Personalized Mortality Risk Stratification in ALD- and MASLD-Related Hepatocellular Carcinoma Using a Machine Learning Approach

<b>Background/Objectives:</b> The epidemiology of hepatocellular carcinoma (HCC) is shifting, with alcohol-associated liver disease (ALD) and metabolic dysfunction-associated steatotic liver disease (MASLD) becoming leading causes in developed countries. This study aimed to identify the main prognostic factors for mortality at diagnosis in HCC patients with ALD and MASLD using machine learning (ML) algorithms. Random Forest (RF) was proposed as reference method. <b>Methods:</b> A multicenter, retrospective cohort of 91 patients diagnosed with HCC due to ALD or MASLD between 2008 and 2023 was analyzed. Demographic, clinical, and biochemical variables were collected. Several ML algorithms were implemented: RF, Support Vector Machine, Decision Tree, Gaussian Naïve Bayes, and K-Nearest Neighbors. Bayesian optimization was applied for hyperparameter tuning. Model performance was evaluated using standard metrics including AUC, precision, recall, and F1 score. <b>Results:</b> RF achieved the highest performance across all metrics (AUC: 0.91, precision: 90.67%, F1 score: 91.05%), surpassing other algorithms by over 10%. The most relevant variables for mortality prediction were serum albumin, CRP/albumin ratio, BCLC stage, and ALBI score. MELD 3.0 showed superior predictive value compared to other MELD variants. Conversely, AFP had limited prognostic utility in this population. <b>Conclusions:</b> In HCC patients related to ALD and MASLD, liver function and systemic inflammation markers outperform tumor markers for early mortality prediction. In this cohort, RF offered the highest predictive performance among the evaluated algorithms and may support personalized prognosis in ALD- and MASLD-related HCC; however, external validation in independent datasets is required before broad clinical implementation.