ALBI-RT as a Radiation Therapy Specific Hepatotoxicity Prediction Model for Hepatocellular Carcinoma Patients

Abstract

The albumin-bilirubin (ALBI) grading system was developed for predicting hepatotoxicity in hepatocellular carcinoma (HCC) patients; however, its value in patients treated with radiotherapy (RT) remains unclear. We sought to develop a modified ALBI grading system, ALBI-RT, as a novel pre-treatment grading system for stratification of hepatotoxicity risk in HCC patients treated with RT. We performed a retrospective review of 92 consecutive HCC patients treated with RT from 2013 – 2018. Raw ALBI scores and Child-Pugh (CP) scores were calculated. Patient deaths were assessed for association with radiation-induced liver disease (RILD). Using raw ALBI score as a continuous variable, ROC analysis was performed to define risk strata of RILD-specific survival (RILD-SS). Univariate Cox regression models of RILD-SS were generated using raw ALBI score, conventional ALBI grade, ALBI-RT grade, CP score, CP class, and CP score increase of 2 or greater (CP+2). The association of ALBI-RT grade with known predictors of hepatotoxicity was assessed in two bivariate Cox models which adjusted for conventional ALBI grade and CP+2, respectively. The patient cohort was comprised of 67% CP-A and 33% CP-B/C. Median follow-up was 11.5 months for all patients and 15.5 months for alive patients. There were 40 deaths, with 8 attributed to possible RILD. ROC analysis identified raw ALBI (AUC = 0.91, p < 0.001) dichotomization at a score of -1.7 to predict RILD with 100% sensitivity and 80% specificity. Based on this dichotomization, ALBI-RT risk categories were divided into low risk grade A < 1.7 (n = 69) and high risk grade B ≥ 1.7 (n = 23). ALBI-RT grade B identified 6% of CP A patients at increased risk for RILD and conversely, ALBI-RT grade A identified 37% of CP B/C patients with decreased risk. On univariate analysis, ALBI-RT grade B (HR 25.5), CP-A5-B7 vs B8+ (HR 46.8), raw ALBI score (HR 5.3), ALBI grade (HR 8.6), CP score (HR 2.3), CP+2 (HR 11.9) were all significantly associated with a higher relative risk of RILD-SS (p ≤ 0.003). On bivariate analysis, ALBI-RT (HR 14, p = 0.030) was independently predictive for RILD-SS compared to conventional ALBI grade (HR 2.2, p = 0.283); furthermore, ALBI-RT (HR 17.3, p = 0.009) was a more significant predictor of RILD-SS than post-treatment toxicity metric CP+2 (HR 5.1, p = 0.025). ALBI-RT is a novel metric for pre-treatment assessment of HCC patients that predicts risk of RILD-related death. ALBI-RT showed a larger effect size and retained its statistical significance relative to known hepatotoxicity predictors (ALBI grade, CP+2). Multi-institution external validation has the potential to strengthen the generalizability of ALBI-RT and facilitate clinical translation.