Inclusion of biological factors in parallel-architecture normal-tissue complication probability model for radiation-induced liver disease

Purpose: To include biologic factors in parallel-architecture normal-tissue complication probability (NTCP) model for radiation-induced liver disease (RILD) after three-dimensional conformal radiotherapy (3D-CRT) for gastric or hepatic cancer. Methods and Materials: A total of 151 patients (89 with hepatocellular carcinoma and 62 with gastric cancer) who received 3D-CRT to the liver were included (isocenter dose range 33.0 to 66.0 Gy; mean 48.0 Gy). RILD was defined as grade 3 or higher liver toxicity according to Common Toxicity Criteria Version 2.0 of the National Cancer Institute within 4 months after 3D-CRT. Possible correlations of patient-related or dosimetric factors with RILD were tested. Maximum-likelihood analysis estimated NTCP model parameters for group and subgroups. Goodness-of-fit analysis estimated deviance of NTCP model parameters between subgroups. Results: RILD developed in 25 patients. Hepatitis B virus carrier status (p < 0.001) was the only significant independent factor. The 4 parallel NTCP model parameters, mean functional reserve (V50), width of functional reserve distribution (σ), dose damage to 50% of liver subunits (D50), and slope parameter for subunit dose-response (k), were respectively, 0.54, 0.14, 50 Gy, 0.18 (group); 0.53, 0.07, 50 Gy, 4.6 x 10-7 (carriers); 0.59, 0.12, 25 Gy, 59.8 (noncarriers). In carrier-state subgroups, goodness-of-fit deviance with 1 subgroup's parameter set would have been worse in the other group. Across subgroups, patients with RILD all had liver fraction damage (f) greater than 0.4 compared with wider distribution for the whole group. Conclusions: RILD is described with a parallel-architecture NTCP model for HBV carriers and noncarriers with a threshold effect greater than 0.4. The main difference is in slope parameter for subunit dose-response