Machine learning of systemic metabolic profiles and longitudinal trajectories for risk stratification in HBV-related chronic liver disease

Existing clinical indices, such as the albumin–bilirubin (ALBI) score and fibrosis-4 (FIB-4) index, serve as pragmatic bedside benchmarks but face limitations in capturing the dynamic nature of hepatic decompensation. This study aimed to investigate whether integrating broad systemic metabolic profiles with early longitudinal ALBI trajectories can refine risk stratification within a large-scale, heterogeneous hepatitis B virus (HBV)-related chronic liver disease cohort. We conducted a retrospective cohort study tracking 11,768 patients with HBV-related conditions (including inactive carriers, chronic hepatitis B, compensated cirrhosis, and decompensated cirrhosis). The cohort was partitioned into a training/validation set ( n = 8,267) and an independent test set ( n = 3,501). Machine learning frameworks (LASSO–Cox regression) were employed to construct a metabolic risk score. Trajectory clustering characterized ALBI measurements available by day 180, and a strict landmark sensitivity analysis assessed the feasibility of subsequent outcome comparisons. The random forest model showed high discrimination for severe ALBI grade classification (AUC 0.986), while the prognostic nomogram for event-free survival achieved time-dependent AUCs of 72.4%, 77.9%, and 82.1% at 3, 6, and 12 months. SHAP analysis indicated that systemic metabolic markers, especially the albumin-to-globulin (A/G) ratio, prealbumin, calcium, and magnesium, contributed strongly to ALBI severity classification. After restricting month 6 measurements to days 150–180, 96 patients had complete trajectories and 36 entered the landmark analysis. No subsequent events were recorded using the existing diagnosis-text endpoint, precluding between-group survival comparisons. Broad systemic metabolic profiling may contribute to risk stratification, although incremental prognostic value beyond conventional static scores remains to be established. Longitudinal ALBI trajectories describe heterogeneous biochemical patterns; their prognostic value could not be established because the strict day 180 landmark analysis had no subsequent recorded events.