Normal Tissue Complication Probability Modeling of Severe Radiation-Induced Lymphopenia Using Blood Dose for Patients With Hepatocellular Carcinoma

Abstract

Purpose: This study aimed to develop a normal tissue complication probability (NTCP) model to estimate the risk of severe radiation-induced lymphopenia (SRIL; absolute lymphocyte count [ALC] < 500/ <mu> L) by using the blood dose of patients with hepatocellular carcinoma (HCC). Methods and Materials: We retrospectively collected data from 75 patients with HCC who received radiation therapy (RT) between 2015 and 2018. The hematological dose framework calculated blood dose -volume histograms (DVHs) using a predefi ned blood fl ow model, organ DVHs, the number of treatment fractions, and beam delivery time. A Lyman-Kutcher-Burman model with a generalized equivalent dose was used to establish the NTCP model, re fl ecting the whole-blood DVHs. Optimization of the Lyman-Kutcher-Burman parameters was conducted by minimizing a negative log-likelihood function. Results: There were 6, 4, 18, 33, and 14 patients in the groups with radiation-induced lymphopenia grades 0, 1, 2, 3, and 4, respectively. The median pre- and post-RT ALC values were 1410/ mu L (range, 520-3710/ mu L) and 470/ mu L (range, 60-1760/ mu L), respectively. There was a correlation between mean blood dose and ALC depletion (Pearson r = - 0.664; P < .001). The average mean blood doses in each radiation-induced lymphopenia group were 2.90 Gy (95% CI, 1.96-3.85 Gy) for grade 0 to 1, 5.29 Gy (95% CI, 4.12-6.45 Gy) for grade 2, 8.81 Gy (95% CI, 7.55-10.07 Gy) for grade 3, and 11.69 Gy (95% CI, 9.82-17.57 Gy) for grade 4. When applying the developed NTCP model to predict SRIL, the area under the receiver operating characteristic curve and Brier score values were 0.89 and 0.12, respectively. Conclusions: We developed the fi rst NTCP model based on whole-blood DVHs for estimating SRIL after abdominal RT in patients with HCC. Our results showed a strong correlation between blood dose and ALC depletion, suggesting the potential to predict the risk of SRIL occurrence using blood dose.