Purpose: To compare the fits of various normal tissue complication probability (NTCP) models to a common set of late rectal toxicity data, with the aim of identifying the best model for predicting late rectal injury after irradiation.
Methods and materials: Late toxicity data from 128 prostate cancer patients treated on protocol with three-dimensional conformal radiotherapy at The University of Texas M.D. Anderson Cancer Center (UTMDACC) were analyzed. The dose-volume histogram for total rectal volume, including contents, was obtained for each patient, and the presence or absence of Grade 2 or worse rectal bleeding within 2 years of treatment was scored. Five different NTCP models were fitted to the data using maximum likelihood analysis: the Lyman model, the mean dose model, a parallel architecture model, and models based on either a cutoff dose or a cutoff volume.
Results: All five of the NTCP models considered provided very similar fits to the UTMDACC rectal bleeding data. In particular, none of the more highly parameterized models (the four-parameter parallel model, three-parameter Lyman model, or three-parameter cutoff dose and volume models) provided a better fit than the simplest of the models, the two-parameter NTCP model describing rectal bleeding as a probit function of mean dose to rectum.
Conclusion: No dose-volume response model has yet been identified that provides a better description of the UTMDACC rectal toxicity data than the mean dose model. Because this model has relatively low predictive accuracy, the need to identify a better model remains.