Differential NTCP Modeling of Radiation-Induced Lymphopenia In Lung Cancer Patients Undergoing IMRT and IMPT
Abstract
Purpose
Severe radiation-induced lymphopenia (SRIL) is a detrimental prognostic factor in lung cancer. This study aimed to develop and validate normal tissue complication probability (NTCP) models for SRIL based on hematologic dose in patients receiving intensity-modulated radiotherapy (IMRT) and intensity-modulated proton therapy (IMPT).
Methods
We retrospectively analyzed 157 lung cancer patients treated with curative-intent radiotherapy (94 IMRT, 63 IMPT). Whole-body blood dose-volume histograms were computed using the HEDOS framework. The Lyman-Kutcher-Burman NTCP model parameters were optimized via maximum likelihood estimation. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) and Brier score.
Results
SRIL incidence was 61.7% in the IMRT cohort and 33.3% in the IMPT cohort. Blood generalized equivalent uniform dose (gEUD) was an independent predictor of SRIL in the IMRT and IMPT cohort (OR=1.830, p<0.001 and OR=2.661, p<0.001). The NTCP models demonstrated strong predictive performance (IMRT: AUC=0.82; IMPT: AUC=0.81). Model parameters differed significantly between modalities, particularly the volume-effect parameter (a=19.85 for IMRT vs. 1.58 for IMPT). External validation of the IMRT-derived model in the IMPT cohort revealed suboptimal calibration (slope=0.54), indicating systematic risk overestimation and reflecting distinct blood dose distributions between photon and proton therapy.
Conclusion
Modality-specific NTCP models are essential for accurate prediction of SRIL in lung cancer radiotherapy. The developed IMRT and IMPT models exhibited robust performance within their respective cohorts. Differences in model parameters underscore the impact of modality-dependent blood dose distributions, supporting the need for separate risk models for photon and proton therapies to guide personalized treatment planning.