Artificial intelligence for early prediction of alcohol-related liver disease: Advances, challenges, and clinical applications

Table 1 Key studies on artificial intelligence in alcohol-related liver disease prediction

Study focus	Methodology	Key findings	Citation
Multi-omics integration	Machine learning with transcriptomics and proteomics data	90% accuracy for liver tissue classification, 89% for PBMCs	(Listopad et al[4], 2024) (Listopad et al[3], 2022)
Gut microbiota-based diagnosis	Supervised ML algorithms with feature reduction techniques	AUCs > 0.90 for ALD and NAFLD diagnosis	(Park et al[5], 2024)
Gradient boosting for mortality	Gradient boosting with laboratory and microbiome data	AUC of 0.87 for 30-day mortality prediction, outperforming MELD score	(Gao et al[2], 2022)
Proteomic panels for fibrosis	Machine learning on liver and plasma proteomics data	AUC of 0.90 for detecting F2 or greater fibrosis	(Mezzacappa and Bhat[8], 2023)
Bayesian optimization for classifiers	Bayesian optimization of Random Forest and XGBoost	81.06% accuracy for liver disease prediction	(Kumar and Rani[7], 2024)
Extra Tree for early detection	Extra Tree algorithm with oversampling techniques	92% accuracy for early liver disease detection	(Lima et al[10], 2024)

ALD: Alcoholic-associated liver disease; AUC: Area under the curve; ML: Machine learning; NAFLD: Non-alcoholic fatty liver disease; PBMCs: Peripheral blood mononuclear cells. MELD: Model for end-stage liver disease; XGBoost: eXtreme Gradient Boosting.

Table 2 Comparative analysis of artificial intelligence models in alcohol-related liver disease prediction

AI model	Methodology	Performance metrics	Key limitations	Citation
Gradient Boosting	MICE imputation, SMOTE, feature selection	AUC = 0.87 (30-day mortality prediction)	Small sample size, Lack of external validation, missing data	Gao et al[2], 2022
Stacked Ensemble (XGBoost + Logistic Regression)	Multi-omics + clinical features	Accuracy = 93.86% (alcoholic cirrhosis prediction)	Lack of external validation	Vinutha et al[6], 2022
Random Forest/XGBoost	Bayesian optimization	Accuracy = 81.06% (Random Forest), 79.85% (XGBoost)	Data heterogeneity	Kumar and Rani[7], 2024
Extra Tree with Oversampling	Liver stiffness + clinical data	Accuracy = 92% (early ARLD detection)	Requires high-resolution imaging	Lima et al[10], 2024

AUC: Area under the curve; MICE: Multivariate imputation by chained equations; SMOTE: Synthetic minority oversampling technique; XGBoost: eXtreme Gradient Boosting; ARLD: Alcohol-related liver disease; AI: Artificial intelligence.