Comparing different NTCP models that predict the incidence of radiation pneumonitis

To compare different normal tissue complication probability (NTCP) models to predict the incidence of radiation pneumonitis on the basis of the dose distribution in the lung.The data from 382 breast cancer, malignant lymphoma, and inoperable non-small-cell lung cancer patients from two centers were studied. Radiation pneumonitis was scored using the Southwestern Oncology Group criteria. Dose-volume histograms of the lungs were calculated from the dose distributions that were corrected for dose per fraction effects. The dose-volume histogram of each patient was reduced to a single parameter using different local dose-effect relationships. Examples of single parameters were the mean lung dose (MLD) and the volume of lung receiving more than a threshold dose (V(Dth)). The parameters for the different NTCP models were fit to patient data using a maximum likelihood analysis.The best fit resulted in a linear local dose-effect relationship, with the MLD as the resulting single parameter. The relationship between the MLD and NTCP could be described with a median toxic dose (TD(50)) of 30.8 Gy and a steepness parameter m of 0.37. The best fit for the relationship between the V(Dth) and the NTCP was obtained with a D(th) of 13 Gy. The MLD model was found to be significantly better than the V(Dth) model (p <0.03). However, for 85% of the studied patients, the difference in NTCP calculated with both models was <10%, because of the high correlation between the two parameters. For dose distributions outside the range of the studied dose-volume histograms, the difference in NTCP, using the two models could be >35%. For arbitrary dose distributions, an estimate of the uncertainty in the NTCP could be determined using the probability distribution of the parameter values of the Lyman-Kutcher-Burman model.The maximum likelihood method revealed that the underlying local dose-effect relation for radiation pneumonitis was linear (the MLD model), rather than a step function (the V(Dth) model). Thus, for the studied patient population, the MLD was the most accurate predictor for the incidence of radiation pneumonitis.