Can NTCP Models Derived From 3D-CRT Data Be Used to Predict Gastrointestinal Toxicities for IMRT Prostate Cancer Patients?
A Troeller1,2*, M Soehn2, D Yan1,(1) William Beaumont Hospital, Royal Oak, MI, (2) Ludwig-Maximilians-Universitaet, MunichTU-G-108-9 Tuesday 4:30PM - 6:00PM Room: 108
Purpose: It is unclear whether NTCP models derived from one treatment modality are applicable to patients receiving a different treatment technique. In this large prostate cancer patient cohort study, we compare the model parameters of patients treated with 3D-CRT vs. IMRT, for EUD-based and a dose-volume-based NTCP model.
Methods: Rectal-wall DVHs of 1116 patients treated for prostate cancer under an adaptive radiotherapy protocol were used to model gastrointestinal (GI) toxicity grade>=2, as defined by CTCAE v.3. Of the total cohort, 458 patients were treated with 3D-CRT, the remaining 658 with IMRT.
Three NTCP models were investigated: the Lyman-gEUD model, the Lyman-EUDlog model, and a dose-volume-based cutoff-dose logistic regression model. All parameter estimates and their standard-errors were determined using maximum-likelihood-estimation. Models were fit to the whole group, as well as to the 3D-CRT and IMRT sub-groups separately.
Results: 128 (27.9%) of the patients who received 3D-CRT, and 48 (7.3%) of IMRT patients experienced GI toxicities grade>=2. Median follow-up was 5.8 years (2.0-12.5 years).
For all models, parameter estimates derived from conformal data were in agreement with values found in the literature, but did not predict IMRT data well (p>0.05). Models developed for the whole population overestimate the NTCP for IMRT patients and underestimate the NTCP for 3D-CRT patients. The Lyman-EUDlog model converged to the Lyman-gEUD model for all three patient groups.
Conclusion: Our study suggests that using the NTCP models investigated here, it is not possible to predict toxicities for IMRT and 3D-CRT patients with the same model. This holds for both the EUD-based as well as the dose-volume-based models. The reason for these findings have yet to be investigated. Since IMRT is increasingly utilized clinically, it is of high importance that appropriate NTCP model parameters are determined for this treatment modality, especially when using NTCP models in practice.