2469 - Incorporation of Variable RBE in Radiation-Induced Lymphopenia Prediction Risk Modeling for Proton Therapy
Presenter(s)
M. Grayson1, Y. Chen2, B. Gao1, and R. Mohan1; 1Department of Radiation Physics, The University of Texas MD Anderson Cancer Center, Houston, TX, 2Department of Epidemiology and Biostatistics, Texas A&M University, College Station, TX
Purpose/Objective(s): Radiation-induced lymphopenia (RIL) is a common adverse event of radiation therapy and is associated with decreased overall survival and reduced response to anti-PD-1 immunotherapy. Our group previously developed a multivariable RIL risk prediction model incorporating dosimetric and clinical parameters. However, the model assumes constant relative biological effectiveness (RBE) and does not account for variations in linear energy transfer (LET). As lymphocytes are highly radiosensitive, elevated LET in low-dose regions surrounding the treatment volume may increase lymphocyte killing beyond that predicted by 1.1-weighted dose alone. We hypothesized that incorporating LET-dependent RBE-weighted dose-volume histograms (DVHs) into the RIL risk model would improve prediction of absolute lymphocyte count (ALC) nadir during treatment.
Materials/Methods:
Monte Carlo dose and LET distributions were calculated for twenty esophageal cancer patients treated with intensity-modulated proton therapy using a fast Monte Carlo dose engine. DVHs were generated using a constant RBE of 1.1 and variable RBE models including McNamara, Wedenberg, and repair–misrepair–fixation (RMF). Variable RBE-weighted DVHs replaced 1.1 weighted DVH inputs within our previously validated multivariable RIL risk prediction model. Predicted ALC nadirs were compared with clinically measured nadir values.Results: Variable RBE-weighted DVH metrics improved agreement between predicted and measured ALC nadir values in 11 of 20 patients (55%) compared with predictions based on 1.1 weighted DVHs. Among the evaluated models, the McNamara RBE model provided the most consistent improvement in prediction accuracy. Despite the model having been trained using 1.1 weighted DVHs, variable RBE weighting improved model performance for a subset of patients. This suggests that LET-dependent RBE variations in low-dose regions may contribute to RIL. Improvements in prediction accuracy were not limited to patients with large baseline model error, suggesting that LET-dependent effects provide information not captured by the original model. Notably, patients with improved prediction generally had elevated LET in the heart volume. This highlights the potential clinical relevance of incorporating LET into lymphopenia modeling.
Conclusion: Incorporating variable RBE into DVHs shows promise for improving prediction of radiation-induced lymphopenia in proton therapy. Ongoing work includes retraining the RIL prediction model using variable RBE-weighted DVHs in a cohort of 477 esophageal cancer patients. Improvements were particularly observed in patients with elevated LET in the heart, suggesting that LET-informed modeling may identify regions at higher risk for lymphocyte depletion. This may support the implementation of biologically optimized proton therapy planning for RIL risk reduction and improve patient risk prediction.