Machine learning-based normal tissue complication probability model for predicting albumin-bilirubin (ALBI) grade increase in hepatocellular carcinoma patients

Table 2 Internal validation of each NTCP model for predicting ALBI1+.

	AUROC (mean ± SD)	Input Features
ALBI score until next grade	0.7657 ± 0.0738	ALBI score until next grade
Logistic regression (PLR)	0.7864 ± 0.0848	Age, ALBI score until next grade, baseline ALBI grade, total bilirubin, portal vein thrombosis, total dose, and MLD
Random forest (RF)	0.8076 ± 0.0583	AST, ALBI score until next grade, baseline ALBI score, total bilirubin, normal liver volume, and MLD
Gradient-boosted tree (GBT)	0.8166 ± 0.0709	AST, ALBI score until next grade, baseline ALBI score, total bilirubin, normal liver volume, and MLD
Ensemble model (PLR + GBT)	0.8214 ± 0.0605	PLR and GBT Inputs

Abbreviation: PLR = Penalized logistic regression; RF = Random forest; GBT = Gradient-boosted tree; AUROC = area under the receiver operating characteristic curve; SD = standard deviation; ALBI = albumin-bilirubin score; AST = aspartate aminotransferase; MLD = mean liver dose (gEUD at a = 1.0)

ISSN: 1748-717X