The rates of esophagitis are 2.5, 7, 9and 13.4%, respectively, when the NTCP values are <10%, <15%, <20% and <25%. A significant association is found between esophagitis and dose/volume parameters V40 (P=0.001) to V70 (P=0.024). Conclusions: New values of TD50, n and m offer a good description of the esophagitis distribution in our population. Compared to the use of this model with previously published parameters (associated with late toxicity) predictions of the model for acute esophagitis using the new parameters would indicate that, for a population of patients, the distribution of events as a function of uniform dose would occur with a lower mean uniform dose value (smaller TD50), over a wider range of uniform doses (larger m), while also exhibiting a bigger volume effect (larger n). These new parameter values are supported in essence by the correlations found for the dose/volume threshold parameters. (c) 2005 Elsevier Ireland Ltd. All rights reserved.